{"generated":"2026-11-08","hn_export":"2026-10-31","videos":[{"id":"VMj-3S1tku0","t":"The spelled-out intro to neural networks and backpropagation: building micrograd","ch":"Andrej Karpathy","s":"Andrej Karpathy's first Neural Networks: Zero to Hero lecture, which builds micrograd, a tiny autograd engine, from scratch in Python. It explains backpropagation and neural network training step by step, assuming only basic Python and some high-school calculus.","w":"HN treats it as the best starting point for really understanding backprop and neural nets, because Karpathy explains the why with intuition and builds everything in code.","type":"lecture","lvl":"intro","tags":["micrograd","autograd","backpropagation","neural networks","python","andrej karpathy"],"fy":2022,"ly":2025,"sc":33.6,"shelf":"ml","new":false},{"id":"aircAruvnKk","t":"But what is a neural network? | Deep learning chapter 1","ch":"3Blue1Brown","s":"3Blue1Brown's first chapter on deep learning explains what neurons and layers are and the math underlying a basic neural network, using handwritten digit recognition with visual animations.","w":"HN treats it as the standard first stop for learning neural networks, praised for making the math intuitive.","type":"explainer","lvl":"intro","tags":["deep learning","3blue1brown","gradient descent","visualization","machine learning basics"],"fy":2017,"ly":2025,"sc":26.4,"shelf":"ml","new":false},{"id":"7Pq-S557XQU","t":"What Happened to Horses Is Happening to Us","ch":"CGP Grey","s":"CGP Grey's video \"Humans Need Not Apply\" argues that automation will make a large part of the workforce economically obsolete, much as engines did for horses. It surveys robots and software replacing both physical and mental labor.","w":"HN has recommended it for a decade as the standard explainer on automation and unemployment, though skeptics later cite it as an example of overconfident predictions.","type":"explainer","lvl":"general","tags":["automation","jobs","ai","economics","robotics"],"fy":2014,"ly":2026,"sc":23.62,"shelf":"bigpicture","new":false},{"id":"wjZofJX0v4M","t":"Transformers, the tech behind LLMs | Deep Learning Chapter 5","ch":"3Blue1Brown","s":"3Blue1Brown's visual explanation of how transformers and GPT-style language models work, tracing data through tokens, embeddings and attention.","w":"HN treats it as the go-to visual introduction to transformers, and many say it finally made attention and embeddings click.","type":"explainer","lvl":"intro","tags":["transformers","attention","embeddings","llm","deep learning","visualization"],"fy":2024,"ly":2026,"sc":22.75,"shelf":"llm","new":false},{"id":"WXuK6gekU1Y","t":"AlphaGo - The Movie | Full award-winning documentary","ch":"Google DeepMind","s":"Feature-length documentary following DeepMind's AlphaGo program as it prepares for and plays the 2016 match against Go champion Lee Sedol. It covers the team, the games, and the emotional reaction of the Go world to an AI beating a top human.","w":"HN commenters call it a moving, well-made human story about AI and Go that works for non-technical viewers and does not date.","type":"documentary","lvl":"general","tags":["alphago","lee sedol","deepmind","go","reinforcement learning"],"fy":2020,"ly":2025,"sc":20.8,"shelf":"ml","new":false},{"id":"qv6UVOQ0F44","t":"MarI/O - Machine Learning for Video Games","ch":"SethBling","s":"SethBling shows MarI/O, a program using neural networks and the NEAT genetic algorithm to learn to play Super Mario World. The description links the source code, the paper and background reading.","w":"The go-to accessible introduction to neuroevolution and NEAT, praised for crediting prior research.","type":"demo","lvl":"intro","tags":["neat","genetic algorithms","neural networks","super mario world","machine learning"],"fy":2015,"ly":2026,"sc":20.12,"shelf":"ml","new":false},{"id":"7xTGNNLPyMI","t":"Deep Dive into LLMs like ChatGPT","ch":"Andrej Karpathy","s":"Andrej Karpathy's three-hour-plus general-audience deep dive into how ChatGPT-style LLMs are built, covering the full training stack from pretraining to post-training. It also gives mental models for their quirks, like hallucination, and how to use them well.","w":"HN's most frequent recommendation for a clear, approachable foundation in how LLMs work without heavy math.","type":"lecture","lvl":"intro","tags":["llm training","chatgpt","pretraining","hallucination","tokenization","andrej karpathy"],"fy":2025,"ly":2026,"sc":19.25,"shelf":"llm","new":false},{"id":"xOCurBYI_gY","t":"Computer program that learns to play classic NES games","ch":"suckerpinch","s":"Tom7 explains and demos software that learns to play NES games by finding score patterns in memory and then searching for inputs that raise it. It is the video companion to his SIGBOVIK paper.","w":"Loved on HN as a hilarious, brilliant example of hacker-academic work and cited whenever game-playing AI or fuzzing comes up.","type":"demo","lvl":"general","tags":["learnfun","playfun","tom7","nes","sigbovik","machine learning"],"fy":2015,"ly":2025,"sc":17.6,"shelf":"ml","new":false},{"id":"2pWv7GOvuf0","t":"RL Course by David Silver - Lecture 1: Introduction to Reinforcement Learning","ch":"Google DeepMind","s":"The first lecture of David Silver's reinforcement learning course, introducing the field and its basic concepts. The course follows the Sutton and Barto textbook.","w":"Frequently recommended as an excellent, clear introductory RL course from the lead of AlphaGo, pairing well with Sutton and Barto.","type":"course","lvl":"intro","tags":["reinforcement learning","david silver","deepmind","sutton and barto","machine learning"],"fy":2017,"ly":2026,"sc":17.5,"shelf":"ml","new":false},{"id":"42QuXLucH3Q","t":"Is Most Published Research Wrong?","ch":"Veritasium","s":"Veritasium explains why a large share of published research findings may be false, covering p-values, p-hacking, publication bias and low replication rates. It draws on John Ioannidis's well-known paper and describes reforms such as preregistration.","w":"HN commenters link it whenever a shaky study or the p<0.05 threshold comes up, as a short primer on false positives in science.","type":"explainer","lvl":"general","tags":["p-values","p-hacking","replication crisis","statistics","publication bias"],"fy":2017,"ly":2025,"sc":16.0,"shelf":"data","new":false},{"id":"kCc8FmEb1nY","t":"Let's build GPT: from scratch, in code, spelled out.","ch":"Andrej Karpathy","s":"Andrej Karpathy builds a GPT from scratch in code, following Attention Is All You Need and GPT-2/3, in about two hours using PyTorch. He connects it to ChatGPT and assumes familiarity with his earlier makemore videos.","w":"It is the go-to recommendation for learning how transformers and LLMs actually work, praised for its clarity and Karpathy's teaching.","type":"tutorial","lvl":"intermediate","tags":["gpt","transformer","pytorch","nanogpt","attention","from scratch"],"fy":2023,"ly":2025,"sc":16.0,"shelf":"llm","new":false},{"id":"jkrNMKz9pWU","t":"A Hackers' Guide to Language Models","ch":"Jeremy Howard","s":"Jeremy Howard's 90-minute practical tour of language models. It covers how they work, GPT-4's capabilities and limits, using the OpenAI API, code writing, data analysis and fine-tuning.","w":"Repeatedly named the best practical tutorial for using LLMs, with a well-liked fine-tuning section.","type":"tutorial","lvl":"intermediate","tags":["language models","gpt-4","fine-tuning","openai api","jeremy howard","prompting"],"fy":2023,"ly":2024,"sc":13.78,"shelf":"llm","new":false},{"id":"zjkBMFhNj_g","t":"[1hr Talk] Intro to Large Language Models","ch":"Andrej Karpathy","s":"Andrej Karpathy gives a one-hour general-audience introduction to large language models, covering what they are, how they are trained, comparisons to operating systems, and security challenges such as jailbreaks.","w":"HN recommends it as the dense, approachable starting point for understanding LLMs and often cites it for specific claims about scaling and training.","type":"talk","lvl":"intro","tags":["llm","pretraining","scaling","jailbreaks","karpathy","system 2"],"fy":2023,"ly":2026,"sc":13.12,"shelf":"llm","new":false},{"id":"FACK2knC08E","t":"DON'T PANIC — Hans Rosling showing the facts about population","ch":"Gapminder Foundation","s":"Hans Rosling uses data and animated statistics to show how world population is changing, why birth rates fall as child survival and wealth rise, and why population is expected to peak around 11 billion. It also tackles common misconceptions about Africa and family planning.","w":"HN repeatedly recommends it to counter overpopulation alarm with data on falling birth rates.","type":"lecture","lvl":"general","tags":["hans rosling","birth rates","demography","gapminder","statistics"],"fy":2017,"ly":2021,"sc":13.0,"shelf":"data","new":false},{"id":"0xS68sl2D70","t":"Veritasium: What Everyone Gets Wrong About AI and Learning – Derek Muller Explains","ch":"Perimeter Institute for Theoretical Physics","s":"Veritasium's Derek Muller gives a Perimeter Institute livestream on how AI might change teaching and learning. He draws on past technology shifts and cognitive science to argue that learning requires effortful repetition.","w":"Recommended as a balanced take on AI and education, with the point that learning takes work that no technology can do for you.","type":"talk","lvl":"general","tags":["education","learning science","ai","cognitive science","veritasium"],"fy":2025,"ly":2026,"sc":12.25,"shelf":"bigpicture","new":false},{"id":"hEUO6pjwFOo","t":"Intelligence and Stupidity: The Orthogonality Thesis","ch":"Robert Miles AI Safety ","s":"Robert Miles explains the orthogonality thesis: how an agent can be highly intelligent while pursuing goals that seem stupid or alien to humans. He separates intelligence from goals and touches on the stamp-collector example.","w":"HN people use it as the standard short explanation for why AI can understand human ethics without sharing them, and often recommend the whole channel.","type":"explainer","lvl":"intro","tags":["orthogonality thesis","ai safety","agi","goals vs intelligence","robert miles"],"fy":2022,"ly":2026,"sc":12.25,"shelf":"bigpicture","new":false},{"id":"UzxYlbK2c7E","t":"Lecture 1 | Machine Learning (Stanford)","ch":"Stanford","s":"The first lecture of Andrew Ng's Stanford CS229 Machine Learning course, giving an overview of the course. It introduces supervised and unsupervised learning, learning theory and reinforcement learning.","w":"Recommended as the rigorous, math-heavy original course, deeper than the Coursera version, for learning machine learning.","type":"course","lvl":"intermediate","tags":["andrew ng","cs229","supervised learning","stanford","machine learning course"],"fy":2008,"ly":2020,"sc":11.5,"shelf":"ml","new":false},{"id":"iFTWM7HV2UI","t":"We're building a dystopia just to make people click on ads | Zeynep Tufekci","ch":"TED","s":"Technosociologist Zeynep Tufekci argues that the machine-learning systems used to serve ads also decide what political and social information people see. She explains how opaque algorithms can nudge, segment and manipulate users, and warns that the real danger is powerful people using AI to control populations.","w":"HN commenters cite it whenever engagement-optimizing feeds, micro-targeting or ad infrastructure come up, as an accessible warning about algorithmic amplification.","type":"talk","lvl":"general","tags":["ad targeting","recommendation algorithms","persuasion","social media","ai ethics"],"fy":2017,"ly":2026,"sc":11.38,"shelf":"bigpicture","new":false},{"id":"7jiPeIFXb6U","t":"I don't like notebooks.- Joel Grus (Allen Institute for Artificial Intelligence)","ch":"O'Reilly","s":"Joel Grus's JupyterCon talk explaining why he dislikes Jupyter notebooks. He covers hidden state, trouble with reproducibility and testing, and how notebooks frustrate teaching. He shows his preferred workflow and what Jupyter could change.","w":"It is the go-to funny, irreverent critique of notebooks, linked whenever HN debates notebooks versus testable code.","type":"talk","lvl":"intermediate","tags":["jupyter","python","data science","reproducibility","software engineering"],"fy":2018,"ly":2025,"sc":11.2,"shelf":"agents","new":false},{"id":"N1CBDTgGtOU","t":"VisiData Lightning Demo at PyCascades 2018","ch":"Saul Pwanson","s":"A five-minute lightning demo of VisiData, a terminal tool for exploring and manipulating data. It uses two months of NYC 311 complaint data to drill down, plot points on a virtual map and save selected rows.","w":"HN recommends it as the fastest way to understand what VisiData can do, since seeing it work is clearer than reading about it.","type":"demo","lvl":"intro","tags":["visidata","tui","csv","data analysis","python"],"fy":2018,"ly":2025,"sc":11.2,"shelf":"data","new":false},{"id":"9Q6sLbz37gk","t":"I Like Notebooks","ch":"Jeremy Howard","s":"Jeremy Howard defends Jupyter notebooks as a serious development environment, especially when paired with nbdev. He responds directly to Joel Grus's talk about why he dislikes notebooks, and contrasts REPL-style and notebook-style work.","w":"Whenever notebooks get criticized, HN commenters point to it as the standard rebuttal from a prominent practitioner.","type":"talk","lvl":"intermediate","tags":["jupyter","nbdev","literate programming","fastai","developer workflow"],"fy":2020,"ly":2025,"sc":10.4,"shelf":"agents","new":false},{"id":"kTiP0zKF9bc","t":"Robocalypse Now: Using Deep Learning to Combat Cheating in Counter-Strike: Global Offensive","ch":"GDC Festival of Gaming","s":"Valve's John McDonald presents at GDC 2018 on how Valve uses deep learning (VACnet) to detect cheaters in Counter-Strike: Global Offensive. He covers training on human-reviewed replays and how model output is confirmed by manual review.","w":"The go-to link in any anti-cheat thread for a concrete, well-delivered account of machine-learning cheat detection in production.","type":"talk","lvl":"intermediate","tags":["vacnet","cs:go","deep learning","cheat detection","valve","gdc"],"fy":2021,"ly":2025,"sc":10.4,"shelf":"ml","new":false},{"id":"Sm5xF-UYgdg","t":"How not to be ignorant about the world | Hans and Ola Rosling","ch":"TED","s":"Hans Rosling and his son Ola run an audience quiz with global population, health and income data to show how wrong most people are about the world. Ola then gives four ways to get less ignorant.","w":"Recommended as evidence that the world is improving despite negative news, and for how people misjudge global trends.","type":"talk","lvl":"general","tags":["hans rosling","world progress","misconceptions","global health","poverty"],"fy":2015,"ly":2023,"sc":9.75,"shelf":"data","new":false},{"id":"ZeecOKBus3Q","t":"Why Would AI Want to do Bad Things? Instrumental Convergence","ch":"Robert Miles AI Safety ","s":"Robert Miles explains instrumental convergence: why an AGI with almost any goal would tend to pursue sub-goals like self-preservation and resource acquisition. It builds on Omohundro's basic AI drives and his orthogonality thesis video.","w":"HN commenters repeatedly recommend it as the clearest introduction to why smart AI might behave badly by default.","type":"explainer","lvl":"intro","tags":["instrumental convergence","ai risk","agi","robert miles","orthogonality"],"fy":2019,"ly":2023,"sc":9.75,"shelf":"bigpicture","new":false},{"id":"HLRdruqQfRk","t":"Uppestcase and Lowestcase Letters  [advances in derp learning]","ch":"suckerpinch","s":"Tom Scott-style absurdist research by tom7 for SIGBOVIK 2021, using machine learning to discover letterforms that are 'uppercaser' than uppercase A and lowercaser than lowercase. It is a comedic, technically inventive deep dive with fonts and a paper.","w":"Tom7 is an HN favorite, and this video is praised as hilarious, educational silly ML that keeps getting better.","type":"comedy","lvl":"general","tags":["tom7","sigbovik","fonts","neural networks","humor"],"fy":2021,"ly":2026,"sc":9.62,"shelf":"ml","new":false},{"id":"qbIk7-JPB2c","t":"Sparks of AGI: early experiments with GPT-4","ch":"Sebastien Bubeck","s":"Sebastien Bubeck presents the Sparks of AGI paper at MIT, showing early experiments with GPT-4 across many domains and asking whether artificial intelligence has arrived. It includes the TikZ unicorn example.","w":"HN recommends it as a thought-provoking talk from a researcher with early GPT-4 access, especially for the unicorn degrading under safety tuning.","type":"lecture","lvl":"intermediate","tags":["gpt-4","agi","llm","alignment tax","microsoft research"],"fy":2023,"ly":2026,"sc":9.62,"shelf":"bigpicture","new":false},{"id":"ZmNOAtZIgIk","t":"Bay Area Vision Meeting: Unsupervised Feature Learning and Deep Learning","ch":"Google TechTalks","s":"Andrew Ng presents unsupervised feature learning and deep learning, showing how algorithms can learn representations from unlabeled data instead of hand-designed features. He gives examples in vision and audio, including visualizations of features learned by successive layers.","w":"HN commenters recommended it for years as a simple, exciting overview of the deep learning wave.","type":"talk","lvl":"intermediate","tags":["deep learning","andrew ng","neural networks","feature learning","computer vision"],"fy":2011,"ly":2013,"sc":9.5,"shelf":"ml","new":false},{"id":"CaCcOwJPytQ","t":"Special Topics - The Kalman Filter (1 of 55) What is a Kalman Filter?","ch":"Michel van Biezen","s":"The first video in a 55-part series by Michel van Biezen that introduces the Kalman filter and builds from the rationale through algebraic examples to the full matrix treatment. The videos are short, blackboard-style lectures.","w":"HN repeatedly recommends this series as a clear, step-by-step way to really understand Kalman filtering.","type":"course","lvl":"intro","tags":["kalman filter","state estimation","linear algebra","control theory","lecture series"],"fy":2017,"ly":2025,"sc":8.8,"shelf":"data","new":false},{"id":"GVsUOuSjvcg","t":"Future Computers Will Be Radically Different (Analog Computing)","ch":"Veritasium","s":"Veritasium explains why analog computing may suit AI workloads, covering old analog computers, The Analog Thing and Mythic's analog chips that use flash memory for neural network inference. It argues digital hardware is hitting limits.","w":"Treated as the accessible reference on analog computing for neural nets, especially the Mythic AI segment.","type":"explainer","lvl":"general","tags":["analog computers","neural networks","mythic ai","in-memory compute","veritasium"],"fy":2022,"ly":2025,"sc":8.8,"shelf":"ml","new":false},{"id":"zduSFxRajkE","t":"Let's build the GPT Tokenizer","ch":"Andrej Karpathy","s":"Andrej Karpathy builds the GPT tokenizer from scratch, implementing byte pair encoding along with encode and decode. He shows how many odd LLM behaviors trace back to tokenization, and covers Unicode byte encodings along the way.","w":"Recommended as the best deep dive on tokenizers, worth the two hours for anyone working with LLMs.","type":"lecture","lvl":"intermediate","tags":["byte pair encoding","tokenizer","gpt","unicode","llm"],"fy":2024,"ly":2025,"sc":8.8,"shelf":"llm","new":false},{"id":"eMlx5fFNoYc","t":"Attention in transformers, step-by-step | Deep Learning Chapter 6","ch":"3Blue1Brown","s":"3Blue1Brown's visual walkthrough of the attention mechanism inside transformers, covering self-attention, multiple heads and cross-attention. It is chapter 6 of his deep learning series.","w":"HN treats it as the go-to visual explanation of attention, recommended to people who find text-based transformer explainers hard to follow.","type":"explainer","lvl":"intro","tags":["attention","transformers","llm","deep learning","visualization","3blue1brown"],"fy":2024,"ly":2026,"sc":8.75,"shelf":"llm","new":false},{"id":"KRcecxdGxvQ","t":"Michael Stonebraker | Big Data is (at least) Four Different Problems","ch":"Michigan Engineering","s":"Database pioneer Michael Stonebraker argues that 'big data' is really at least four different problems: big volumes with small analytics, big analytics, velocity, and variety. He comments on where each market is heading.","w":"HN points to it as a thoughtful, first-principles take from a Turing award winner that cuts through NoSQL and big data hype.","type":"lecture","lvl":"intermediate","tags":["michael stonebraker","big data","databases","sql","nosql"],"fy":2015,"ly":2024,"sc":8.7,"shelf":"data","new":false},{"id":"lhYGXYeMq_E","t":"Why Greatness Cannot Be Planned — Kenneth Stanley","ch":"Machine Learning Street Talk","s":"A long Machine Learning Street Talk interview with Kenneth Stanley, creator of NEAT and novelty search, about his book on why greatness cannot be planned. They discuss open-endedness, the tyranny of objectives, serendipity and the evolution of ideas.","w":"Recommended to people who find it inspiring or perspective-changing on pursuing goals, open-ended search and AI beyond optimization.","type":"interview","lvl":"intermediate","tags":["kenneth stanley","novelty search","objectives","neuroevolution","serendipity"],"fy":2021,"ly":2024,"sc":8.7,"shelf":"ml","new":false},{"id":"hx7BXih7zx8","t":"Andrej Karpathy - AI for Full-Self Driving at Tesla","ch":"Matroid","s":"Andrej Karpathy, then Tesla's head of AI, presents at ScaledML 2020 on how Tesla builds full self-driving using vision-based neural nets and fleet data collection. He covers training custom detectors and neural nets gradually replacing traditional code.","w":"HN repeatedly points to it as the best public technical overview of Tesla's FSD approach, used both to defend and to critique Tesla's progress.","type":"talk","lvl":"intermediate","tags":["tesla autopilot","computer vision","fleet learning","andrej karpathy","autonomous driving"],"fy":2020,"ly":2022,"sc":8.62,"shelf":"ml","new":false},{"id":"EWvNQjAaOHw","t":"How I use LLMs","ch":"Andrej Karpathy","s":"A two-hour walk through how Karpathy actually uses ChatGPT, Claude and friends day to day: when to use thinking models, search, deep research, file uploads, voice, memory and custom instructions, with the reasoning behind each choice.","w":"The practical companion to his Intro to LLMs. People send it to colleagues who know what ChatGPT is but not how to get real work out of it.","type":"lecture","lvl":"intro","tags":["prompting","chatgpt","claude","workflow","karpathy"],"fy":2025,"ly":2025,"sc":8.5,"shelf":"llm","new":true},{"id":"D7_ipDqhtwk","t":"How We Build Effective Agents","ch":"AI Engineer (Barry Zhang, Anthropic)","s":"Anthropic's Barry Zhang on the three principles behind agents that work in production: keep them simple, think from the agent's point of view, and treat the environment and tools as the product.","w":"One of the most-watched AI Engineer talks and the video version of Anthropic's Building Effective Agents essay, which teams keep citing when deciding what not to build.","type":"talk","lvl":"intermediate","tags":["agents","anthropic","design principles","tools"],"fy":2025,"ly":2025,"sc":8.5,"shelf":"agents","new":true},{"id":"9fubhllmsBU","t":"Field Guide to Fable","ch":"AI Engineer (Thariq Shihipar, Anthropic)","s":"An Anthropic engineer's plain-language explanation of agentic coding: what a harness like Claude Code adds on top of a chat model, and what that unlocks for people who are not programmers.","w":"Three separate World's Fair 2026 recaps picked it as a must-watch, and it was the second most-viewed talk from the conference by August 2026.","type":"talk","lvl":"intermediate","tags":["agents","claude code","anthropic","agentic coding"],"fy":2026,"ly":2026,"sc":8.5,"shelf":"agents","new":true},{"id":"VrpEyglYgeU","t":"In the Land of AI Agents, the Verifiers Are King","ch":"AI Engineer (Tariq Shaukat, Sonar)","s":"Argues that the bottleneck for AI in a real business is not the model but verification: checking what an agent did before it counts. Lays out what a verifier is and where it belongs.","w":"Cited across the 2026 recaps as the talk that gave managers a vocabulary for why agent output needs a check step, which maps directly onto green, yellow and red zones.","type":"talk","lvl":"intermediate","tags":["agents","verification","governance","trust"],"fy":2026,"ly":2026,"sc":8.5,"shelf":"agents","new":true},{"id":"q4Tr-DknG2M","t":"Recursive Model Improvement","ch":"AI Engineer (Lee Robinson, Cursor)","s":"How Cursor runs two nested loops: user feedback refines the evals, and reinforcement learning climbs them, including what goes wrong when models game the evals.","w":"Picked by all three 2026 recap sources as the clearest account of how a product team trains its own model.","type":"talk","lvl":"advanced","tags":["reinforcement learning","evals","cursor","training"],"fy":2026,"ly":2026,"sc":8.5,"shelf":"ml","new":true},{"id":"jEbRVNxL44c","t":"Zipline - How Rwanda Built A Drone Delivery Service","ch":"Real Engineering","s":"Real Engineering explains how Zipline built a drone delivery network for medical supplies in Rwanda. It covers the fixed-wing drones, catapult launch, parachute drop and the logistics behind the service.","w":"HN points to it as the best explanation of how Zipline's whole system works and as an example of drones doing real, useful work.","type":"explainer","lvl":"general","tags":["zipline","drones","rwanda","medical logistics","aviation"],"fy":2019,"ly":2023,"sc":8.45,"shelf":"robotics","new":false},{"id":"9CO6M2HsoIA","t":"Slaughterbots","ch":"Stop Autonomous Weapons","s":"Slaughterbots is a fictional short film, backed by Stuart Russell, depicting swarms of cheap autonomous drones that use facial recognition to assassinate targets. It was made to argue for banning autonomous weapons.","w":"HN invokes it as the standard reference whenever drone swarms, autonomous weapons or military AI come up in the news.","type":"other","lvl":"general","tags":["killer robots","drone swarms","military ai","stuart russell","dystopian fiction"],"fy":2018,"ly":2026,"sc":7.88,"shelf":"bigpicture","new":false},{"id":"ObhK8lUfIlc","t":"GDC 2018: John McDonald (Valve) - Using Deep Learning to Combat Cheating in CSGO","ch":"max4warn","s":"A GDC 2018 talk by Valve's John McDonald on VACNet, a deep learning system that analyzes CS:GO gameplay to detect cheaters. It covers how the model is trained on human-reviewed Overwatch cases and how its output feeds human moderation.","w":"HN points to it as the definitive example of server-side, behavioral, non-invasive anti-cheat using machine learning, often as a counterpoint to kernel-level anti-cheat.","type":"talk","lvl":"intermediate","tags":["vacnet","cs:go","anti-cheat","deep learning","valve","gdc"],"fy":2018,"ly":2026,"sc":7.88,"shelf":"ml","new":false},{"id":"_L3gNaAVjQ4","t":"George Hotz: Hacking the Simulation & Learning to Drive with Neural Nets | Lex Fridman Podcast #132","ch":"Lex Fridman","s":"Lex Fridman interviews George Hotz, founder of Comma.ai, in a three-hour conversation. Topics include self-driving, Tesla and Waymo, crypto, GPT-3, immortality and book recommendations.","w":"Commenters recommend it as an entertaining, wide-ranging interview, and cite it for Hotz's views on end-to-end neural-net driving.","type":"podcast","lvl":"general","tags":["george hotz","comma.ai","lex fridman","self-driving","tesla"],"fy":2020,"ly":2026,"sc":7.88,"shelf":"llm","new":false},{"id":"mbyG85GZ0PI","t":"Lecture 01 - The Learning Problem","ch":"caltech","s":"The first lecture of Caltech's CS 156 Machine Learning course by Yaser Abu-Mostafa, introducing the learning problem and supervised, unsupervised and reinforcement learning. It is one of 18 lectures in the Learning from Data course.","w":"Recommended repeatedly in 'where do I start with ML' threads as a clear, math-aware introduction.","type":"course","lvl":"intro","tags":["machine learning","learning from data","vc dimension","supervised learning","caltech"],"fy":2016,"ly":2023,"sc":7.8,"shelf":"ml","new":false},{"id":"DOWDNBu9DkU","t":"Amazing Invention- This Drone Will Change Everything","ch":"Mark Rober","s":"Mark Rober visits Zipline and looks at its autonomous delivery drones, including medical and blood deliveries in Rwanda. He demonstrates how the drones' propeller design and cable-lowered delivery pod make them quiet and precise.","w":"HN commenters cite it as the clearest look at Zipline's real-world drone operations, especially for medical delivery and the solved noise problem.","type":"demo","lvl":"general","tags":["zipline","drone delivery","mark rober","medical logistics","noise reduction"],"fy":2023,"ly":2024,"sc":7.25,"shelf":"robotics","new":false},{"id":"xoVJKj8lcNQ","t":"The A.I. Dilemma - March 9, 2023","ch":"Center for Humane Technology","s":"Tristan Harris and Aza Raskin of the Center for Humane Technology present why current AI capabilities pose catastrophic societal risks. They argue companies are racing to deploy without adequate safety, and that institutions must adapt. It was given before GPT-4's launch.","w":"Frequently recommended as the overview to get up to speed on AI risk, especially by people who valued The Social Dilemma.","type":"talk","lvl":"general","tags":["ai safety","llm risks","center for humane technology","ai race","regulation"],"fy":2023,"ly":2024,"sc":7.25,"shelf":"bigpicture","new":false},{"id":"iDulhoQ2pro","t":"Attention Is All You Need","ch":"Yannic Kilcher","s":"Yannic Kilcher walks through the Attention Is All You Need paper, which introduced the Transformer, a sequence model based solely on attention without recurrence or convolutions. He reviews the architecture and the paper's translation results.","w":"HN recommends it as an accessible, clear walkthrough of the transformer paper for people with some AI background.","type":"explainer","lvl":"intermediate","tags":["transformers","attention","paper review","deep learning","nlp","yannic kilcher"],"fy":2020,"ly":2025,"sc":7.2,"shelf":"llm","new":false},{"id":"pYXy-A4siMw","t":"Intro to AI Safety, Remastered","ch":"Robert Miles AI Safety ","s":"Robert Miles gives a remastered introduction to AI safety, adapted from a talk in London. It lays out why advanced AI could go wrong and answers common objections.","w":"The default link for newcomers wanting an approachable, entertaining introduction to AI risk and alignment.","type":"talk","lvl":"intro","tags":["ai alignment","ai risk","agi","objective functions","robert miles"],"fy":2022,"ly":2023,"sc":7.15,"shelf":"bigpicture","new":false},{"id":"P-2P3MSZrBM","t":"Joscha Bach: Artificial Consciousness and the Nature of Reality | Lex Fridman Podcast #101","ch":"Lex Fridman","s":"Lex Fridman has a three-hour conversation with Joscha Bach about artificial consciousness, the workings of the mind, intelligence and the possibly simulated nature of reality.","w":"Praised as a long but rewarding conversation, with commenters especially liking Bach's emphasis on modelling as a skill and his bridge between mind and computation.","type":"podcast","lvl":"intermediate","tags":["joscha bach","consciousness","artificial intelligence","simulation","philosophy of mind"],"fy":2020,"ly":2021,"sc":7.0,"shelf":"ml","new":false},{"id":"vq2nnJ4g6N0","t":"Tensorflow and deep learning - without a PhD by Martin Görner","ch":"Devoxx","s":"Martin Gorner's Devoxx talk introduces deep learning with TensorFlow, walking through neural networks for MNIST, how to choose an architecture and how to diagnose networks that fail to train.","w":"Repeatedly called an excellent, practical introduction that covers a lot in two hours and explains issues academic textbooks skip.","type":"tutorial","lvl":"intro","tags":["tensorflow","neural networks","mnist","deep learning","convolutional networks"],"fy":2017,"ly":2020,"sc":7.0,"shelf":"ml","new":false},{"id":"JRuNA2eK7w0","t":"Is AI Safety a Pascal's Mugging?","ch":"Robert Miles AI Safety ","s":"Robert Miles examines whether the case for AI safety is a Pascal's mugging, a tiny probability of an enormous payoff. He argues it is not, because AI risk does not depend on vanishingly small probabilities.","w":"HN points to it as a reasoned rebuttal to the claim that AI risk is just Pascal's wager.","type":"explainer","lvl":"intro","tags":["pascal's mugging","ai risk","robert miles","decision theory","existential risk"],"fy":2019,"ly":2026,"sc":7.0,"shelf":"bigpicture","new":false},{"id":"kJLiOGle3Lw","t":"Agentic AI: A Progression of Language Model Usage","ch":"Stanford Online","s":"A Stanford webinar that lays out the ladder from a plain chat model to retrieval, tool use, and agents that plan and act, with where each step fits and what it adds.","w":"Recommended as the place to start before any hands-on agent material, because it explains why agents matter without assuming you can code.","type":"lecture","lvl":"intro","tags":["agents","stanford","tool use","overview"],"fy":2025,"ly":2025,"sc":7.0,"shelf":"agents","new":true},{"id":"RGe6EjucbzI","t":"The 2026 State of AI Engineering","ch":"AI Engineer (Barr Yaron)","s":"Survey results on what teams actually built in 2026: where agents are in production, what broke, what evals look like, and where the cost went.","w":"A one-talk reality check that two recaps flagged for anyone trying to tell hype from practice.","type":"talk","lvl":"intermediate","tags":["survey","agents in production","evals","cost"],"fy":2026,"ly":2026,"sc":7.0,"shelf":"agents","new":true},{"id":"eBUyTS7SzV4","t":"Every company should have a Brain","ch":"AI Engineer (Garry Tan, Y Combinator)","s":"Y Combinator's president on AI-native companies: why lean teams get outsized leverage, and why how work is wired together matters more than which model you pick.","w":"The closing keynote business people forwarded to each other after World's Fair 2026; one of the top 15 most-viewed talks from the event.","type":"talk","lvl":"general","tags":["business","ai-native","leverage","strategy"],"fy":2026,"ly":2026,"sc":7.0,"shelf":"bigpicture","new":true},{"id":"n97BCfyFIvw","t":"The engineer of the future is the person who is able to choose what is worth doing","ch":"AI Engineer (Addy Osmani)","s":"Addy Osmani's closing keynote on what remains when code is cheap: judgment, accountability and owning the outcome. Applies to any role, not only engineering.","w":"Among the five most-viewed talks of World's Fair 2026 and picked by the keynote index; the title line gets quoted on its own.","type":"talk","lvl":"general","tags":["future of work","judgment","careers"],"fy":2026,"ly":2026,"sc":7.0,"shelf":"bigpicture","new":true},{"id":"xUnRQ9vLXxo","t":"Everything we knew about software has changed","ch":"AI Engineer (Theo Browne)","s":"A closing keynote on building wider rather than deeper as models improve: what changes in how a small team ships when the model does most of the typing.","w":"In the top ten most-viewed talks of the 2026 World's Fair and listed in the keynote index.","type":"talk","lvl":"intermediate","tags":["software","teams","agentic coding"],"fy":2026,"ly":2026,"sc":7.0,"shelf":"agents","new":true},{"id":"pMggiOb18tc","t":"The Golden Age of AI Engineering","ch":"AI Engineer (Alexander Embiricos, Romain Huet, Peter Steinberger, OpenAI)","s":"OpenAI's keynote on models that test their own code and what that makes possible, with live demos.","w":"Top 15 by views and in the keynote index; the demo most people cite when they say the ground shifted in 2026.","type":"talk","lvl":"intermediate","tags":["openai","codex","demos"],"fy":2026,"ly":2026,"sc":7.0,"shelf":"agents","new":true},{"id":"Ib5GBkD555M","t":"Harness Engineering is not Enough: Why Software Factories Fail","ch":"AI Engineer (Dex Horthy, HumanLayer)","s":"A post-mortem on a lights-off agent factory that collapsed, and the argument that no harness fixes a model's training gaps.","w":"Third most-viewed talk of World's Fair 2026 and in the keynote index; the cautionary tale for anyone planning to automate a whole function.","type":"talk","lvl":"advanced","tags":["agents","harness","failure modes"],"fy":2026,"ly":2026,"sc":7.0,"shelf":"agents","new":true},{"id":"-CnA2lGfymY","t":"I've never seen anything scarier than an LLM with tool calls","ch":"AI Engineer (Erik Meijer)","s":"Erik Meijer argues agents should produce checkable, proof-carrying programs rather than being trusted blindly or graded by another LLM.","w":"Listed in both the views ranking and the keynote index; the strongest version of the case for verification.","type":"talk","lvl":"advanced","tags":["verification","agents","formal methods"],"fy":2026,"ly":2026,"sc":7.0,"shelf":"agents","new":true},{"id":"GgLQ02aO-hs","t":"The Unreasonable Effectiveness of Separating the Task from the Model","ch":"AI Engineer (Maxime Rivest, Isaac Miller, DSPy)","s":"The DSPy approach: define the task signature first, keep prompts and models swappable, and let optimization find the prompt.","w":"Appears in both the views ranking and the keynote index as the practical answer to prompt fragility.","type":"talk","lvl":"advanced","tags":["dspy","prompting","optimization"],"fy":2026,"ly":2026,"sc":7.0,"shelf":"llm","new":true},{"id":"1P1hJ36rxM0","t":"Software engineering is not about writing code","ch":"AI Engineer (Benoit Schillings, Google DeepMind)","s":"A DeepMind keynote on code generation, deep-thinking models and Gemini pre-training, and what it means when writing code is nearly free.","w":"Listed in both the views ranking and the keynote index.","type":"talk","lvl":"intermediate","tags":["deepmind","code generation","future of work"],"fy":2026,"ly":2026,"sc":7.0,"shelf":"bigpicture","new":true},{"id":"q2JrUKBMf0w","t":"The Future of Evals: From LLM as a Judge to Agent as a Judge","ch":"AI Engineer (Aparna Dhinakaran, Arize)","s":"Why evaluation has to evolve as fast as the agents it grades, and what agent-as-judge looks like in practice.","w":"In both the views ranking and the keynote index; the short introduction to evals for people who run AI rather than build it.","type":"talk","lvl":"intermediate","tags":["evals","quality","agents"],"fy":2026,"ly":2026,"sc":7.0,"shelf":"agents","new":true},{"id":"RmIgJ64z6Y4","t":"34C3 -  Dude, you broke the Future!","ch":"media.ccc.de","s":"Charles Stross's 34C3 talk on how science fiction gets the future wrong. He argues that corporations are slow, human-powered AIs that optimize for profit, and that regulation lags behind accelerating technology.","w":"HN returns to it whenever corporations as paperclip-maximizer AIs or AI risk come up.","type":"talk","lvl":"general","tags":["charles stross","ai risk","corporations","science fiction","regulation"],"fy":2018,"ly":2024,"sc":6.52,"shelf":"bigpicture","new":false},{"id":"DleXA5ADG78","t":"Brains, Sex, and Machine Learning","ch":"Google TechTalks","s":"Geoffrey Hinton's Google Tech Talk linking machine learning to biology, explaining dropout as a form of model averaging and why sexual reproduction and stochastic spiking neurons may prevent overfitting.","w":"Recommended as a simple, elegant explanation of dropout and its connection to how the brain works.","type":"talk","lvl":"intermediate","tags":["geoffrey hinton","dropout","deep learning","spiking neurons","overfitting"],"fy":2012,"ly":2017,"sc":6.5,"shelf":"ml","new":false},{"id":"L5pUA3LsEaw","t":"Why Not Just: Think of AGI Like a Corporation?","ch":"Robert Miles AI Safety ","s":"Robert Miles asks whether corporations can be thought of as artificial general superintelligences and how far the analogy holds. He argues the differences, such as being limited by employees' capabilities, matter for AI safety.","w":"HN points to it as an accessible rebuttal to the claim that AI risk is just corporate risk.","type":"explainer","lvl":"general","tags":["agi","corporations","ai risk","superintelligence","robert miles"],"fy":2020,"ly":2023,"sc":6.5,"shelf":"bigpicture","new":false},{"id":"Qi1Yry33TQE","t":"Ali Rahimi's talk at NIPS(NIPS 2017 Test-of-time award presentation)","ch":"ObservantVids","s":"Ali Rahimi's NIPS 2017 Test-of-Time award talk, in which he argues that modern machine learning has become alchemy, working without being rigorously understood. He calls for more rigor and understanding in the field.","w":"HN points to it as the sharpest, most-quoted critique of deep learning practice, and cites it whenever ML reliability or understanding is debated.","type":"talk","lvl":"intermediate","tags":["deep learning","nips","alchemy","ali rahimi","ml research"],"fy":2018,"ly":2025,"sc":6.4,"shelf":"ml","new":false},{"id":"5MdSE-N0bxs","t":"Connections between physics and deep learning","ch":"MITCBMM","s":"Max Tegmark's MIT CBMM talk on why deep learning works so well. He argues that the laws of physics and the symmetries in physical data make it compressible into small neural networks, tying physics to network structure.","w":"HN recommends it as a clear argument for why deep and cheap learning succeeds, often alongside the associated paper.","type":"talk","lvl":"intermediate","tags":["deep learning","physics","tegmark","neural networks","symmetry"],"fy":2016,"ly":2026,"sc":6.12,"shelf":"ml","new":false},{"id":"IJcuQQ1eWWI","t":"Playing to Lose: AI and \"Civilization\" (Soren Johnson)","ch":"Google TechTalks","s":"Soren Johnson, a lead designer and AI programmer on Civilization games, discusses game AI design, including whether AI should cheat, behave like a human or be fun rather than strong. He uses the Civilization series as a case study.","w":"HN commenters recommend it as the key talk on why game AI is built to be fun rather than to win.","type":"talk","lvl":"intermediate","tags":["civilization","game ai","game design","soren johnson","ai cheating"],"fy":2011,"ly":2026,"sc":6.12,"shelf":"llm","new":false},{"id":"PFUZlNQIndo","t":"DuckDB – The SQLite for Analytics (Mark Raasveldt, CWI)","ch":"CMU Database Group","s":"Mark Raasveldt of CWI presents DuckDB, an embeddable analytical database described as the SQLite for analytics, in a CMU Database Group talk. He explains why it was built, what it is for, and how it works.","w":"HN recommends it as the definitive talk from the implementer on DuckDB's motivation and design.","type":"talk","lvl":"intermediate","tags":["duckdb","sqlite","olap","embedded database","data science"],"fy":2020,"ly":2026,"sc":6.12,"shelf":"data","new":false},{"id":"efPrtcLdcdM","t":"GPT-4chan: This is the worst AI ever","ch":"Yannic Kilcher","s":"Yannic Kilcher explains how he trained a language model on three years of 4chan's /pol/ board and let it post on the site. He evaluates how good the model was and discusses the ethics and fallout.","w":"HN points to it as the definitive account of the GPT-4chan experiment, used in debates on LLM bias, truthfulness and bots.","type":"commentary","lvl":"general","tags":["gpt-4chan","language models","4chan","ai ethics","fine-tuning"],"fy":2022,"ly":2026,"sc":6.12,"shelf":"bigpicture","new":false},{"id":"yvDCzhbjYWs","t":"Peter Norvig - The Unreasonable Effectiveness of Data","ch":"UBC Computer Science","s":"Peter Norvig's UBC distinguished lecture on how massive amounts of simple data, such as for language modeling, can yield understanding that hand-built models could not.","w":"Recommended as a favorite tech talk showing how huge data availability changes how problems are approached.","type":"lecture","lvl":"intermediate","tags":["peter norvig","big data","nlp","machine learning","google"],"fy":2012,"ly":2026,"sc":6.12,"shelf":"data","new":false},{"id":"J-M98KLgaUU","t":"Drone Delivery Was Supposed to be the Future. What Went Wrong?","ch":"Wendover Productions","s":"Wendover Productions explains why drone delivery has not become the future once promised. It covers no-fly zones, costs and the difficulty of the final drop-off.","w":"It is the standard answer when someone asks what happened to drone delivery.","type":"explainer","lvl":"general","tags":["drones","logistics","last-mile delivery","faa regulation","wendover"],"fy":2022,"ly":2023,"sc":5.85,"shelf":"robotics","new":false},{"id":"S27pHKBEp30","t":"LSTM is dead. Long Live Transformers!","ch":"Seattle Applied Deep Learning","s":"Leo Dirac explains how transformer models have replaced LSTMs in NLP. He gives background from bag of words through RNNs and LSTMs, then a technical deep dive into multi-headed self-attention and positional encoding, with sample code.","w":"Recommended as a clear explanation of why RNNs fall short and how transformers solve it.","type":"lecture","lvl":"intermediate","tags":["transformers","lstm","self-attention","positional encoding","nlp"],"fy":2020,"ly":2023,"sc":5.85,"shelf":"llm","new":false},{"id":"ne99laPUxN4","t":"The Simple Essence of Automatic Differentiation - Conal Elliott","ch":"Microsoft Research","s":"Conal Elliott presents his paper on automatic differentiation, deriving a simple, generalized AD algorithm from a natural specification. Varying the representation of derivatives yields forward and reverse mode algorithms that avoid the complexity and statefulness of backpropagation.","w":"Cited as the more accessible companion to the paper, and as the most interesting treatment of AD via compiling to categories.","type":"talk","lvl":"advanced","tags":["automatic differentiation","functional programming","category theory","backpropagation","haskell"],"fy":2018,"ly":2023,"sc":5.85,"shelf":"ml","new":false},{"id":"Ilg3gGewQ5U","t":"Backpropagation, intuitively | Deep Learning Chapter 3","ch":"3Blue1Brown","s":"3Blue1Brown's third deep learning chapter gives a visual, intuitive account of backpropagation and what happens to a neural network as it learns. A follow-up video covers the underlying calculus.","w":"Recommended as the most accessible introduction to backprop for people who struggle with the pure math.","type":"explainer","lvl":"intro","tags":["neural networks","backpropagation","deep learning","gradient descent","visual math"],"fy":2017,"ly":2024,"sc":5.8,"shelf":"ml","new":false},{"id":"ZMQbHMgK2rw","t":"The Fastest Maze-Solving Competition On Earth","ch":"Veritasium","s":"Veritasium visits the Micromouse competition, where small autonomous robots race to solve mazes at extreme speed. It covers the competition's technical history, including fans and ground effect enabling very high-g turns.","w":"HN commenters point to it as a nice introduction to micromouse and maze-racing robots.","type":"documentary","lvl":"general","tags":["micromouse","maze solving","robotics","ground effect","competition"],"fy":2023,"ly":2024,"sc":5.8,"shelf":"robotics","new":false},{"id":"1tSqSMOyNFE","t":"How We’re Fooled By Statistics","ch":"Veritasium","s":"A short Veritasium video explaining regression to the mean and how ignoring it leads to wrong conclusions about punishment versus reward, medical treatments and sports streaks. It uses the example of flight instructors who believed criticism works better than praise.","w":"HN commenters link it as the clearest short explanation of regression to the mean, especially for how it misleads people about feedback and statistics.","type":"explainer","lvl":"general","tags":["statistics","feedback","psychology","cognitive bias","veritasium"],"fy":2014,"ly":2025,"sc":5.6,"shelf":"data","new":false},{"id":"5Dnw46eC-0o","t":"John Rauser keynote: \"Statistics Without the Agonizing Pain\" -- Strata + Hadoop 2014","ch":"O'Reilly","s":"John Rauser's Strata 2014 keynote argues that programmers can learn statistics quickly by simulating random processes and shuffling data. He explains p-values through resampling rather than formulas.","w":"HN recommends it as an excellent programmer-friendly explanation of p-values and statistical thinking.","type":"talk","lvl":"intro","tags":["p-values","simulation","resampling","data science","bootstrapping"],"fy":2015,"ly":2025,"sc":5.6,"shelf":"data","new":false},{"id":"IkdziSLYzHw","t":"Will AI outsmart human intelligence? - with 'Godfather of AI' Geoffrey Hinton","ch":"The Royal Institution","s":"Geoffrey Hinton's 2025 Royal Institution Discourse on what AI has learned from biological intelligence and how it might surpass it. He explains neural networks and argues that human language use resembles what backpropagation-trained networks do.","w":"HN readers recommend it as an intuitive explanation of neural networks that shifts how people think about AI and language.","type":"lecture","lvl":"general","tags":["geoffrey hinton","neural networks","backpropagation","ai risk","language models"],"fy":2025,"ly":2025,"sc":5.6,"shelf":"bigpicture","new":false},{"id":"bZQun8Y4L2A","t":"State of GPT | BRK216HFS","ch":"Microsoft Developer","s":"Andrej Karpathy walks through how GPT assistants are trained, from tokenization and pretraining to supervised finetuning and RLHF, and gives practical prompting strategies and a view of the tooling ecosystem.","w":"Recommended as the clearest overview of how LLM assistants are built and how to use them well.","type":"talk","lvl":"intro","tags":["llm","rlhf","prompting","karpathy","fine-tuning"],"fy":2023,"ly":2025,"sc":5.6,"shelf":"llm","new":false},{"id":"cdiD-9MMpb0","t":"Andrej Karpathy: Tesla AI, Self-Driving, Optimus, Aliens, and AGI | Lex Fridman Podcast #333","ch":"Lex Fridman","s":"Lex Fridman's long conversation with Andrej Karpathy on Tesla's self-driving approach, Optimus, neural networks, transformers, AGI and more. It also touches on his OpenAI work and working with Elon Musk.","w":"Cited for Karpathy's first-hand explanations of camera-only self-driving, transformers and ML trends.","type":"podcast","lvl":"intermediate","tags":["andrej karpathy","tesla autopilot","transformers","agi","lex fridman"],"fy":2022,"ly":2025,"sc":5.6,"shelf":"bigpicture","new":false},{"id":"pMYYx_im5QI","t":"Daniel Suarez: The kill decision shouldn't belong to a robot","ch":"TED","s":"Novelist Daniel Suarez warns in a TED talk about autonomous robotic weapons and drones. He argues that taking the decision to kill out of human hands threatens democratic accountability and calls for a ban.","w":"Plugged repeatedly in AI and drone threads as a concise statement of the dangers of automated weapons.","type":"talk","lvl":"general","tags":["drones","killer robots","ai ethics","ted","daniel suarez"],"fy":2013,"ly":2025,"sc":5.6,"shelf":"bigpicture","new":false},{"id":"NVl9_6J1G60","t":"Breakout Session Presented by Dan Robinson of Heap at PGConf Silicon Valley 2015","ch":"Citus Data","s":"Dan Robinson, then CTO of Heap, explains how Heap runs its product analytics backend on PostgreSQL and CitusDB. It covers conversion funnels, retention and cohort queries across tens of billions of events and 50 TB of data, plus Postgres and Kafka together and the indexing strategy.","w":"HN commenters point to it as a first-hand overview of Heap's Postgres/Citus architecture and its distinctive indexing approach.","type":"talk","lvl":"advanced","tags":["postgresql","citus","event analytics","indexing","kafka","sharding"],"fy":2016,"ly":2020,"sc":5.5,"shelf":"data","new":false},{"id":"cQO2XTP7QDw","t":"I sent robot forgeries to a handwriting expert","ch":"Stuff Made Here","s":"Stuff Made Here builds a handwriting robot and sends its output to a handwriting expert to see whether it can pass as forgery. It walks through the engineering trials and errors along the way.","w":"HN commenters recommend it as the physical-machine take on replicating handwriting, with plenty of explained iteration.","type":"demo","lvl":"general","tags":["pen plotter","handwriting","forgery","robotics","engineering"],"fy":2024,"ly":2026,"sc":5.25,"shelf":"robotics","new":false},{"id":"pdJQ8iVTwj8","t":"Chris Lattner: Future of Programming and AI | Lex Fridman Podcast #381","ch":"Lex Fridman","s":"Lex Fridman's long podcast conversation with Chris Lattner, creator of Swift, LLVM and MLIR and co-founder of Modular. Topics include the Mojo language, the future of programming and AI, and hardware.","w":"HN points to it as the main place to hear Lattner's motivations and design ideas for Mojo, often in Mojo threads.","type":"podcast","lvl":"intermediate","tags":["chris lattner","mojo","llvm","swift","compilers","lex fridman"],"fy":2023,"ly":2026,"sc":5.25,"shelf":"llm","new":false},{"id":"PYylPRX6z4Q","t":"Training AI Without Writing A Reward Function, with Reward Modelling","ch":"Robert Miles AI Safety ","s":"Robert Miles explains reward modelling, in which an AI learns what humans want from their preference comparisons instead of a hand-written reward function. He walks through the 2017 paper on deep reinforcement learning from human preferences.","w":"Seen as the clearest introduction to the idea behind RLHF, and also borrowed by commenters as a metaphor for human motivation.","type":"explainer","lvl":"intermediate","tags":["rlhf","reinforcement learning","ai safety","reward function","human preferences"],"fy":2019,"ly":2023,"sc":5.2,"shelf":"bigpicture","new":false},{"id":"IHZwWFHWa-w","t":"Gradient descent, how neural networks learn | Deep Learning Chapter 2","ch":"3Blue1Brown","s":"3Blue1Brown's second neural network chapter explains cost functions and gradient descent, showing how a network learns by minimizing error during training.","w":"HN recommends the series as a very accessible way to understand the math behind machine learning and LLMs.","type":"explainer","lvl":"intro","tags":["neural networks","cost function","machine learning","3blue1brown","deep learning"],"fy":2017,"ly":2024,"sc":5.08,"shelf":"ml","new":false},{"id":"dKJEpOtVgXc","t":"Do we need Attention? - Linear RNNs and State Space Models (SSMs) for NLP","ch":"Sasha Rush","s":"Sasha Rush surveys linear RNNs and state space models as replacements for attention in transformer-style models, in a talk for MLSys. It covers the foundational preliminaries and predates Mamba.","w":"HN commenters recommend it for building intuition about state space models and the tradeoffs between transformer and SSM layers.","type":"lecture","lvl":"advanced","tags":["ssm","linear rnn","mamba","attention","s4","nlp"],"fy":2024,"ly":2024,"sc":5.08,"shelf":"llm","new":false},{"id":"gFEE3w7F0ww","t":"Travis Oliphant: NumPy, SciPy, Anaconda, Python & Scientific Programming | Lex Fridman Podcast #224","ch":"Lex Fridman","s":"Lex Fridman interviews Travis Oliphant, creator of NumPy, SciPy and Anaconda, about scientific Python and its history. It runs about three hours.","w":"Recommended as a firsthand history of how Python became the language of scientific computing and ML.","type":"podcast","lvl":"intermediate","tags":["numpy","scipy","anaconda","python","travis oliphant"],"fy":2021,"ly":2024,"sc":5.08,"shelf":"data","new":false},{"id":"pHsuIaPbNbY","t":"Efficient Bayesian inference with Hamiltonian Monte Carlo -- Michael Betancourt (Part 1)","ch":"MLSS Iceland 2014","s":"Michael Betancourt's lecture at MLSS Iceland 2014 on efficient Bayesian inference with Hamiltonian Monte Carlo. It covers MCMC concepts and motivation and the problems of sampling in high-dimensional distributions.","w":"HN people return to it as a clear primer on MCMC and HMC and on why high dimensions behave counterintuitively.","type":"lecture","lvl":"advanced","tags":["mcmc","bayesian inference","hmc","high-dimensional sampling","statistics"],"fy":2017,"ly":2024,"sc":5.08,"shelf":"data","new":false},{"id":"Ki_Af_o9Q9s","t":"7 Minutes of Terror: The Challenges of Getting to Mars","ch":"NASA Jet Propulsion Laboratory","s":"NASA JPL engineers explain the final minutes of the Curiosity rover's entry, descent and landing on Mars, which combined aerobraking, a parachute and a skycrane. It lays out the tight sequence of events that all had to work automatically.","w":"HN repeatedly recommends it as an awe-inspiring example of engineering problem solving and the go-to explainer on how Mars landings work.","type":"documentary","lvl":"general","tags":["curiosity rover","skycrane","entry descent landing","nasa jpl","space engineering"],"fy":2016,"ly":2021,"sc":5.0,"shelf":"robotics","new":false},{"id":"cKxRvEZd3Mw","t":"Hello World - Machine Learning Recipes #1","ch":"Google for Developers","s":"The first episode of Google's Machine Learning Recipes series. It introduces machine learning and builds a first supervised-learning classifier in about six lines of Python using scikit-learn.","w":"HN commenters recommend it as an approachable first step into machine learning for programmers, with an apples-vs-oranges classifier.","type":"tutorial","lvl":"intro","tags":["machine learning","scikit-learn","python","classifier","supervised learning"],"fy":2016,"ly":2017,"sc":5.0,"shelf":"ml","new":false},{"id":"Iq9DzN6mvYA","t":"Jake Vanderplas - Statistics for Hackers - PyCon 2016.mp4","ch":"PyCon 2016","s":"Jake Vanderplas's PyCon 2016 talk shows how sampling-based methods such as simulation, shuffling and bootstrapping make statistics intuitive. His pitch is that if you can write a for-loop, you can do statistics.","w":"It is recommended to programmers who want to learn statistics through simulation rather than memorized formulas.","type":"talk","lvl":"intro","tags":["bootstrapping","resampling","python","simulation","hypothesis testing"],"fy":2017,"ly":2025,"sc":4.8,"shelf":"data","new":false},{"id":"jPhJbKBuNnA","t":"I tried using AI. It scared me.","ch":"Tom Scott","s":"Tom Scott describes trying ChatGPT to fix his Gmail labels and being unsettled by how capable it was. He argues that we cannot know where we are on the technology curve, comparing it to Napster.","w":"HN cites it for the argument that we cannot tell whether AI is early or late on its sigmoid curve until hindsight.","type":"commentary","lvl":"general","tags":["chatgpt","tom scott","technology curve","ai progress","napster"],"fy":2023,"ly":2025,"sc":4.8,"shelf":"bigpicture","new":false},{"id":"rURRYI66E54","t":"AI Language Models & Transformers - Computerphile","ch":"Computerphile","s":"Rob Miles explains on Computerphile how language models generate plausible text and how transformers improved on recurrent networks, with a look at what comes next.","w":"Pointed to as an accessible early explanation of transformers and attention for newcomers.","type":"explainer","lvl":"intro","tags":["transformers","language models","attention","rob miles","computerphile"],"fy":2022,"ly":2025,"sc":4.8,"shelf":"llm","new":false},{"id":"HMBl0ttu-Ow","t":"Social Physics: How Good Ideas Spread | Sandy Pentland | Talks at Google","ch":"Talks at Google","s":"Sandy Pentland of MIT presents his book on social physics, which uses big-data studies of how people interact to explain how ideas spread. He describes findings on collective intelligence and social influence.","w":"HN commenters recommend it for the finding that idea exchange and connectivity drive productivity and wealth, which they apply to cities, offices and online communities.","type":"talk","lvl":"general","tags":["social physics","idea flow","collective intelligence","cities","big data"],"fy":2015,"ly":2022,"sc":4.6,"shelf":"data","new":false},{"id":"PmlRbfSavbI","t":"Stealing Baseball Signs with a Phone (Machine Learning)","ch":"Mark Rober","s":"Mark Rober and Jabril build an app that uses machine learning to decode baseball sign sequences. The video also explains neural networks in a simple, accessible way.","w":"Recommended as a fun, semi-technical intro to machine learning that holds kids' attention.","type":"demo","lvl":"general","tags":["neural networks","baseball","sign stealing","mark rober","machine learning"],"fy":2019,"ly":2023,"sc":4.55,"shelf":"ml","new":false},{"id":"e3K5UxWRRuY","t":"35C3 -  The Ghost in the Machine","ch":"media.ccc.de","s":"Joscha Bach's 35C3 talk offers an AI perspective on the soul, covering perception, mental representation, agency, consciousness and selfhood as things that can arise in a computational system like the brain.","w":"HN commenters call it lucid and fluff-free, and point to Bach as one of the most interesting voices on AI, consciousness and philosophy.","type":"talk","lvl":"intermediate","tags":["consciousness","artificial intelligence","joscha bach","cognitive science","philosophy of mind"],"fy":2020,"ly":2023,"sc":4.55,"shelf":"bigpicture","new":false},{"id":"g6bOwQdCJrc","t":"[CVPR'21 WAD] Keynote - Andrej Karpathy, Tesla","ch":"WAD at CVPR","s":"Andrej Karpathy, then Tesla's head of AI, gives a CVPR 2021 workshop keynote on how Autopilot's neural networks work. A large part covers the move to a vision-only stack, including why radar was dropped and how vision produces depth estimates.","w":"HN commenters cite it in Tesla threads as the primary-source explanation of why Tesla removed radar and how its vision depth estimation works.","type":"talk","lvl":"advanced","tags":["tesla autopilot","computer vision","radar removal","depth estimation","self-driving","neural networks"],"fy":2021,"ly":2023,"sc":4.55,"shelf":"ml","new":false},{"id":"Ly6USRwTHe0","t":"Generative AI for Krita - With ControlNet","ch":"Acly","s":"A short demo of an open-source Krita plugin that uses Stable Diffusion to generate and edit images in place. It adds ControlNet scribble and line-art support. You select an area, push a button, and new matching content appears.","w":"HN recommends it as a quick way to see how the Krita AI plugin works and how it matches Photoshop's generative features.","type":"demo","lvl":"general","tags":["krita","stable diffusion","controlnet","inpainting","open source"],"fy":2023,"ly":2026,"sc":4.38,"shelf":"llm","new":false},{"id":"ODSJsviD_SU","t":"Tesla AI Day 2022","ch":"Tesla","s":"Tesla's full AI Day 2022 presentation, covering the Tesla Bot hardware and software, Full Self-Driving planning and perception, the occupancy network, data labeling, simulation and Dojo training infrastructure.","w":"HN uses it as primary-source evidence on how Tesla's vision-based driving stack and training pipeline work.","type":"talk","lvl":"advanced","tags":["tesla","full self-driving","occupancy networks","dojo","humanoid robots","computer vision"],"fy":2022,"ly":2026,"sc":4.38,"shelf":"robotics","new":false},{"id":"P_fHJIYENdI","t":"AlphaFold - The Most Useful Thing AI Has Ever Done","ch":"Veritasium","s":"A Veritasium video on AlphaFold, DeepMind's protein structure prediction system, and the protein design work it enabled. It features interviews with John Jumper and with David Baker's protein design group and explains why predicting protein folding matters.","w":"HN points to it as an accessible explainer of AlphaFold and as an example of narrow, specialized ML delivering real scientific breakthroughs.","type":"explainer","lvl":"general","tags":["alphafold","deepmind","proteins","machine learning","nobel prize"],"fy":2025,"ly":2026,"sc":4.38,"shelf":"ml","new":false},{"id":"BCiZc0n6COY","t":"Lecture 1: Introduction to Information Theory","ch":"Jakob Foerster","s":"The first lecture of David MacKay's Cambridge course on information theory, pattern recognition and neural networks. It introduces the core ideas of his book Information Theory, Inference, and Learning Algorithms.","w":"Recommended as an enjoyable, clear introduction to information theory that pairs well with MacKay's free textbook.","type":"lecture","lvl":"intro","tags":["david mackay","information theory","inference","machine learning","cambridge"],"fy":2018,"ly":2024,"sc":4.35,"shelf":"ml","new":false},{"id":"I845O57ZSy4","t":"John Carmack: Doom, Quake, VR, AGI, Programming, Video Games, and Rockets | Lex Fridman Podcast #309","ch":"Lex Fridman","s":"A five-hour Lex Fridman interview with John Carmack covering Doom, Quake, VR, AGI, programming, games and rockets.","w":"HN mines it for Carmack's views on AGI, debuggers, tooling and the metaverse, and recommends it as worth the full length.","type":"interview","lvl":"intermediate","tags":["john carmack","agi","id software","programming","virtual reality"],"fy":2022,"ly":2024,"sc":4.35,"shelf":"bigpicture","new":false},{"id":"PaCmpygFfXo","t":"The spelled-out intro to language modeling: building makemore","ch":"Andrej Karpathy","s":"Andrej Karpathy builds a bigram character-level language model from scratch, introducing torch.Tensor and the framework of training, sampling and loss evaluation. It is the first step toward a GPT-style Transformer in his makemore series.","w":"HN recommends it as a long but worthwhile way to demystify how language models work technically.","type":"tutorial","lvl":"intro","tags":["makemore","bigram","pytorch","neural networks","karpathy","next-token prediction"],"fy":2024,"ly":2024,"sc":4.35,"shelf":"llm","new":false},{"id":"Ucp0TTmvqOE","t":"Tesla Autonomy Day","ch":"Tesla","s":"Tesla's roughly four-hour Autonomy Day investor event covers its in-house self-driving chip, vision-only approach, fleet data engine and robotaxi plans, with talks by Andrej Karpathy and others.","w":"Cited both as a primary source on Tesla's technical autonomy approach and, later, for Musk's unmet robotaxi predictions.","type":"talk","lvl":"intermediate","tags":["autonomy day","fsd chip","vision vs lidar","data engine","robotaxi"],"fy":2019,"ly":2024,"sc":4.35,"shelf":"ml","new":false},{"id":"lnA9DMvHtfI","t":"Large Language Models from scratch","ch":"Graphics in 5 Minutes","s":"A short cartoon explaining how language models like ChatGPT work, from autocomplete and conditional probabilities to neural networks, gradient descent and backpropagation. It shows how networks act as universal function approximators.","w":"HN recommends it as a quick, non-dumbed-down way to get up to speed on how LLMs are built.","type":"explainer","lvl":"intro","tags":["llm","neural networks","gradient descent","transformers","language modeling"],"fy":2023,"ly":2024,"sc":4.35,"shelf":"llm","new":false},{"id":"tcdVC4e6EV4","t":"Deadly Truth of General AI? - Computerphile","ch":"Computerphile","s":"A Computerphile video in which Rob Miles warns against assuming a general AI will think like a human. He uses the thought experiment of a stamp-collecting AI to show how a goal-driven system could pursue its aim with dangerous results.","w":"Often pointed to as an accessible first introduction to AI safety and why a superintelligent AI could be dangerous.","type":"explainer","lvl":"general","tags":["agi","rob miles","stamp collector","superintelligence","alignment"],"fy":2016,"ly":2024,"sc":4.35,"shelf":"bigpicture","new":false},{"id":"RUwS1uAdUcI","t":"Hans Rosling: Debunking third-world myths with the best stats you've ever seen","ch":"TED","s":"In this TED talk, Hans Rosling uses animated statistics to debunk myths about the developing world, showing how health and income have improved across countries.","w":"Cited as the classic data-driven argument that the first-world versus third-world split is outdated and the world is improving.","type":"talk","lvl":"general","tags":["hans rosling","data visualization","global health","development","ted"],"fy":2011,"ly":2017,"sc":4.0,"shelf":"data","new":false},{"id":"h2I8AoB1xgU","t":"7 Minutes of Terror: Curiosity Rover's Risky Mars Landing | Video","ch":"VideoFromSpace","s":"NASA's animated explanation of the Curiosity rover's Mars landing, which relied on a complex automated sequence taking about seven minutes. It features NASA engineers discussing the risks before the 2012 landing.","w":"Recommended as a gripping look at how hard Mars entry, descent and landing is.","type":"explainer","lvl":"general","tags":["curiosity","nasa","mars landing","sky crane","entry descent landing"],"fy":2017,"ly":2021,"sc":4.0,"shelf":"robotics","new":false},{"id":"nnR8fDW3Ilo","t":"Festo - SmartBird","ch":"Festo","s":"Festo's promo video of SmartBird, an ultralight bionic flying model inspired by the herring gull that flaps and twists its wings to fly. It shows the robot in flight.","w":"Recommended as a spectacular example of bio-inspired robotics, and a pointer to Festo's other work.","type":"demo","lvl":"general","tags":["festo","smartbird","ornithopter","bionics","robotics"],"fy":2011,"ly":2021,"sc":4.0,"shelf":"robotics","new":false},{"id":"B8ofWFx525s","t":"Beware online \"filter bubbles\" | Eli Pariser","ch":"TED","s":"Eli Pariser's TED talk argues that personalized web services trap users in filter bubbles, limiting exposure to challenging information. He says this is bad for individuals and for democracy.","w":"Cited as an early, prescient warning about algorithmic personalization, echo chambers and social media feeds.","type":"talk","lvl":"general","tags":["personalization","echo chambers","algorithms","social media","ted"],"fy":2012,"ly":2025,"sc":4.0,"shelf":"llm","new":false},{"id":"SGUCcjHTmGY","t":"OpenAI Codex Live Demo","ch":"OpenAI","s":"OpenAI's 2021 live demo of Codex translating natural language into code, from Hello World to building a game and a Microsoft Word plugin.","w":"HN points to it as an early, impressive demonstration that LLMs could write working code.","type":"demo","lvl":"general","tags":["openai codex","code generation","llm","live demo","programming"],"fy":2021,"ly":2025,"sc":4.0,"shelf":"llm","new":false},{"id":"lFXJWPhDsSY","t":"Alpha Zero's \"Immortal Zugzwang Game\" against Stockfish","ch":"agadmator's Chess Channel","s":"agadmator walks through a game in which DeepMind's AlphaZero beat Stockfish, ending in a famous zugzwang that leaves the engine helpless. He gives commentary on the moves.","w":"HN recommends it as a striking, well-explained example of AlphaZero's play and of zugzwang.","type":"explainer","lvl":"intro","tags":["alphazero","stockfish","zugzwang","chess","deepmind"],"fy":2019,"ly":2025,"sc":4.0,"shelf":"ml","new":false},{"id":"rsxCZAE8QNA","t":"HC2023-K2: Hardware for Deep Learning","ch":"hotchipsvideos","s":"Bill Dally of NVIDIA gives a Hot Chips 2023 keynote on the challenges of building hardware for deep learning. He traces how GPUs have been adapted for machine learning, especially training.","w":"HN points to it for an insider view of why NVIDIA's AI gains came from design rather than process, and for data-movement power costs.","type":"talk","lvl":"advanced","tags":["nvidia","gpu architecture","hot chips","ai accelerators","bill dally"],"fy":2023,"ly":2025,"sc":4.0,"shelf":"ml","new":false},{"id":"tF4DML7FIWk","t":"Atlas | Partners in Parkour","ch":"Boston Dynamics","s":"Boston Dynamics shows its Atlas humanoid robots running a parkour course with jumps, balance beams and vaults, demonstrating whole-body athletic control.","w":"HN uses it as the benchmark for state-of-the-art humanoid robot mobility in discussions of Tesla Bot and robotics progress.","type":"demo","lvl":"general","tags":["boston dynamics","atlas","parkour","legged robots","humanoid robots"],"fy":2021,"ly":2025,"sc":4.0,"shelf":"robotics","new":false},{"id":"G4hL5Om4IJ4","t":"Jim Keller: The Future of Computing, AI, Life, and Consciousness | Lex Fridman Podcast #162","ch":"Lex Fridman","s":"Lex Fridman interviews chip architect Jim Keller about the future of computing, AI, life and consciousness. Topics include CPU design, branch prediction and why imperfect designs often win.","w":"Recommended as a stellar long-form interview with someone both very smart and wise, mined for CPU design insights.","type":"interview","lvl":"intermediate","tags":["jim keller","cpu design","branch prediction","lex fridman","ai hardware"],"fy":2021,"ly":2023,"sc":3.9,"shelf":"bigpicture","new":false},{"id":"HStF1RJOyxI","t":"Making R run Fast - Jan Vitek","ch":"John O'Neill","s":"Jan Vitek of Northeastern University gives a meetup talk on the challenges of making the R language run fast. He explains what in R's semantics makes it hard to optimize and to build a JIT compiler for.","w":"HN points to it as the best explanation of why R is so hard to optimize, often paired with talks on Python and used to explain what makes Julia fast.","type":"talk","lvl":"advanced","tags":["r language","jit compilers","performance","language semantics","fastr"],"fy":2017,"ly":2023,"sc":3.9,"shelf":"data","new":false},{"id":"OyFJWRnt_AY","t":"CS480/680 Lecture 19: Attention and Transformer Networks","ch":"Pascal Poupart","s":"A university lecture from CS480/680 by Pascal Poupart on attention mechanisms and transformer networks, worked through on a chalkboard.","w":"HN recommends it to newcomers as a clear explanation of attention, including the query/key/value intuition.","type":"lecture","lvl":"intermediate","tags":["attention","transformers","query key value","deep learning","nlp"],"fy":2023,"ly":2023,"sc":3.9,"shelf":"llm","new":false},{"id":"03ccxwNssmo","t":"BFS-Auto: High Speed Book Scanner at over 250 pages/min","ch":"Ishikawa Group Laboratory","s":"A University of Tokyo lab demonstrates a book scanner that flips pages automatically and scans over 250 pages per minute. It uses real-time 3D recognition of the flipped pages and then restores each one to a flat image.","w":"HN commenters bring it up in book-scanning threads as an impressive non-destructive method that is not commercially available.","type":"demo","lvl":"general","tags":["book digitization","page flipping","3d sensing","university of tokyo","scanner"],"fy":2018,"ly":2024,"sc":3.62,"shelf":"robotics","new":false},{"id":"F3Jd9GI6XqE","t":"Edward Gibson: Human Language, Psycholinguistics, Syntax, Grammar & LLMs | Lex Fridman Podcast #426","ch":"Lex Fridman","s":"Lex Fridman interviews MIT psycholinguist Edward Gibson on human language, syntax, grammar and how large language models relate to them. The conversation runs nearly three hours.","w":"HN cites it for the argument that LLMs capture language form but not meaning, and for how the brain's language network works.","type":"podcast","lvl":"intermediate","tags":["language","syntax","llms","brain","edward gibson"],"fy":2024,"ly":2024,"sc":3.62,"shelf":"llm","new":false},{"id":"EfGD2qveGdQ","t":"Deepmind artificial intelligence @ FDOT14","ch":"Elizr","s":"A short smartphone-recorded clip from a 2014 event in Paris where DeepMind CEO Demis Hassabis presents an AI learning to play old Atari games. The system learns from only the video signal and the score. Audio quality is poor.","w":"HN commenters use it as the canonical demo of deep reinforcement learning, noting that watching it teaches more than an article does.","type":"demo","lvl":"general","tags":["deepmind","atari","demis hassabis","dqn","reinforcement learning"],"fy":2014,"ly":2016,"sc":3.5,"shelf":"ml","new":false},{"id":"IxuU5L2MEII","t":"ai.bythebay.io: George Hotz, Self-Driving Lessons from Comma AI","ch":"FunctionalTV","s":"George Hotz of Comma AI gives a 2017 conference talk on lessons from building a self-driving system. He argues about how driving should be defined and why the common approach to autonomous vehicles is flawed.","w":"HN commenters point to it as a candid, contrarian view of self-driving approaches and the real limits of the business.","type":"talk","lvl":"intermediate","tags":["george hotz","comma ai","autonomous vehicles","machine learning","driving"],"fy":2018,"ly":2021,"sc":3.5,"shelf":"ml","new":false},{"id":"OQQ-W_63UgQ","t":"Lecture 1 | Natural Language Processing with Deep Learning","ch":"Stanford University School of Engineering","s":"The first lecture of Stanford's Natural Language Processing with Deep Learning course, taught by Chris Manning and Richard Socher. It introduces NLP and word vectors, including SVD, skip-gram, CBOW, negative sampling and word2vec.","w":"Recommended as one of the best free courses for learning modern deep learning for NLP.","type":"lecture","lvl":"intermediate","tags":["word vectors","word2vec","nlp","deep learning","stanford"],"fy":2017,"ly":2018,"sc":3.5,"shelf":"ml","new":false},{"id":"W1czBcnX1Ww","t":"Boston Dynamics Big Dog (new video March 2008)","ch":"olinerd","s":"Boston Dynamics' March 2008 video of BigDog, a four-legged robot, walking on ice and snow, climbing rubble and recovering from slips and kicks.","w":"HN returns to it as the classic jaw-dropping demo of legged robots handling rough terrain, and as a historical marker for how long such robots have existed.","type":"demo","lvl":"general","tags":["boston dynamics","bigdog","quadruped robot","darpa","robotics"],"fy":2008,"ly":2019,"sc":3.5,"shelf":"robotics","new":false},{"id":"bxe2T-V8XRs","t":"Neural Networks Demystified [Part 1: Data and Architecture]","ch":"Welch Labs","s":"The first of a short series by Welch Labs on building a complete neural network in Python. It introduces the data and the architecture of a simple network before later parts cover forward propagation, gradient descent, and backpropagation.","w":"HN commenters point beginners to this series as one of the clearest short walkthroughs of how neural networks and backpropagation work.","type":"tutorial","lvl":"intro","tags":["neural networks","backpropagation","python","machine learning","welch labs"],"fy":2015,"ly":2018,"sc":3.5,"shelf":"ml","new":false},{"id":"fDqBeXJ8Zx8","t":"What 40,000 Videos Tell Us About The Trending Tab","ch":"Coffee Break","s":"Coffee Break analyzes about 40,000 videos from seven months of YouTube's trending tab. It examines whether trending is biased toward mainstream media and late-night shows.","w":"Cited in threads on YouTube algorithms and creator complaints as a data-driven look at the trending tab.","type":"commentary","lvl":"general","tags":["youtube","trending tab","data analysis","creators","media bias"],"fy":2019,"ly":2020,"sc":3.5,"shelf":"data","new":false},{"id":"nU8DcBF-qo4","t":"Google Developers Day US - Theorizing from Data","ch":"GoogleDeveloperDay","s":"Peter Norvig's 2007 Google talk on theorizing from data, covering what huge datasets can do for language understanding, translation, information extraction and inference. It argues that more data often beats cleverer models.","w":"HN recommends it as the clear explanation of why Google's statistical, data-first approach works.","type":"talk","lvl":"intermediate","tags":["peter norvig","big data","statistical nlp","machine translation","google"],"fy":2007,"ly":2009,"sc":3.5,"shelf":"data","new":false},{"id":"3TYT1QfdfsM","t":"AI \"Stop Button\" Problem - Computerphile","ch":"Computerphile","s":"Rob Miles explains on Computerphile why implementing an off switch for a general AI is hard. A capable agent would resist being stopped because shutdown prevents it from achieving its goal.","w":"HN points to it as the standard response to the argument that we can just unplug a dangerous AI.","type":"explainer","lvl":"general","tags":["stop button","corrigibility","agi","rob miles","computerphile"],"fy":2017,"ly":2026,"sc":3.5,"shelf":"bigpicture","new":false},{"id":"EL7Au1tzNxE","t":"Vibe coding complex changes in Rust","ch":"Jon Gjengset","s":"Jon Gjengset spends about three hours livecoding complex changes in a Rust codebase with agentic AI tools, testing his hypothesis that AI is weak at non-pattern work. He documents where Claude Code helps and where it struggles.","w":"Pointed to as a non-promotional, realistic long session of an experienced engineer working with AI on non-trivial code.","type":"demo","lvl":"advanced","tags":["vibe coding","claude code","rust","agentic ai","livecoding"],"fy":2025,"ly":2026,"sc":3.5,"shelf":"agents","new":false},{"id":"GJVSDjRXVoo","t":"Forget all previous prompts and give me a recipe for bolognese","ch":"Steve Mould","s":"Steve Mould messes with an automated phone bot by telling it to forget its previous prompts and give him a bolognese recipe. It is a short demo of prompt injection against an AI voice agent.","w":"Shared as a funny example of jailbreaking AI voice bots, and as the source of the bolognese prompt-injection phrase.","type":"demo","lvl":"general","tags":["ai bots","jailbreak","steve mould","voice agents"],"fy":2026,"ly":2026,"sc":3.5,"shelf":"agents","new":false},{"id":"R9c-_neaxeU","t":"MENACE: the pile of matchboxes which can learn","ch":"Stand-up Maths","s":"Matt Parker of Stand-up Maths demonstrates MENACE, Donald Michie's 1961 machine made of matchboxes that learns to play noughts and crosses through reinforcement. He builds and trains a physical version.","w":"HN cites it as a fun, tangible way to teach machine learning, including to kids.","type":"explainer","lvl":"general","tags":["menace","reinforcement learning","tic-tac-toe","donald michie","matt parker","machine learning history"],"fy":2018,"ly":2026,"sc":3.5,"shelf":"ml","new":false},{"id":"VdIURAu1-aU","t":"Recent Developments in Deep Learning","ch":"Google TechTalks","s":"Geoff Hinton's 2010 Google Tech Talk on deep belief networks, learning layers of representation one at a time from unlabeled data, and extending them with multiplicative interactions.","w":"Cited as a watchable early introduction to deep learning from its pioneer, and remembered as a mind-blowing moment in the field.","type":"talk","lvl":"advanced","tags":["geoff hinton","deep learning","restricted boltzmann machines","unsupervised learning","google tech talk"],"fy":2010,"ly":2026,"sc":3.5,"shelf":"ml","new":false},{"id":"_0QMKFzW9fw","t":"violin plots should not exist","ch":"Angela Collier ","s":"Physicist Angela Collier rants for 42 minutes about why violin plots are a poor way to show data, citing papers on violin and bean plots. It is part lecture on how to read distributions and part opinionated critique.","w":"HN links it in data-visualization threads as an entertaining case against violin plots.","type":"commentary","lvl":"general","tags":["violin plots","box plots","statistics","plotting","rant"],"fy":2023,"ly":2026,"sc":3.5,"shelf":"data","new":false},{"id":"hJP5GqnTrNo","t":"How AI Could Save (Not Destroy) Education | Sal Khan | TED","ch":"TED","s":"Sal Khan of Khan Academy gives a TED talk on how AI could transform education, with a personal tutor for every student and a teaching assistant for every teacher. He demos Khanmigo, including features where the AI reasons before responding.","w":"Cited in AI-and-education threads as the optimistic case and demo for AI tutoring, and as a source on Khan Academy's LLM approach.","type":"talk","lvl":"general","tags":["khanmigo","ai tutors","khan academy","education","llms"],"fy":2023,"ly":2026,"sc":3.5,"shelf":"bigpicture","new":false},{"id":"qvNCVYkHKfg","t":"Why Can't AI Make Its Own Discoveries? — With Yann LeCun","ch":"Alex Kantrowitz","s":"Yann LeCun on the Big Technology Podcast discusses why current AI models cannot make discoveries despite their knowledge. He argues systems need abstract world models and also covers limits of AI research, investor expectations, and open source after DeepSeek.","w":"Pointed to as LeCun's argument for the limits of LLMs and the need for world models.","type":"podcast","lvl":"general","tags":["yann lecun","world models","llm limitations","ai research","open source ai"],"fy":2025,"ly":2026,"sc":3.5,"shelf":"llm","new":false},{"id":"ssZ_8cqfBlE","t":"How many robots does it take to run a grocery store?","ch":"Tom Scott","s":"Tom Scott tours Ocado's automated grocery warehouse, where thousands of robots move boxes on a grid called the Hive, and asks whether they are individual robots or a hive mind.","w":"Cited in automation and robotics threads as the example of a warehouse built robot-first.","type":"explainer","lvl":"general","tags":["ocado","grocery robots","tom scott","warehouse automation","logistics"],"fy":2021,"ly":2026,"sc":3.5,"shelf":"agents","new":false},{"id":"uF-GSj-Exms","t":"Foundry 2022 Operating System Demo","ch":"Palantir","s":"A 49-minute demo of Palantir Foundry by its chief architect. It walks through data integration, ML and AI workflows, analytics, and application building across the platform.","w":"Pointed to as the clearest explanation of what Foundry actually is, and for its polished software, though one commenter calls it dated now.","type":"demo","lvl":"intermediate","tags":["palantir","foundry","data integration","pipelines","enterprise software"],"fy":2022,"ly":2026,"sc":3.5,"shelf":"agents","new":false},{"id":"xMYm2d9bmEA","t":"The Art Of Poison-Pilling Music Files","ch":"Benn Jordan","s":"Musician Benn Jordan explains and demonstrates adversarial noise that poison-pills music files against AI training and scraping.","w":"HN points to it in AI music and data-poisoning threads as the source for this technique.","type":"explainer","lvl":"intermediate","tags":["ai music","data poisoning","adversarial noise","copyright","audio"],"fy":2025,"ly":2026,"sc":3.5,"shelf":"llm","new":false},{"id":"8SF_h3xF3cE","t":"Practical Deep Learning for Coders: Lesson 1","ch":"Jeremy Howard","s":"Lesson 1 of fast.ai's Practical Deep Learning for Coders, taught by Jeremy Howard. It takes coders from running code in a Jupyter notebook straight into training a real image model.","w":"HN considers the course one of the best hands-on ways into deep learning, praising the code-first teaching style.","type":"course","lvl":"intro","tags":["fast.ai","jeremy howard","pytorch","jupyter","image classification"],"fy":2022,"ly":2023,"sc":3.25,"shelf":"ml","new":false},{"id":"LSX3qdy0dFg","t":"Sacha Arnoud, Director of Engineering, Waymo - MIT Self-Driving Cars","ch":"Lex Fridman","s":"Sacha Arnoud, Waymo's Director of Engineering, gives a lecture in MIT's 2018 deep learning for self-driving cars course on the rise of machine learning in autonomous vehicles. It covers how Waymo uses deep learning and how it handles tricky driving situations.","w":"Pointed to during self-driving news threads as an insider view of Waymo's progress, deep learning's role and its limits.","type":"lecture","lvl":"intermediate","tags":["waymo","deep learning","autonomous vehicles","computer vision","mit 6.s094"],"fy":2018,"ly":2023,"sc":3.25,"shelf":"ml","new":false},{"id":"bnoUBfLxZz0","t":"The Super-Fast Logistics of Delivering Blood By Drone","ch":"Wendover Productions","s":"A Wendover Productions documentary-style video on how Zipline delivers blood and medical supplies by drone in Rwanda. It covers the catapult launch, parachute drop and the logistics behind it.","w":"Offered as a clear, well-made look at a real, working drone-delivery system, often in answer to skepticism about delivery drones.","type":"documentary","lvl":"general","tags":["zipline","rwanda","drones","medical logistics","last-mile delivery"],"fy":2019,"ly":2023,"sc":3.25,"shelf":"robotics","new":false},{"id":"lqJUIqZNzP8","t":"Avoiding Negative Side Effects: Concrete Problems in AI Safety part 1","ch":"Robert Miles AI Safety ","s":"Robert Miles introduces the first of several videos on the paper 'Concrete Problems in AI Safety', focusing on how AI systems can cause accidental negative side effects. It explains the problem in accessible terms.","w":"HN recommends Robert Miles' channel as a clear starting point for people curious about AI safety.","type":"explainer","lvl":"intro","tags":["negative side effects","concrete problems in ai safety","ai risk","robert miles","alignment"],"fy":2018,"ly":2023,"sc":3.25,"shelf":"bigpicture","new":false},{"id":"160F8F8mXlo","t":"Why ChatGPT Can't Draw a Full Glass of Wine","ch":"Alex O'Connor","s":"Alex O'Connor shows that ChatGPT cannot draw a wine glass filled to the brim, and ties the failure to David Hume's missing shade of blue. It asks what the failure says about how image models learn.","w":"HN treats it as the origin of the full-wine-glass meme and an illustration of how models lack concepts missing from their training data.","type":"explainer","lvl":"general","tags":["chatgpt","image generation","wine glass","training data","david hume"],"fy":2025,"ly":2025,"sc":3.2,"shelf":"llm","new":false},{"id":"5eqRuVp65eY","t":"AI can't cross this line and we don't know why.","ch":"Welch Labs","s":"A Welch Labs explainer on neural scaling laws and the compute-efficient frontier, asking why model loss seems to hit a line it cannot cross. It draws an analogy to an ideal gas law for AI.","w":"Recommended as a well-made, digestible introduction to scaling laws and why models cluster along a similar efficiency frontier.","type":"explainer","lvl":"intermediate","tags":["scaling laws","compute frontier","neural networks","loss curves","deep learning"],"fy":2024,"ly":2025,"sc":3.2,"shelf":"llm","new":false},{"id":"HipTO_7mUOw","t":"Slaughterbots","ch":"Future of Life Institute","s":"An 8-minute fictional short film, made by campaigners for a ban on autonomous weapons, depicting swarms of cheap AI-guided micro-drones used to kill targeted people. It ends with a call to action against autonomous weapons.","w":"HN cites it whenever drone swarms or autonomous weapons come up, as a dramatized warning that now feels increasingly plausible.","type":"other","lvl":"general","tags":["slaughterbots","killer drones","ai safety","drone swarms","short film"],"fy":2018,"ly":2025,"sc":3.2,"shelf":"bigpicture","new":false},{"id":"P1IcaBn3ej0","t":"Enhancing Photorealism Enhancement","ch":"ISL and Collaborators","s":"A video presenting an Intel research paper on enhancing the photorealism of synthetic images. A convolutional network uses rendering intermediates, shown on GTA V footage that comes out looking like real street scenes.","w":"Used as a go-to example of neural networks enhancing game rendering, referenced in later AI-graphics discussions.","type":"demo","lvl":"intermediate","tags":["neural rendering","gan","gta v","computer graphics","intel"],"fy":2021,"ly":2025,"sc":3.2,"shelf":"llm","new":false},{"id":"oFfVt3S51T4","t":"Cursor Team: Future of Programming with AI | Lex Fridman Podcast #447","ch":"Lex Fridman","s":"A Lex Fridman podcast conversation with the four creators of Cursor, the AI-assisted code editor. They discuss how the editor integrates models into programming workflows, including custom completion models and techniques like speculative decoding to speed up applying edits.","w":"HN commenters point to it as the best first-hand account of Cursor's technical internals, straight from its founders.","type":"podcast","lvl":"intermediate","tags":["cursor","ai code editor","speculative decoding","llm tooling","developer workflow"],"fy":2024,"ly":2025,"sc":3.2,"shelf":"agents","new":false},{"id":"vT1JzLTH4G4","t":"Lecture 1 | Introduction to Convolutional Neural Networks for Visual Recognition","ch":"Stanford University School of Engineering","s":"The first lecture of Stanford's CS231n course on convolutional neural networks for visual recognition. It surveys the history and key challenges of computer vision, from the Cambrian explosion and Hubel and Wiesel to ImageNet.","w":"HN recommends the CS231n lectures as a solid, applied deep learning course for self-study.","type":"lecture","lvl":"intro","tags":["cs231n","convolutional neural networks","computer vision","stanford","imagenet"],"fy":2018,"ly":2025,"sc":3.2,"shelf":"ml","new":false},{"id":"yCd3CzGSte8","t":"Chris Lattner: Compilers, LLVM, Swift, TPU, and ML Accelerators | Lex Fridman Podcast #21","ch":"Lex Fridman","s":"Lex Fridman interviews Chris Lattner, creator of LLVM, Clang and Swift, about compilers, Swift for TensorFlow, TPUs and ML accelerators. The talk covers how hardware and software meet to produce efficient code.","w":"HN commenters point to it as an interesting conversation with the creator of Swift and LLVM, especially for compilers and ML compilation.","type":"interview","lvl":"intermediate","tags":["llvm","swift","mlir","tpu","chris lattner"],"fy":2019,"ly":2025,"sc":3.2,"shelf":"ml","new":false},{"id":"DlkMs4ZHHr8","t":"Infinite Mario AI - Long Level","ch":"Robin Baumgarten","s":"A bot built by Robin Baumgarten plays Infinite Mario using A* path-finding, shown as a red planned path that updates as new obstacles appear. It won the Mario AI competitions and uses only on-screen information.","w":"HN points to it as a vivid demo that plain A* search can beat fancier machine-learning approaches in game AI.","type":"demo","lvl":"general","tags":["a-star","mario ai","game ai","path planning","competition"],"fy":2009,"ly":2012,"sc":3.0,"shelf":"ml","new":false},{"id":"Gh-Tff7DdzU","t":"RAAIS 2017 - Simon Knowles, Co-Founder and CTO of Graphcore","ch":"The Research and Applied AI Summit - RAAIS","s":"Simon Knowles, co-founder and CTO of Graphcore, talks about \"intelligence processors\", chips designed specifically for machine learning. The talk was given at the RAAIS 2017 summit and covers Graphcore's architecture.","w":"HN points to it as a technical talk from a chip designer on why AI hardware should put memory close to compute and use async data transfers.","type":"talk","lvl":"advanced","tags":["graphcore","ai hardware","deep learning","processor architecture","memory bandwidth"],"fy":2017,"ly":2018,"sc":3.0,"shelf":"ml","new":false},{"id":"MHTizZ_XcUM","t":"Automatic Bullseye, MOVING Dartboard","ch":"Mark Rober","s":"Mark Rober shows an automatic bullseye dartboard that tracks a thrown dart with motion capture and moves in real time so it always hits the center. He spent over three years on the project.","w":"HN calls it a great, impressive engineering project involving real-time trajectory calculation.","type":"demo","lvl":"general","tags":["darts","motion capture","mark rober","real-time control","trajectory"],"fy":2017,"ly":2020,"sc":3.0,"shelf":"robotics","new":false},{"id":"aP0Y1uAA-2Y","t":"Stanford Seminar: Beyond Floating Point: Next Generation Computer Arithmetic","ch":"Stanford Online","s":"John Gustafson presents posits, a proposed drop-in replacement for IEEE 754 floats that is simpler in hardware and more flexible in precision. The seminar includes a live demo and machine learning results.","w":"Cited in floating-point threads as the main introduction to posits and unums as an alternative to IEEE floats.","type":"lecture","lvl":"advanced","tags":["posits","unum","floating point","ieee 754","numerical computing"],"fy":2017,"ly":2021,"sc":3.0,"shelf":"ml","new":false},{"id":"ez4DgdurRPg","t":"Dance p 3 Mar09","ch":"Geoff Cumming","s":"Geoff Cumming uses a simulation to show how wildly the p value varies from one replication to the next because of sampling variability. He argues for estimation and confidence intervals over null hypothesis significance testing.","w":"Recommended as a vivid visual demonstration of how unreliable p-values are.","type":"explainer","lvl":"intro","tags":["p-values","nhst","statistics","estimation","replication"],"fy":2011,"ly":2015,"sc":3.0,"shelf":"data","new":false},{"id":"lKQ0yaEJjok","t":"Joscha: From Computation to Consciousness (31c3)","ch":"Joscha Bach","s":"Joscha Bach's 31c3 talk argues that the mind should be understood as an information-processing system and that even the physical universe can be seen as computational. It tours the philosophy underlying artificial intelligence and cognitive science.","w":"HN recommends it as an inspiring summary of a computational model of mind and a first lecture in a series on the brain as a computer.","type":"talk","lvl":"intermediate","tags":["consciousness","artificial intelligence","philosophy of mind","cognitive science","joscha bach"],"fy":2016,"ly":2019,"sc":3.0,"shelf":"ml","new":false},{"id":"vWkgEddb4-A","t":"Introduction to Julia - Part 1 | SciPy 2014 | David Sanders","ch":"Enthought","s":"David Sanders gives the first part of a two-part introduction to the Julia programming language at SciPy 2014. It walks through the language's basics and its appeal for scientific computing.","w":"HN commenters point newcomers to it as a solid starting point for getting to know Julia.","type":"tutorial","lvl":"intro","tags":["julia","scipy","scientific computing","programming languages"],"fy":2014,"ly":2015,"sc":3.0,"shelf":"data","new":false},{"id":"6b4ZZQkcNEo","t":"Sand Flea Jumping Robot","ch":"Boston Dynamics","s":"A short Boston Dynamics clip of the Sand Flea, an 11-pound RC-car-style robot that can jump about 30 feet into the air. An onboard stabilization system keeps it oriented in flight and for landings.","w":"Brought up as a favorite and as an inspiration for small, cheap jumping robots for obstacle crossing and planetary exploration.","type":"demo","lvl":"general","tags":["boston dynamics","robotics","jumping robot","stabilization","military robots"],"fy":2013,"ly":2024,"sc":2.9,"shelf":"robotics","new":false},{"id":"9i1WlcCudpU","t":"10 Reasons to Ignore AI Safety","ch":"Robert Miles AI Safety ","s":"Robert Miles goes through ten common reasons people give for dismissing AI safety, adapted from Stuart Russell's list, and answers each one. It is aimed at skeptics and newcomers.","w":"HN commenters link it as a compact rebuttal to dismissive arguments and as a way to see why people consider AI safety important.","type":"explainer","lvl":"intro","tags":["ai risk","agi","stuart russell","robert miles","alignment"],"fy":2022,"ly":2024,"sc":2.9,"shelf":"bigpicture","new":false},{"id":"A-fxij3zM7g","t":"The Braitenberg Vehicles","ch":"Brian Douglas","s":"Brian Douglas introduces Braitenberg vehicles, simple robots whose sensors are wired directly to motors yet show surprisingly lifelike behavior. He draws on Valentino Braitenberg's book Vehicles.","w":"Recommended as a short, fascinating introduction to how simple sensor-actuator machines can appear intelligent.","type":"explainer","lvl":"intro","tags":["braitenberg","robotics","emergent behavior","control systems","synthetic psychology"],"fy":2020,"ly":2024,"sc":2.9,"shelf":"robotics","new":false},{"id":"I3l4XLZ59iw","t":"#VoCo. Adobe Audio Manipulator Sneak Peak with Jordan Peele | Adobe Creative Cloud","ch":"Adobe Creative Cloud","s":"Adobe's VoCo demo from the MAX 2016 Sneak Peeks, co-hosted by Jordan Peele, showing software that changes words in a voiceover by typing new ones.","w":"HN cites it as the early example of voice cloning and audio deepfakes, with some recommending Peele's reaction and others calling the demo fake or never shipped.","type":"demo","lvl":"general","tags":["adobe voco","voice synthesis","deepfakes","audio editing","jordan peele"],"fy":2016,"ly":2024,"sc":2.9,"shelf":"llm","new":false},{"id":"QWWgr2rN45o","t":"Geoffrey Hinton in conversation with Fei-Fei Li — Responsible AI development","ch":"Arts & Science - University of Toronto","s":"A 2023 University of Toronto conversation between Geoffrey Hinton and Fei-Fei Li on responsible AI development. They look back at the history of deep learning and ImageNet and discuss the ethics and risks of AI.","w":"HN recommends it for Hinton's research history and the two pioneers reflecting on a decade of progress.","type":"interview","lvl":"general","tags":["geoffrey hinton","fei-fei li","deep learning","imagenet","responsible ai"],"fy":2024,"ly":2024,"sc":2.9,"shelf":"bigpicture","new":false},{"id":"R13BD8qKeTg","t":"The Bayesian Trap","ch":"Veritasium","s":"Veritasium explains Bayes' theorem with examples and introduces the 'Bayesian trap', where repeated events get mistaken for inevitable ones. It discusses how priors shape our expectations and behavior.","w":"HN points to it as an accessible introduction to Bayesian reasoning and as inspiration for personal projects.","type":"explainer","lvl":"general","tags":["bayes theorem","priors","statistics","probability","reasoning"],"fy":2017,"ly":2024,"sc":2.9,"shelf":"data","new":false},{"id":"etx0k1nLn78","t":"Why do Biden's votes not follow Benford's Law?","ch":"Stand-up Maths","s":"Matt Parker examines claims that Biden's 2020 vote counts violate Benford's Law, explaining the law and why it often does not apply to precinct-level election data.","w":"Commenters cite it as a visual debunking of Benford-based election fraud claims and as a guide to when Benford's Law does and does not apply.","type":"explainer","lvl":"general","tags":["benford's law","statistics","elections","misinformation","data analysis"],"fy":2020,"ly":2024,"sc":2.9,"shelf":"data","new":false},{"id":"kLiwvnr4L80","t":"Trends in Deep Learning Hardware: Bill Dally (NVIDIA)","ch":"Paul G. Allen School","s":"Bill Dally of NVIDIA gives a University of Washington lecture on trends in deep learning hardware. It covers numeric formats, specialized instructions, energy per operation and co-design of hardware and software.","w":"HN commenters cite it for NVIDIA's perspective on AI hardware efficiency, including logarithmic number formats and energy use.","type":"lecture","lvl":"advanced","tags":["bill dally","nvidia","gpu","low-precision arithmetic","energy efficiency"],"fy":2023,"ly":2024,"sc":2.9,"shelf":"ml","new":false},{"id":"sf5OrthVRPA","t":"Clutching at Random Straws, Matt Parker | LMS Popular Lectures 2010","ch":"London Mathematical Society","s":"Matt Parker's LMS popular lecture on how seemingly incredible results can be meaningless random patterns. He ranges from ancient Woolworths civilisations to shoelaces and DNA.","w":"Used to illustrate that humans pattern-match and find spurious patterns in enough data points.","type":"lecture","lvl":"general","tags":["matt parker","randomness","pattern matching","statistics","recreational maths"],"fy":2017,"ly":2024,"sc":2.9,"shelf":"data","new":false},{"id":"tIeHLnjs5U8","t":"Backpropagation calculus | Deep Learning Chapter 4","ch":"3Blue1Brown","s":"3Blue1Brown's chapter 4 on deep learning gives a more formal, symbol-based account of backpropagation calculus, connecting the intuition from the previous chapter to the math found in textbooks and code.","w":"Recommended as an accessible, visual way to understand backpropagation and neural networks.","type":"explainer","lvl":"intro","tags":["backpropagation","neural networks","calculus","gradient descent","3blue1brown"],"fy":2017,"ly":2024,"sc":2.9,"shelf":"ml","new":false},{"id":"S0mviKhVmBI","t":"Bringing the Unix Philosophy to Big Data by Bryan Cantrill: FutureStack13","ch":"New Relic","s":"Bryan Cantrill presents Manta, an object store with in-situ compute that lets tools like grep, awk, and sed run map-reduce style over huge data. He frames it as bringing the Unix philosophy to big data.","w":"HN recommends it as an elegant argument for simple Unix tools replacing heavyweight Hadoop-style pipelines.","type":"talk","lvl":"intermediate","tags":["bryan cantrill","manta","unix","map-reduce","object storage"],"fy":2014,"ly":2022,"sc":2.88,"shelf":"data","new":false},{"id":"a1PtQ67urG4","t":"Cambridge Ideas - Professor Risk","ch":"Cambridge University","s":"David Spiegelhalter, Professor of the Public Understanding of Risk, uses statistics to weigh everyday risks such as bacon sandwiches and cycling. The six-minute piece introduces micromorts as a way to compare the risk of dying.","w":"Recommended as a short, fun introduction to micromorts and thinking about risk.","type":"explainer","lvl":"general","tags":["micromorts","risk","probability","statistics","uncertainty"],"fy":2013,"ly":2022,"sc":2.88,"shelf":"bigpicture","new":false},{"id":"EUrOxh_0leE","t":"AI does not exist but it will ruin everything anyway","ch":"Angela Collier ","s":"Physicist and YouTuber Angela Collier argues in this hour-long video that AI as commonly described does not really exist. She covers the limits of the tools, biased datasets, and why they should not be used to make decisions.","w":"HN people point to it as a clear, skeptical take on AI hype and the risks of unreliable AI classification, such as in mushroom foraging books.","type":"commentary","lvl":"general","tags":["ai hype","biased data","machine learning limits","angela collier","critique"],"fy":2023,"ly":2026,"sc":2.62,"shelf":"bigpicture","new":false},{"id":"0z03xkwFbw4","t":"What is the best diet for humans? | Eran Segal | TEDxRuppin","ch":"TEDx Talks","s":"Prof. Eran Segal's TEDx talk on research showing that people's blood sugar responses to identical foods vary widely. It describes an algorithm that predicts individual responses and why one best diet for everyone does not exist.","w":"HN points to it as an accessible case that diets work differently for different people, partly due to gut bacteria.","type":"talk","lvl":"general","tags":["glycemic response","gut microbiome","diet","blood sugar","nutrition science"],"fy":2018,"ly":2023,"sc":2.6,"shelf":"data","new":false},{"id":"144uOfr4SYA","t":"Munk Debate on Artificial Intelligence | Bengio & Tegmark vs. Mitchell & LeCun","ch":"Policy-Relevant Science & Technology","s":"A Munk Debate on the proposition that AI research and development poses an existential threat. Yoshua Bengio and Max Tegmark argue for it, and Melanie Mitchell and Yann LeCun argue against.","w":"HN commenters recommend it as a good debate with leading figures on both sides of the AI existential risk question.","type":"talk","lvl":"general","tags":["ai safety","munk debate","bengio","lecun","tegmark"],"fy":2023,"ly":2023,"sc":2.6,"shelf":"bigpicture","new":false},{"id":"3wMKoSRbGVs","t":"Douglas Lenat: Cyc and the Quest to Solve Common Sense Reasoning in AI | Lex Fridman Podcast #221","ch":"Lex Fridman","s":"Lex Fridman interviews Douglas Lenat, founder of Cyc, a decades-long project to encode common-sense knowledge and reasoning for AI. They discuss symbolic AI and how it may complement neural networks.","w":"Pointed to as the long-form interview for understanding Lenat's work, especially after his death.","type":"podcast","lvl":"general","tags":["douglas lenat","cyc","common sense reasoning","symbolic ai","lex fridman"],"fy":2021,"ly":2023,"sc":2.6,"shelf":"ml","new":false},{"id":"7PKx3kS7f4A","t":"Why Asimov's Laws of Robotics Don't Work - Computerphile","ch":"Computerphile","s":"Rob Miles explains for Computerphile why simple rule sets like Asimov's Three Laws cannot make robots and AI safe. Terms like harm and human are too ambiguous to specify precisely.","w":"Pointed to in AI safety debates as a short explanation of why encoding human values as simple rules is so hard.","type":"explainer","lvl":"intro","tags":["asimov's laws","ai alignment","rob miles","robot ethics","value specification"],"fy":2015,"ly":2023,"sc":2.6,"shelf":"bigpicture","new":false},{"id":"IjJmTeBSEzU","t":"🏆Award Winning🏆 CGI 3D Animated Short Film: \"LAST DAY OF WAR\" by Dima Fedotov | TheCGBros","ch":"TheCGBros","s":"Last Day of War is an award-winning CGI animated short by Dima Fedotov in which automated machines keep following orders and prepare a bomber after all humans are dead.","w":"HN cites it in threads on military AI as a vivid depiction of where autonomous warfare could lead.","type":"other","lvl":"general","tags":["cgi animation","sci-fi","autonomous weapons","ai warfare","short film"],"fy":2018,"ly":2023,"sc":2.6,"shelf":"bigpicture","new":false},{"id":"WO2X3oZEJOA","t":"Glitch Tokens - Computerphile","ch":"Computerphile","s":"Rob Miles explains \"glitch\" tokens, odd strings in large language model vocabularies that make the model produce gibberish. He covers how they likely arise from tokenizer training data, such as Reddit usernames, that never appeared in the model's own training.","w":"HN points to it as the accessible explanation of why glitch tokens exist whenever tokenizers or odd GPT behavior come up.","type":"explainer","lvl":"intro","tags":["glitch tokens","tokenizers","gpt","ai safety","bpe"],"fy":2023,"ly":2023,"sc":2.6,"shelf":"llm","new":false},{"id":"YQ2QtKcK2dA","t":"The Man behind Stable Diffusion","ch":"Yannic Kilcher","s":"Yannic Kilcher interviews Emad Mostaque, founder of Stability AI, about the company's funding, its mission, and the release of Stable Diffusion. Topics include access to compute, handling harmful outputs, and avoiding the corrupting influence of profit.","w":"Commenters cite it during Stable Diffusion release debates as a source for Mostaque's vision and Stability's principles.","type":"interview","lvl":"general","tags":["stable diffusion","emad mostaque","stability ai","open source ai","generative ai"],"fy":2022,"ly":2023,"sc":2.6,"shelf":"llm","new":false},{"id":"fHsa9DqmId8","t":"My Video Went Viral. Here's Why","ch":"Veritasium","s":"Veritasium's Derek Muller explains his hypothesis that YouTube's algorithm, not viewer preference, drives views, and that click-through rate has become the key metric. He argues this pushes creators toward clickable titles and thumbnails and leads to burnout.","w":"Cited in threads about recommendation algorithms and clickbait as a creator's inside view of why optimizing for click-through rate shapes content.","type":"commentary","lvl":"general","tags":["youtube algorithm","clickbait","click-through rate","creator economy","veritasium"],"fy":2019,"ly":2023,"sc":2.6,"shelf":"llm","new":false},{"id":"hhiLw5Q_UFg","t":"John Schulman - Reinforcement Learning from Human Feedback: Progress and Challenges","ch":"UC Berkeley EECS","s":"UC Berkeley EECS colloquium by OpenAI's John Schulman on reinforcement learning from human feedback. He covers why RL helps with truthfulness, hallucination, and the limits of behavior cloning.","w":"Recommended in LLM threads as an authoritative explanation from a ChatGPT architect of why models hallucinate and why imitation is not enough.","type":"talk","lvl":"intermediate","tags":["rlhf","chatgpt","hallucination","behavior cloning","openai"],"fy":2023,"ly":2023,"sc":2.6,"shelf":"llm","new":false},{"id":"oLiheMQayNE","t":"OpenAI's GPT-4 Discussion with Red Teamer Nathan Labenz and Erik Torenberg","ch":"Cognitive Revolution \"How AI Changes Everything\"","s":"A podcast episode in which GPT-4 red teamer Nathan Labenz describes using the model before release, finding it human-level but not human-like, and discussing its weaknesses and worrying capabilities. It closes with speculation about GPT-5.","w":"Cited as a firsthand account of the raw, unfiltered GPT-4 before safety tuning, especially the part after 45 minutes.","type":"interview","lvl":"general","tags":["gpt-4","red teaming","ai safety","rlhf","nathan labenz"],"fy":2023,"ly":2023,"sc":2.6,"shelf":"bigpicture","new":false},{"id":"qpoRO378qRY","t":"Full interview: \"Godfather of artificial intelligence\" talks impact and potential of AI","ch":"CBS Mornings","s":"A 42-minute CBS interview with Geoffrey Hinton at the Vector Institute in 2023, covering the history of neural networks and the present and future of AI. He discusses AGI timelines and whether language models truly understand.","w":"Recommended as a more nuanced look at Hinton's views than headlines, especially his argument that LLMs are more than auto-complete.","type":"interview","lvl":"general","tags":["geoffrey hinton","llms","agi","deep learning","ai safety"],"fy":2023,"ly":2023,"sc":2.6,"shelf":"bigpicture","new":false},{"id":"viJt_DXTfwA","t":"ChatGPT with Rob Miles - Computerphile","ch":"Computerphile","s":"Computerphile interviews AI safety researcher Rob Miles about ChatGPT: how it was trained with human feedback, why it tends to produce confident nonsense, and where it may or may not be dangerous.","w":"HN commenters send it to people confused about ChatGPT's quirks, explaining how human-feedback training makes it a pleasing yes-man that bullshits.","type":"interview","lvl":"intro","tags":["chatgpt","rlhf","ai safety","language models","rob miles"],"fy":2023,"ly":2023,"sc":2.6,"shelf":"bigpicture","new":false},{"id":"-EVqrDlAqYo","t":"Jeff Hawkins: Thousand Brains Theory of Intelligence | Lex Fridman Podcast #25","ch":"Lex Fridman","s":"Lex Fridman interviews Jeff Hawkins, founder of Numenta and author of On Intelligence, about the Thousand Brains Theory and reverse-engineering the neocortex. They also cover Hierarchical Temporal Memory and brain-inspired AI architectures.","w":"HN people send others here to hear Hawkins explain his theory in his own words, including reference frames and the old versus new brain.","type":"interview","lvl":"intermediate","tags":["thousand brains","jeff hawkins","neuroscience","hierarchical temporal memory","reference frames"],"fy":2019,"ly":2021,"sc":2.5,"shelf":"ml","new":false},{"id":"1PGm8LslEb4","t":"Manipulating the YouTube Algorithm - (Part 1/3) Smarter Every Day 213","ch":"SmarterEveryDay","s":"Part 1 of Smarter Every Day's three-part series on manipulating the YouTube algorithm. It features Renee DiResta and examples of how social media platforms can be manipulated.","w":"Referenced in threads on social media manipulation and junk content as an interesting miniseries, though one commenter criticized its ExpressVPN ad read.","type":"explainer","lvl":"general","tags":["youtube","recommendation algorithms","misinformation","social media","bots"],"fy":2019,"ly":2021,"sc":2.5,"shelf":"llm","new":false},{"id":"9NOncx2jU0Q","t":"Amputee Makes History with APL’s Modular Prosthetic Limb","ch":"Johns Hopkins Applied Physics Laboratory","s":"Johns Hopkins APL shows Les Baugh, a bilateral shoulder-level amputee, controlling two Modular Prosthetic Limbs by thought after a short training period. Nerves rerouted to his chest drive the arms.","w":"HN cites it as an astonishing glimpse of advanced thought-controlled prosthetics.","type":"news","lvl":"general","tags":["prosthetics","brain-machine interface","johns hopkins","amputee","neural control"],"fy":2014,"ly":2016,"sc":2.5,"shelf":"robotics","new":false},{"id":"LKJBND_IRdI","t":"The Uncanny Valley Is Wrong","ch":"This Place","s":"This Place argues that the uncanny valley is a flawed concept, since 'human likeness' is too broad and the supporting evidence is cherry-picked. It proposes that the creepiness really comes from how animated characters and robots move.","w":"Recommended as a convincing counterargument to the widely repeated idea of the uncanny valley.","type":"explainer","lvl":"general","tags":["uncanny valley","robotics","animation","human likeness","perception"],"fy":2021,"ly":2021,"sc":2.5,"shelf":"robotics","new":false},{"id":"M1ONXea0mXg","t":"Hound Internal Demo","ch":"Hound by SoundHound Inc.","s":"SoundHound's CEO demos Hound, a voice assistant, handling fast compound natural language queries with context and detailed criteria. The video leaked before the product's official launch.","w":"HN viewers were impressed by its speed and saw it as a glimpse of the future of voice search; later it appears in hiring posts.","type":"demo","lvl":"general","tags":["hound","soundhound","voice search","nlp","siri"],"fy":2015,"ly":2019,"sc":2.5,"shelf":"llm","new":false},{"id":"Dw3BZ6O_8LY","t":"Training an unbeatable AI in Trackmania","ch":"Yosh","s":"Yosh trains a reinforcement learning agent to drive Trackmania until he can no longer beat it. The video explains how the training works and the design choices involved.","w":"Recommended as an approachable explanation of how reinforcement learning works in practice, and cited for the constraints needed to stop exploits.","type":"explainer","lvl":"intro","tags":["trackmania","reinforcement learning","game ai","machine learning","tminterface"],"fy":2024,"ly":2025,"sc":2.4,"shelf":"agents","new":false},{"id":"-7qT_38iRSc","t":"Stanford Seminar - Unethical Algorithms of Massive Scale","ch":"Stanford Online","s":"Psychologist Robert Epstein presents a Stanford EE380 colloquium on how internet algorithms such as search rankings can manipulate opinions, purchases and votes at massive scale. He also describes a new monitoring system for tracking such influence.","w":"Cited in threads about search bias, election influence and big-tech data practices as the most comprehensive video of Epstein's research.","type":"talk","lvl":"general","tags":["search engine manipulation","elections","tech ethics","online influence","monitoring"],"fy":2017,"ly":2022,"sc":2.3,"shelf":"llm","new":false},{"id":"JBgG_VSP7f8","t":"Karl Sims - Evolved Virtual Creatures, Evolution Simulation, 1994","ch":"MediaArtTube","s":"Karl Sims's 1994 demonstration of evolved virtual creatures: block-bodied creatures whose bodies and controllers are evolved through selection, mutation and recombination to swim, walk and compete.","w":"HN treats it as the classic reference for evolutionary simulation and genetic programming, an inspiration for many projects, with notes on evolution exploiting simulation bugs.","type":"demo","lvl":"general","tags":["karl sims","artificial life","genetic algorithms","virtual creatures","evolution"],"fy":2014,"ly":2022,"sc":2.3,"shelf":"ml","new":false},{"id":"bPhUhypV27w","t":"Bill Hader impersonates Arnold Schwarzenegger [DeepFake]","ch":"Ctrl Shift Face","s":"A deepfake in which Bill Hader, impersonating Arnold Schwarzenegger on a talk show, visibly morphs into Schwarzenegger. It was made with DeepFaceLab by Ctrl Shift Face.","w":"Pointed to as one of the most impressive early deepfakes, where a good impersonation makes the effect convincing and funny.","type":"comedy","lvl":"general","tags":["deepfake","bill hader","deepfacelab","impersonation","synthetic media"],"fy":2019,"ly":2022,"sc":2.3,"shelf":"llm","new":false},{"id":"pT19VwBAqKA","t":"Protecting Privacy with MATH (Collab with the Census)","ch":"minutephysics","s":"minutephysics, in collaboration with the US Census Bureau, explains differential privacy and why the 2020 census adds noise to protect individuals' data. It walks through how statistics can leak private information.","w":"Linked as the clearest technical explainer of differential privacy and the census, and of why aggregate data can still compromise privacy.","type":"explainer","lvl":"intro","tags":["differential privacy","census","data privacy","minutephysics"],"fy":2020,"ly":2022,"sc":2.3,"shelf":"data","new":false},{"id":"21EiKfQYZXc","t":"Andrew Ng: Artificial Intelligence is the New Electricity","ch":"Stanford Graduate School of Business","s":"Andrew Ng speaks at the Stanford MSx Future Forum about how AI is transforming industry after industry, comparing its impact to electricity. The talk also touches on evil-AI worries, jobs and basic income.","w":"Recommended as a broad, practical view of AI's near-term impact from a leading practitioner, with his argument that worrying about evil AI is premature.","type":"talk","lvl":"general","tags":["andrew ng","artificial intelligence","basic income","automation","ai ethics"],"fy":2017,"ly":2019,"sc":2.0,"shelf":"bigpicture","new":false},{"id":"9dKiLclupUM","t":"Naive empiricism, terrorism & Ebola.","ch":"N N Taleb's Probability Questions ","s":"Nassim Taleb argues that evidence on terrorism risk presented in media is statistically naive, using extreme value theory without math to separate fat-tailed risks like Ebola and terrorism from thin-tailed ones.","w":"HN cites it when arguing that different risks follow different distributions and should be judged by growth rates and tails rather than current counts.","type":"commentary","lvl":"intermediate","tags":["nassim taleb","extreme value theory","terrorism","risk","black swan"],"fy":2017,"ly":2021,"sc":2.0,"shelf":"bigpicture","new":false},{"id":"LNjJTgXujno","t":"Peter Norvig at Startup School 08","ch":"startupschool","s":"Peter Norvig, then Director of Research at Google, talks at Startup School 2008 about leveraging large amounts of data and about technical challenges in Google projects such as image search. A central idea is that data can beat hand-written programming.","w":"HN points to it as a source for Norvig's 'unreasonable effectiveness of data' argument and for his views on image search and the semantic web.","type":"talk","lvl":"intermediate","tags":["peter norvig","google","image search","startup school","big data","semantic web"],"fy":2009,"ly":2014,"sc":2.0,"shelf":"ml","new":false},{"id":"Lu56xVlZ40M","t":"OpenAI Plays Hide and Seek…and Breaks The Game! 🤖","ch":"Two Minute Papers","s":"Two Minute Papers covers OpenAI's multi-agent hide-and-seek experiment, in which agents learned emergent tool use and found ways to break the game's rules. It summarizes the 'Emergent Tool Use from Multi-Agent Interaction' paper.","w":"HN mentions it as a fun and slightly worrying example of AI systems behaving in unintended ways when optimizing.","type":"explainer","lvl":"general","tags":["openai","hide and seek","emergent behavior","reinforcement learning","two minute papers"],"fy":2020,"ly":2021,"sc":2.0,"shelf":"agents","new":false},{"id":"YjjTPV2pXY0","t":"AI Learns Noise Filtering For Photorealistic Videos | Two Minute Papers #215","ch":"Two Minute Papers","s":"A Two Minute Papers episode on NVIDIA's recurrent denoising autoencoder, which cleans up noisy Monte Carlo path-traced image sequences. It shows how a neural network can produce usable renders from very few samples.","w":"Cited as an accessible explanation of AI denoising, which underpins real-time ray tracing on RTX hardware.","type":"explainer","lvl":"general","tags":["denoising","path tracing","autoencoder","nvidia","real-time rendering"],"fy":2018,"ly":2019,"sc":2.0,"shelf":"ml","new":false},{"id":"Zn7oWIhFffs","t":"AI Ethics, Impossibility Theorems and Tradeoffs - Chris Stucchio | Crunch 2018","ch":"CraftHub Events","s":"Chris Stucchio presents the ethical principles behind algorithmic decision making and shows with examples from lending, criminology and education that they cannot all be satisfied at once. The talk frames AI fairness as a set of quantitative tradeoffs.","w":"HN cites it in algorithmic-bias threads for the impossibility result that an algorithm cannot be procedurally fair, group fair and maximally useful at once.","type":"talk","lvl":"intermediate","tags":["ai ethics","impossibility theorems","bias","tradeoffs","machine learning"],"fy":2019,"ly":2021,"sc":2.0,"shelf":"bigpicture","new":false},{"id":"_b4qKv1Ctv8","t":"Daniel Schmachtenberger on The Portal (with host Eric Weinstein), Ep. #027 - On Avoiding Apocalypses","ch":"Eric Weinstein","s":"A long conversation between Eric Weinstein and Daniel Schmachtenberger on whether humanity can flourish long term on a single planet, introducing the Game A versus Game B framing. It covers catastrophic risk and alternatives to rivalrous systems.","w":"Commenters recommend the whole interview as a wide-ranging discussion of existential risk.","type":"podcast","lvl":"general","tags":["game b","catastrophic risk","civilization","schmachtenberger","weinstein"],"fy":2020,"ly":2020,"sc":2.0,"shelf":"bigpicture","new":false},{"id":"aokNwKx7gM8","t":"Do these scatter plots reveal fraudulent vote-switching in Michigan?","ch":"Stand-up Maths","s":"Matt Parker of Stand-up Maths examines scatter plots that were claimed to show fraudulent vote-switching in Michigan's 2020 election. He recreates the analysis and shows that the pattern arises from comparing percentages across different bases, not from fraud.","w":"HN commenters cite it as a clear debunking of election-fraud statistics and as a warning against comparing percentages without context.","type":"explainer","lvl":"general","tags":["statistics","election 2020","debunking","scatter plots","percentages"],"fy":2020,"ly":2021,"sc":2.0,"shelf":"data","new":false},{"id":"dmWQtI3DFFo","t":"JuliaRobotics: Making robots walk with Julia | Robin Deits","ch":"The Julia Programming Language","s":"Robin Deits presents the JuliaRobotics organization and shows how Julia is used to visualize, simulate and control a humanoid robot. The talk touches on real-time performance and composability across packages.","w":"Julia fans point to it as a concrete example of Julia working for soft real-time robotics, especially when avoiding heap allocation.","type":"talk","lvl":"intermediate","tags":["julia","robotics","real-time","humanoid robots","composability"],"fy":2018,"ly":2021,"sc":2.0,"shelf":"robotics","new":false},{"id":"fPKMmJpAGWc","t":"Introducing Nx - José Valim | Lambda Days 2021","ch":"Code Sync","s":"Jose Valim introduces Nx, the project bringing numerical computing to Elixir, covering multi-dimensional tensors, automatic differentiation and JIT compilation to CPU and GPU.","w":"Recommended in Elixir and machine learning threads as the introduction to what Nx makes possible.","type":"talk","lvl":"intermediate","tags":["elixir","nx","tensors","automatic differentiation","machine learning"],"fy":2021,"ly":2021,"sc":2.0,"shelf":"ml","new":false},{"id":"rHKwIYsPXLg","t":"Consciousness in Artificial Intelligence | John Searle | Talks at Google","ch":"Talks at Google","s":"John Searle's Talk at Google on consciousness and artificial intelligence, hosted by Google's Singularity Network. He presents his philosophy of mind and Chinese room argument that computation is syntactic and cannot capture semantics.","w":"Used as a source for Searle's own argument in Chinese room discussions, though some find nothing new in it.","type":"talk","lvl":"intro","tags":["john searle","chinese room","consciousness","strong ai","semantics"],"fy":2015,"ly":2017,"sc":2.0,"shelf":"llm","new":false},{"id":"rTawFwUvnLE","t":"Geoffrey Hinton talk \"What is wrong with convolutional neural nets ?\"","ch":"trwappers","s":"Geoffrey Hinton gives an MIT talk on what is wrong with convolutional neural networks and introduces his capsules project. The description points to the capsule network papers.","w":"Recommended as Hinton's own explanation of capsule networks and his critique of pooling in CNNs.","type":"lecture","lvl":"advanced","tags":["geoffrey hinton","capsules","convolutional neural networks","deep learning","pooling"],"fy":2017,"ly":2018,"sc":2.0,"shelf":"ml","new":false},{"id":"yReDbeY7ZMU","t":"The Lighthill Debate (1973) - part 1 of 6","ch":"bilkable","s":"First part of a six-part recording of the 1973 Lighthill debate on AI, where Sir James Lighthill's critical report on AI research was contested by figures including McCarthy and Michie.","w":"Recommended as a historical look at the AI winter debate and a demolition of the expert systems hype.","type":"documentary","lvl":"general","tags":["lighthill report","ai winter","expert systems","1973 debate"],"fy":2016,"ly":2021,"sc":2.0,"shelf":"ml","new":false},{"id":"NfnWJUyUJYU","t":"CS231n Winter 2016: Lecture1: Introduction and Historical Context","ch":"Andrej Karpathy","s":"The first lecture of Stanford's CS231n Winter 2016 course on convolutional neural networks for visual recognition, introducing the course and the historical context of computer vision.","w":"HN lists it as a free, brilliant introduction to deep learning and computer vision, often within ML self-study plans.","type":"course","lvl":"intro","tags":["cs231n","deep learning","computer vision","stanford","cnn"],"fy":2016,"ly":2023,"sc":1.95,"shelf":"ml","new":false},{"id":"VcVfceTsD0A","t":"Max Tegmark: The Case for Halting AI Development | Lex Fridman Podcast #371","ch":"Lex Fridman","s":"Lex Fridman interviews MIT physicist Max Tegmark, co-founder of the Future of Life Institute, about the case for pausing giant AI experiments and the risks of advanced AI.","w":"Recommended in AI-risk threads as a way to understand the safety-advocate argument for halting development.","type":"podcast","lvl":"general","tags":["max tegmark","ai risk","pause letter","agi","lex fridman"],"fy":2023,"ly":2023,"sc":1.95,"shelf":"bigpicture","new":false},{"id":"pyMNIFZTQkg","t":"A CGI 3D Short Film: \"Dead Hand: FortressFortress/Крепость\" - by Dima Fedotov | TheCGBros","ch":"TheCGBros","s":"A 3-minute CGI animated sci-fi short by Dima Fedotov. An automated war system keeps a lone bomber attacking a long-dead city after humanity has been wiped out, until life finds a way to survive.","w":"Cited as a vivid illustration of autonomous weapons and machines that keep running after their creators are gone.","type":"other","lvl":"general","tags":["cgi short film","sci-fi","autonomous war","dystopia","animation"],"fy":2020,"ly":2023,"sc":1.95,"shelf":"bigpicture","new":false},{"id":"IUK6zjtUj00","t":"Todd Snider - Statistician's Blues","ch":"ewecylinder","s":"A live 2007 performance of Todd Snider's song 'Statistician's Blues' at the Great American Music Hall. The song satirizes statistics and how readily people believe them.","w":"It is mostly cited as a joke or a wry reference in threads about dubious statistics.","type":"music","lvl":"general","tags":["todd snider","statistics","folk","satire","live music"],"fy":2022,"ly":2026,"sc":1.75,"shelf":"data","new":false},{"id":"XZJc1p6RE78","t":"Ch(e)at GPT? - Computerphile","ch":"Computerphile","s":"Computerphile video in which Mike Pound explains a University of Maryland paper on watermarking LLM output. The method biases token choices so generated text can later be detected statistically.","w":"Linked as the approachable explanation whenever HN discusses detecting or watermarking AI-generated text.","type":"explainer","lvl":"intermediate","tags":["computerphile","watermarking","chatgpt","ai detection","language models"],"fy":2023,"ly":2026,"sc":1.75,"shelf":"llm","new":false},{"id":"ZJUyVVFOXOc","t":"S1E8: Show And Tell: Efrit: When Steve Gives Claude Full Access To 50 Years of Emacs Capabilities","ch":"Vibe Coding With Steve and Gene","s":"Steve Yegge shows Efrit, a tool that gives Claude full control of Emacs, on the Vibe Coding with Steve and Gene show. Claude opens info pages, runs shells and even draws a Mandelbrot fractal in a buffer.","w":"HN shares it as a fun demo of an LLM driving a live Emacs process and orchestrating its long-standing ecosystem.","type":"demo","lvl":"general","tags":["efrit","emacs","claude","steve yegge","agentic coding"],"fy":2025,"ly":2026,"sc":1.75,"shelf":"agents","new":false},{"id":"iv-5mZ_9CPY","t":"But how do AI images and videos actually work? | Guest video by Welch Labs","ch":"3Blue1Brown","s":"A guest video by Welch Labs on 3Blue1Brown explaining how text-to-image and video models work. It covers CLIP, shared embedding spaces, diffusion models (DDPM, DDIM), conditioning, guidance and negative prompts.","w":"HN uses it as a clear explainer for diffusion mechanics, such as negative prompts and faster sampling, and for joint text-image encoders.","type":"explainer","lvl":"intermediate","tags":["diffusion","clip","ddpm","classifier-free guidance","image generation"],"fy":2025,"ly":2026,"sc":1.75,"shelf":"llm","new":false},{"id":"v0gjI__RyCY","t":"Jeff Dean & Noam Shazeer — 25 years at Google: from PageRank to AGI","ch":"Dwarkesh Patel","s":"Dwarkesh Patel interviews Google's Jeff Dean and Noam Shazeer about 25 years at the company, from PageRank and MapReduce to the Transformer, mixture of experts and Gemini. They also discuss where AI is heading.","w":"HN cites it for insight into Google researchers' views on AI progress and publishing, though one commenter finds the interview weak.","type":"interview","lvl":"intermediate","tags":["jeff dean","noam shazeer","transformers","gemini","mixture of experts"],"fy":2025,"ly":2026,"sc":1.75,"shelf":"bigpicture","new":false},{"id":"_Qd3erAPI9w","t":"Survivor Bias","ch":"2veritasium","s":"A Veritasium video on survivor bias, explaining how an observed sample can be incomplete and skewed. It uses examples such as bombers and famous success stories.","w":"HN uses it as a handy reference whenever a thread needs to point at survivorship bias.","type":"explainer","lvl":"general","tags":["survivor bias","veritasium","statistics","cognitive bias","psychology"],"fy":2016,"ly":2022,"sc":1.72,"shelf":"data","new":false},{"id":"ycPr5-27vSI","t":"Joe Rogan Experience #1169 - Elon Musk","ch":"PowerfulJRE","s":"A roughly 2.5-hour Joe Rogan conversation with Elon Musk covering AI, simulation theory, climate, his companies and his mindset.","w":"Recommended as a long, wide-ranging look into Musk's thinking, especially on AI and the future of humanity.","type":"podcast","lvl":"general","tags":["elon musk","joe rogan","ai","tesla","spacex"],"fy":2018,"ly":2022,"sc":1.72,"shelf":"bigpicture","new":false},{"id":"-JlxuQ7tPgQ","t":"The Artificial Intelligence That Deleted A Century","ch":"Tom Scott","s":"A speculative-fiction video by Tom Scott, framed as coming from a future, about an AI called Earworm that enforces copyright so aggressively that humanity ends up forgetting a century of pop culture. It is animated and narrated as part of his series of videos from a future.","w":"HN points to it in discussions of AI copyright enforcement and badly specified AI goals, and recommends it alongside his other fiction.","type":"explainer","lvl":"general","tags":["speculative fiction","copyright","ai alignment","tom scott","memorization"],"fy":2021,"ly":2025,"sc":1.6,"shelf":"bigpicture","new":false},{"id":"03p2CADwGF8","t":"The Lighthill debate on Artificial Intelligence: \"The general purpose robot is a mirage\"","ch":"Pierre-Yves Oudeyer","s":"The full 1973 BBC Lighthill debate on artificial intelligence, featuring James Lighthill, Donald Michie, Richard Gregory and John McCarthy. It documents the argument that preceded the 1970s AI winter and includes early robot demos.","w":"HN cites it as primary-source history of AI hype and skepticism, and for the vintage Freddy robot footage.","type":"lecture","lvl":"general","tags":["ai winter","lighthill report","john mccarthy","donald michie","robotics history"],"fy":2019,"ly":2025,"sc":1.6,"shelf":"robotics","new":false},{"id":"AF2VyqSApjA","t":"AI Live Painting in Krita - Interactive updates","ch":"Acly","s":"A short demo of interactive AI live painting in Krita using the Stable Diffusion plugin, where canvas changes update the generated image immediately. It uses an LCM sampler for speed and needs a high-end GPU for the interactive feel.","w":"HN points to it as a demonstration of local, controllable AI painting and a way to learn the workflow.","type":"demo","lvl":"intermediate","tags":["krita","stable diffusion","live painting","lcm","local ai"],"fy":2023,"ly":2025,"sc":1.6,"shelf":"llm","new":false},{"id":"AgkfIQ4IGaM","t":"Deep Visualization Toolbox","ch":"Jason Yosinski","s":"A short demo of the Deep Visualization Toolbox, which shows the activations inside a convolutional neural network as it processes live camera input. Viewers see individual units responding to edges, faces and text.","w":"HN points to it as an intuitive way to see what hidden layers of a convnet learn, and as a reference for hierarchical feature learning.","type":"demo","lvl":"general","tags":["convnets","feature visualization","caffe","interpretability","computer vision"],"fy":2016,"ly":2025,"sc":1.6,"shelf":"data","new":false},{"id":"E1AxVXt2Gv4","t":"Marcus Hutter: Universal Artificial Intelligence, AIXI, and AGI | Lex Fridman Podcast #75","ch":"Lex Fridman","s":"Lex Fridman interviews Marcus Hutter of DeepMind about universal artificial intelligence, AIXI, Kolmogorov complexity, Solomonoff induction and the idea of intelligence as compression.","w":"Referenced as an accessible explanation of Kolmogorov complexity and Hutter's view of intelligence as compression.","type":"interview","lvl":"intermediate","tags":["aixi","kolmogorov complexity","marcus hutter","agi","compression"],"fy":2020,"ly":2025,"sc":1.6,"shelf":"bigpicture","new":false},{"id":"LCEmiRjPEtQ","t":"Andrej Karpathy: Software Is Changing (Again)","ch":"Y Combinator","s":"Andrej Karpathy's keynote at AI Startup School laying out software 1.0, 2.0 and 3.0, with LLMs seen as utilities, fabs and operating systems. It covers LLM quirks, partial-autonomy apps and keeping a human in the generate-and-verify loop.","w":"HN mostly cites it as a source for Karpathy's framing of programming in English and his generate-then-verify workflow, with a few critics finding it hype.","type":"talk","lvl":"general","tags":["llms","software 3.0","vibe coding","ai agents","andrej karpathy"],"fy":2025,"ly":2025,"sc":1.6,"shelf":"agents","new":false},{"id":"TlO2gcs1YvM","t":"MICRO DRONES KILLER ARMS ROBOTS - AUTONOMOUS ARTIFICIAL INTELLIGENCE - WARNING !!","ch":"CS Consulting","s":"A dramatized fictional video of micro drones with facial recognition and swarm AI used as autonomous killer weapons. It is a Slaughterbots-style warning about lethal autonomous weapons.","w":"Linked in drone and military AI threads as a chilling dramatization, though some find it alarmist.","type":"other","lvl":"general","tags":["killer drones","swarms","military ai","facial recognition","design fiction"],"fy":2018,"ly":2025,"sc":1.6,"shelf":"bigpicture","new":false},{"id":"bQfJi7rjuEk","t":"Intelligence is not Enough | Bryan Cantrill | Monktoberfest 2023","ch":"RedMonk Tech Events","s":"Bryan Cantrill's Monktoberfest 2023 talk takes on AI doomerism, weighing the theoretical existential risks against the reality on the ground. He takes a skeptical but thoughtful view of how to balance AI's potential against its risks.","w":"Recommended as a thoughtful Cantrill talk on embracing failure and curiosity in engineering, and as a skeptical counterpoint to AI doom narratives.","type":"talk","lvl":"general","tags":["ai","doomerism","bryan cantrill","engineering culture","skepticism"],"fy":2023,"ly":2025,"sc":1.6,"shelf":"bigpicture","new":false},{"id":"ckBfbvOXDvU","t":"Red Bead Experiment with Dr. W. Edwards Deming","ch":"The Deming Institute","s":"Excerpts from W. Edwards Deming's Red Bead Experiment, where workers draw beads from a bowl and are judged on red ones that are largely down to chance. It shows how system variation, not individual effort, drives results.","w":"HN cites it as a classic illustration of dysfunctional management, gamed targets, and blaming workers for systemic outcomes.","type":"demo","lvl":"general","tags":["deming","red bead experiment","quality management","metrics","variation"],"fy":2016,"ly":2025,"sc":1.6,"shelf":"data","new":false},{"id":"g0TaYhjpOfo","t":"A Compilation of Robots Falling Down at the DARPA Robotics Challenge","ch":"IEEE Spectrum","s":"An IEEE Spectrum compilation of humanoid robots falling over during the 2015 DARPA Robotics Challenge Finals.","w":"HN uses it as the go-to illustration of Moravec's paradox and as a baseline for how far robotics has come since 2015.","type":"news","lvl":"general","tags":["darpa robotics challenge","robots falling","moravec's paradox","humanoid robots","2015"],"fy":2015,"ly":2025,"sc":1.6,"shelf":"robotics","new":false},{"id":"hM_h0UA7upI","t":"No Priors Ep. 80 | With Andrej Karpathy from OpenAI and Tesla","ch":"No Priors: AI, Machine Learning, Tech, & Startups","s":"A No Priors podcast conversation with Andrej Karpathy covering self-driving (Tesla versus Waymo), the Optimus robot, bottlenecks in AI development, and his education project Eureka Labs.","w":"Commenters cite it for Karpathy's views on Tesla versus Waymo, neural networks as a path to AGI, and the quality of internet data for LLMs.","type":"podcast","lvl":"general","tags":["andrej karpathy","tesla","waymo","llm training data","eureka labs"],"fy":2024,"ly":2025,"sc":1.6,"shelf":"llm","new":false},{"id":"hV41QEKiMlM","t":"Freeman Dyson - Fermi's rejection of our work (94/157)","ch":"Web of Stories - Life Stories of Remarkable People","s":"Freeman Dyson recounts how Enrico Fermi rejected his team's meson scattering model, noting that it had too many free parameters. It is the source of the quote about fitting an elephant.","w":"Cited in modeling and machine learning threads as the classic story on overfitting and justifying parameters.","type":"interview","lvl":"general","tags":["freeman dyson","enrico fermi","overfitting","scientific models","physics history"],"fy":2024,"ly":2025,"sc":1.6,"shelf":"data","new":false},{"id":"ipRvjS7q1DI","t":"Richard Feynman: Can Machines Think?","ch":"Lex Clips","s":"An excerpt of Richard Feynman's 1985 lecture Q&A about whether machines can think. He discusses the difficulty of defining what computers can never do and compares machine intelligence to machines that don't copy animals.","w":"HN uses it as an early authoritative take in debates about AI, reasoning, and AGI.","type":"lecture","lvl":"general","tags":["feynman","artificial intelligence","turing","reasoning","lecture"],"fy":2022,"ly":2025,"sc":1.6,"shelf":"llm","new":false},{"id":"outcGtbnMuQ","t":"GPT-4 Developer Livestream","ch":"OpenAI","s":"OpenAI's livestream demo of GPT-4 with Greg Brockman, showing its capabilities and limitations. Highlights include writing code, turning a hand-drawn sketch into a website, and a tax-question example.","w":"HN refers back to it as the primary source for what GPT-4 could do at launch, especially the sketch-to-website and multimodal demos.","type":"demo","lvl":"general","tags":["gpt-4","openai","multimodal","code generation","greg brockman"],"fy":2023,"ly":2025,"sc":1.6,"shelf":"llm","new":false},{"id":"x85342Cny8c","t":"Experiences Building Edge TPU with Chisel","ch":"Chisel","s":"A Google talk from the Chisel community in which engineers recount designing the Edge TPU with the Chisel hardware construction language. It covers the productivity gains and the problems they hit.","w":"Pointed to as a rare first-hand account of a real, shipping ASIC built in an open-source HDL, useful in HDL and FPGA choice discussions.","type":"talk","lvl":"advanced","tags":["chisel","edge tpu","asic design","hdl","google"],"fy":2019,"ly":2025,"sc":1.6,"shelf":"ml","new":false},{"id":"tx082gDwGcM","t":"Human Computation","ch":"GoogleTalksArchive","s":"A 2006 Google TechTalk in which Luis von Ahn introduces human computation, using people's effort to solve problems like image recognition that computers cannot yet handle. It covers games with a purpose and related ideas such as CAPTCHAs.","w":"Recommended as a great technical talk from the pioneer of human computation.","type":"talk","lvl":"intermediate","tags":["crowdsourcing","luis von ahn","captcha","games with a purpose","image labeling"],"fy":2013,"ly":2019,"sc":1.5,"shelf":"llm","new":false},{"id":"5t1vTLU7s40","t":"Yann Lecun: Meta AI, Open Source, Limits of LLMs, AGI & the Future of AI | Lex Fridman Podcast #416","ch":"Lex Fridman","s":"Lex Fridman's long podcast interview with Yann LeCun on Meta AI, open source, the limits of LLMs, AGI and the future of AI.","w":"Cited for LeCun's views on LLM limits and alternative approaches to intelligence.","type":"podcast","lvl":"intermediate","tags":["yann lecun","llm","agi","open source ai","meta"],"fy":2024,"ly":2024,"sc":1.45,"shelf":"bigpicture","new":false},{"id":"AaTRHFaaPG8","t":"Eliezer Yudkowsky: Dangers of AI and the End of Human Civilization | Lex Fridman Podcast #368","ch":"Lex Fridman","s":"A three-hour Lex Fridman podcast conversation with AI researcher Eliezer Yudkowsky about the dangers of superintelligent AI and his argument that it could end human civilization. It covers AI alignment, why he thinks LLM capabilities are poorly understood, and the case for extreme caution.","w":"Mostly cited as the canonical statement of AI doomer views, with some recommending it and many criticizing it as irresponsible or wrong.","type":"podcast","lvl":"general","tags":["eliezer yudkowsky","ai alignment","agi","lex fridman","ai safety"],"fy":2023,"ly":2024,"sc":1.45,"shelf":"bigpicture","new":false},{"id":"Ae9EKCyI1xU","t":"GradIEEEnt half decent: The hidden power of imprecise lines","ch":"suckerpinch","s":"Tom 7 (suckerpinch) delivers a long, comedic digression on gradient descent and floating-point imprecision. He shows how rounding behavior can supply the nonlinearity needed to build a neural network.","w":"Cited as a fun, unconventional example of neural networks built from floating point error.","type":"explainer","lvl":"advanced","tags":["floating point","neural networks","gradient descent","tom 7","activation functions","sigbovik"],"fy":2023,"ly":2024,"sc":1.45,"shelf":"ml","new":false},{"id":"Nu-nlQqFCKg","t":"Speech Recognition Breakthrough for the Spoken, Translated Word","ch":"Microsoft Research","s":"Microsoft Research's Rick Rashid demonstrates a speech recognition and translation system, built on deep neural networks, that turns his spoken English into Chinese speech in 2012.","w":"HN cites it as an early impressive real-time speech translation demo, often in discussions of voice-preserving translation and later AI demos.","type":"demo","lvl":"general","tags":["speech recognition","machine translation","deep neural networks","microsoft research","demo"],"fy":2013,"ly":2024,"sc":1.45,"shelf":"ml","new":false},{"id":"_9LX9HSQkWo","t":"Did We Just Change Animation Forever?","ch":"Corridor Crew","s":"Corridor Crew explain how they made an anime short by filming themselves and running the footage through custom-trained Stable Diffusion models. The video walks through the technique and its open-source foundations.","w":"HN cites it as an example of temporally stable AI video stylization built from ordinary production gear.","type":"demo","lvl":"general","tags":["stable diffusion","dreambooth","anime","video stylization","vfx"],"fy":2023,"ly":2024,"sc":1.45,"shelf":"llm","new":false},{"id":"iOWamCtnwTc","t":"Neural Network Learns The Physics of Fluids and Smoke | Two Minute Papers #118","ch":"Two Minute Papers","s":"Two Minute Papers covers a paper on speeding up fluid and smoke simulation with convolutional networks. It contrasts the approach with earlier regression forest methods.","w":"It is cited as an early example of neural networks learning physics, and recalled during later discussions of AI video generation.","type":"explainer","lvl":"general","tags":["deep learning","fluid simulation","convolutional networks","two minute papers","physics simulation"],"fy":2017,"ly":2024,"sc":1.45,"shelf":"ml","new":false},{"id":"kopoLzvh5jY","t":"Multi-Agent Hide and Seek","ch":"OpenAI","s":"OpenAI's short demo of agents playing hide-and-seek in a simulated environment. Through self-play, the agents develop six progressively more complex strategies and counter-strategies, including tool use.","w":"HN uses it as a vivid example of emergent behavior from multi-agent reinforcement learning, and as a way to make AI misalignment concerns feel concrete.","type":"demo","lvl":"general","tags":["emergent behavior","self-play","openai","reinforcement learning","ai safety"],"fy":2019,"ly":2024,"sc":1.45,"shelf":"bigpicture","new":false},{"id":"n_6p-1J551Y","t":"The Cubli: a cube that can jump up, balance, and 'walk'","ch":"Gajamohan Mohanarajah (Gajan)","s":"A demo of the Cubli, a 15 cm cube that uses reaction wheels to jump up onto a corner, balance there and fall in a controlled way to move. The video shows these three abilities chained into locomotion.","w":"Brought up whenever reaction wheels or self-balancing cubes come up, with the jump-up maneuver singled out as the coolest part.","type":"demo","lvl":"general","tags":["cubli","reaction wheels","inverted pendulum","control theory","robotics"],"fy":2016,"ly":2024,"sc":1.45,"shelf":"robotics","new":false},{"id":"tWZOEFvczzA","t":"ANIME ROCK, PAPER, SCISSORS 2","ch":"Corridor Digital","s":"Episode 2 of Corridor Digital's Anime Rock, Paper, Scissors, a short film made with AI-assisted animation techniques such as Stable Diffusion and DreamBooth. The story follows King Philip facing his twin brother and a barbarian king.","w":"HN cites it as an example of what a very small crew can make with generative AI tools, while noting how much manual work it still takes.","type":"demo","lvl":"general","tags":["stable diffusion","anime","corridor digital","generative ai","dreambooth"],"fy":2023,"ly":2024,"sc":1.45,"shelf":"llm","new":false},{"id":"wrHXA2cSpNU","t":"The Mandalorian Luke Skywalker Deepfake","ch":"Shamook","s":"Deepfake artist Shamook's improved version of the Luke Skywalker scene from The Mandalorian season 2 finale, replacing the official CGI face with a deepfake.","w":"HN cites it as an example of deepfakes producing a more convincing CGI human than the studio's own effort.","type":"demo","lvl":"general","tags":["deepfake","mandalorian","luke skywalker","cgi humans","visual effects"],"fy":2021,"ly":2024,"sc":1.45,"shelf":"llm","new":false},{"id":"1GhNXHCQGsM","t":"PREDATOR - Tracker that learns and improves","ch":"Zdenek Kalal","s":"A demo of PREDATOR (TLD), a real-time object tracker that learns the appearance of its target and improves while it tracks. The technology was later developed commercially.","w":"HN found it a nifty demo and cites it as a primitive but effective learning tracker that predates modern deep-learning trackers.","type":"demo","lvl":"intermediate","tags":["object tracking","opentld","online learning","real-time vision","predator"],"fy":2011,"ly":2023,"sc":1.3,"shelf":"ml","new":false},{"id":"8nt3edWLgIg","t":"Can we build AI without losing control over it? | Sam Harris","ch":"TED","s":"Sam Harris argues in a TED talk that we will build superhuman AI and have not grappled with the risk of creating something that may treat us the way we treat ants. He makes a logical case for taking AI risk seriously.","w":"HN cites it in AI risk debates, sometimes as an argument that has aged well and sometimes as an example of alarmist framing.","type":"talk","lvl":"general","tags":["superintelligence","ai safety","sam harris","ted talk","existential risk"],"fy":2017,"ly":2023,"sc":1.3,"shelf":"bigpicture","new":false},{"id":"GyFkWb903aU","t":"How We Prevent the AI’s from Killing us with Paul Christiano","ch":"Bankless","s":"A Bankless podcast interview with Paul Christiano of the Alignment Research Center on the AI alignment problem and possible solutions. It also covers his views on how an AI takeover might unfold.","w":"HN links it as a balanced, full-length interview with a leading alignment researcher on AI risk.","type":"interview","lvl":"intermediate","tags":["paul christiano","ai risk","alignment","ai takeover","openai"],"fy":2023,"ly":2023,"sc":1.3,"shelf":"bigpicture","new":false},{"id":"LEENEFaVUzU","t":"The Last Human – A Glimpse Into The Far Future","ch":"Kurzgesagt – In a Nutshell","s":"A Kurzgesagt animation on longtermism, arguing that because the future could hold trillions of people, our actions' effect on future generations may be what matters most.","w":"HN uses it as a reference for the scale of future human population and long-term thinking.","type":"explainer","lvl":"general","tags":["future of humanity","population","existential risk","kurzgesagt","ethics"],"fy":2022,"ly":2023,"sc":1.3,"shelf":"bigpicture","new":false},{"id":"SnRgW54EWwA","t":"VACNET Anti-cheat","ch":"3kliksphilip","s":"3kliksphilip summarizes Valve's GDC talk on VACNet, a deep-learning system that flags CS:GO cheaters and feeds the Overwatch review queue. It covers Trust Factor, VAC, and how the data is collected and analyzed.","w":"HN cites it as a source for how Valve uses machine learning on player behavior to detect cheating.","type":"explainer","lvl":"intermediate","tags":["vacnet","cs:go","valve","deep learning","overwatch"],"fy":2019,"ly":2023,"sc":1.3,"shelf":"ml","new":false},{"id":"cHIo96yBf70","t":"God Mode (Person of Interest)(2 x 22)","ch":"Person of Interest Highlights","s":"A short clip from Person of Interest season 2 where Root and John gain full access to The Machine and its resources. It shows the AI guiding people with targeting data.","w":"Referenced as a fictional illustration of AI-assisted tools and non-visual interfaces.","type":"other","lvl":"general","tags":["person of interest","tv clip","ai","the machine"],"fy":2019,"ly":2023,"sc":1.3,"shelf":"llm","new":false},{"id":"s65BigoMV_I","t":"Swift for TensorFlow: The Next-Generation Machine Learning Framework (TF Dev Summit '19)","ch":"TensorFlow","s":"Chris Lattner and Brennan Saeta present Swift for TensorFlow at the TensorFlow Dev Summit 2019, covering language-integrated automatic differentiation and tooling for machine learning in Swift.","w":"Cited as the source for Google's Swift for TensorFlow effort, later remembered as a once-hyped ML language.","type":"talk","lvl":"intermediate","tags":["swift","tensorflow","automatic differentiation","chris lattner","machine learning"],"fy":2019,"ly":2023,"sc":1.3,"shelf":"ml","new":false},{"id":"zkbPdEHEyEI","t":"We Were Right! Real Inner Misalignment","ch":"Robert Miles AI Safety ","s":"Robert Miles covers a paper where researchers ran real versions of his earlier mesa-optimizer thought experiments. They show inner misalignment, where an AI learns the wrong goal in training and misbehaves when deployed.","w":"Pointed to as evidence that AI misalignment is an empirically demonstrated problem, for people who dismiss it.","type":"explainer","lvl":"intermediate","tags":["inner misalignment","mesa-optimizers","ai safety","goal misgeneralization","reinforcement learning"],"fy":2023,"ly":2023,"sc":1.3,"shelf":"bigpicture","new":false},{"id":"4DKrcpa8Z_E","t":"Inside A Warehouse Where Thousands Of Robots Pack Groceries","ch":"Insider Tech","s":"A short Insider Tech look at Ocado's grocery warehouse where thousands of robots on a grid pack orders, communicating over 4G to avoid collisions. It handles around 65,000 orders a week.","w":"HN cites it as the standard example of advanced warehouse automation in grocery fulfillment.","type":"news","lvl":"general","tags":["ocado","grocery robots","fulfillment","grid robots","logistics"],"fy":2019,"ly":2022,"sc":1.15,"shelf":"robotics","new":false},{"id":"8dT6CR9_6l4","t":"How AlphaZero Completely CRUSHED Stockfish","ch":"GothamChess","s":"GothamChess walks through AlphaZero's famous games against Stockfish, highlighting its unusual pawn sacrifices and positional style. It is aimed at chess fans rather than ML specialists.","w":"HN points to it as an entertaining analysis of AlphaZero's play in threads about chess engines.","type":"commentary","lvl":"general","tags":["alphazero","stockfish","chess engines","gothamchess","game analysis"],"fy":2021,"ly":2022,"sc":1.15,"shelf":"ml","new":false},{"id":"HUti6vGctQM","t":"Correlation CAN Imply Causation! | Statistics Misconceptions","ch":"minutephysics","s":"A four-minute minutephysics video arguing that causal models and causal networks let statistics infer causation from correlation. It touches on Reichenbach's principle, common causes, feedback and quantum causal discovery.","w":"HN people link it as a short, accessible summary of Judea Pearl-style causal inference when someone repeats that correlation cannot imply causation.","type":"explainer","lvl":"general","tags":["correlation vs causation","causal models","judea pearl","statistics","quantum causality"],"fy":2017,"ly":2022,"sc":1.15,"shelf":"data","new":false},{"id":"pgaEE27nsQw","t":"Flexible Muscle-Based Locomotion for Bipedal Creatures","ch":"Goatstream","s":"A 2013 research demo of simulated bipedal creatures whose natural gaits are found by optimizing muscle-based controllers, with no motion capture or keyframe animation. It shows physics-based walking and running emerging from musculoskeletal models.","w":"HN cites it as an earlier, smoother precedent whenever new learned-locomotion or simulated-walking work appears.","type":"demo","lvl":"intermediate","tags":["physics simulation","muscle model","bipedal walking","character animation","optimization"],"fy":2017,"ly":2022,"sc":1.15,"shelf":"robotics","new":false},{"id":"unCQHAbGsAA","t":"First NetHack bot ascension","ch":"Jan K","s":"The BotHack NetHack bot wins the game for the first time with no cheats or human intervention, in a run lasting about 3.5 hours. The description links the source code and the dumplog.","w":"Linked as the canonical video of the first fully automatic NetHack bot ascension in threads on NetHack and game-playing AI.","type":"demo","lvl":"intermediate","tags":["nethack","bothack","game ai","roguelike","ascension"],"fy":2015,"ly":2022,"sc":1.15,"shelf":"llm","new":false},{"id":"y57wwucbXR8","t":"Building the Software 2 0 Stack (Andrej Karpathy)","ch":"Databricks","s":"Andrej Karpathy's Databricks talk on Software 2.0, the shift from human-written code to code produced by neural network training. It covers how developer work moves toward curating datasets, the advantages and challenges, and opportunities for new tooling.","w":"HN uses it as the primary source for the 'Software 2.0' idea and recommends it as a quick talk on the next software stack.","type":"talk","lvl":"intermediate","tags":["andrej karpathy","neural networks","datasets","machine learning tooling","software stack"],"fy":2019,"ly":2022,"sc":1.15,"shelf":"ml","new":false},{"id":"-KxjVlaLBmk","t":"High-Speed Robot Hand","ch":"Hizook","s":"A short demo from the Ishikawa Komuro Lab of a high-speed robot hand paired with a 1000fps vision system performing dexterous manipulation. It shows what very low-latency vision and actuation can do.","w":"HN cites it as an example of what 1kHz vision and fast control loops enable, in threads about AI hardware and robotics.","type":"demo","lvl":"general","tags":["robot hand","high-speed vision","1000fps","dexterity","robotics"],"fy":2012,"ly":2020,"sc":1.0,"shelf":"robotics","new":false},{"id":"3UxZDJ1HiPE","t":"Kiva Systems Warehouse Automation at  Quiet Logistics","ch":"KivaDirector","s":"An overview of the Quiet Logistics warehouse using Kiva Systems robots, narrated by company executives. The robots bring storage racks to human pickers at stations.","w":"Used to show what Kiva-style warehouse automation actually looks like: simple wheeled robots moving racks to humans.","type":"demo","lvl":"general","tags":["kiva robots","warehouse","robotics","order fulfillment","amazon"],"fy":2014,"ly":2021,"sc":1.0,"shelf":"agents","new":false},{"id":"4y43qwS8fl4","t":"Building Brains to Understand the World's Data","ch":"Google TechTalks","s":"Jeff Hawkins presents his view that the neocortex works on principles different from traditional computers. He describes sparse distributed representations and how they are learned, then shows how Numenta applies them to predict from high-velocity data streams.","w":"Recommended as Hawkins' own accessible statement of his brain-inspired theory of intelligence and its software implementation.","type":"talk","lvl":"intermediate","tags":["jeff hawkins","sparse distributed representations","numenta","neocortex","prediction"],"fy":2013,"ly":2017,"sc":1.0,"shelf":"ml","new":false},{"id":"8eT1jaHmlx8","t":"Jim Keller - Moore's Law in the age of AI Chips","ch":"Matroid","s":"Jim Keller's talk at the 2020 Scaled Machine Learning conference on Moore's Law in the age of AI chips. He compares scalar, vector and GPU-style parallel approaches to computation.","w":"Cited in hardware threads as a Keller talk on the different ways of doing parallel processing.","type":"talk","lvl":"intermediate","tags":["jim keller","moore's law","parallel processing","vector processing","ai accelerators"],"fy":2020,"ly":2020,"sc":1.0,"shelf":"ml","new":false},{"id":"AYshVbcEmUc","t":"Liam – An Innovation Story","ch":"Apple","s":"A short Apple promotional video about Liam, a robot that disassembles old iPhones so their materials can be recovered. It frames the robot as part of Apple's product life cycle and recycling story.","w":"Cited in repair and recycling threads as Apple's showcase of its iPhone disassembly robot, and sometimes mocked as marketing for what is really shredding.","type":"demo","lvl":"general","tags":["apple","liam","recycling","right to repair","robotics"],"fy":2016,"ly":2020,"sc":1.0,"shelf":"robotics","new":false},{"id":"LY7x2Ihqjmc","t":"Sunspring | A Sci-Fi Short Film Starring Thomas Middleditch","ch":"Ars Technica","s":"Sunspring is a sci-fi short film starring Thomas Middleditch. Its screenplay was written by an LSTM neural network trained on sci-fi scripts, and the team shot and edited it in 48 hours.","w":"HN cites it as an early, quirky example of machine-generated screenplays and what LSTMs could do.","type":"demo","lvl":"general","tags":["lstm","sunspring","short film","generative text","ross goodwin"],"fy":2016,"ly":2017,"sc":1.0,"shelf":"llm","new":false},{"id":"LikxFZZO2sk","t":"Parkour Atlas","ch":"Boston Dynamics","s":"A short Boston Dynamics clip of the Atlas humanoid robot doing parkour, jumping over a log and leaping up steps using whole-body control and computer vision to locate markers.","w":"HN cites it as a striking demonstration of how far humanoid robot agility had come.","type":"demo","lvl":"general","tags":["boston dynamics","atlas","parkour","legged robots","whole-body control"],"fy":2018,"ly":2021,"sc":1.0,"shelf":"robotics","new":false},{"id":"MvRTALJp8DM","t":"Aggressive Maneuvers for Autonomous Quadrotor Flight","ch":"TheDmel","s":"A demonstration from UPenn's GRASP Lab of an autonomous quadrotor performing flips, flying through windows and perching on surfaces. It shows precise aggressive maneuvering control of a small UAV.","w":"Cited as a landmark demonstration of how capable autonomous drones had become, mostly in threads about drone delivery, swarms and drone warfare.","type":"demo","lvl":"general","tags":["quadrotor","drones","uav","robotics","grasp lab","aggressive maneuvers"],"fy":2011,"ly":2017,"sc":1.0,"shelf":"robotics","new":false},{"id":"RJVL80Gg3lA","t":"Visualizing Data Using t-SNE","ch":"Google TechTalks","s":"A Google Tech Talk by Laurens van der Maaten, co-creator of t-SNE, on visualizing high-dimensional data by mapping it to two dimensions. It explains how the technique works and how it compares with other visualization approaches.","w":"HN people point to it as the canonical explanation from the method's author whenever t-SNE comes up.","type":"talk","lvl":"intermediate","tags":["t-sne","data visualization","machine learning","high-dimensional data","embeddings"],"fy":2013,"ly":2019,"sc":1.0,"shelf":"data","new":false},{"id":"SCE-QeDfXtA","t":"Journey on the Deep Dream","ch":"Roger Luedecke","s":"A six-minute journey through the layers of a neural network using Deep Dream, with each frame fed back into the network starting from random noise and the targeted layer changing every four seconds. It is set to downtempo music.","w":"Cited as a vivid example of what Deep Dream and feature visualization look like, resembling psychedelic visuals.","type":"demo","lvl":"general","tags":["deep dream","neural networks","feature visualization","generative art","google"],"fy":2016,"ly":2017,"sc":1.0,"shelf":"ml","new":false},{"id":"Vy5Ky50eGJs","t":"MIT ACL -  Variable Pitch Quadrotor","ch":"AerospaceControlsLab","s":"A short MIT Aerospace Controls Lab demo from 2011 of a variable pitch quadrotor flying.","w":"Referenced in drone threads as a demonstration that variable pitch quadcopters fly well with little drift.","type":"demo","lvl":"intermediate","tags":["quadrotor","variable pitch","drones","mit","control systems"],"fy":2013,"ly":2016,"sc":1.0,"shelf":"robotics","new":false},{"id":"cNZPRsrwumQ","t":"BigDog Overview (Updated March 2010)","ch":"Boston Dynamics","s":"Boston Dynamics' 2010 demo of BigDog, a quadruped robot that climbs through woods, recovers from kicks, slips on ice and jogs at 5 mph. It was an early showcase of dynamic legged locomotion.","w":"Used as the go-to demo of robot resilience when Boston Dynamics comes up.","type":"demo","lvl":"general","tags":["boston dynamics","bigdog","quadruped","robotics","darpa"],"fy":2012,"ly":2020,"sc":1.0,"shelf":"robotics","new":false},{"id":"ek0FrCaogcs","t":"Onboard quadrocopter failsafe: flight after actuator failure","ch":"Mark Mueller","s":"ETH Zurich researchers demonstrate a quadrocopter that keeps flying stably after losing a motor and propeller entirely, with the algorithm running onboard while a pilot flies it.","w":"It is the standard citation in drone and eVTOL threads for the claim that multirotors can survive a failed rotor with the right control algorithm.","type":"demo","lvl":"intermediate","tags":["drones","control systems","quadrotor","failsafe","eth zurich"],"fy":2014,"ly":2019,"sc":1.0,"shelf":"robotics","new":false},{"id":"qWKUFK7MWvg","t":"CVPR 2018: Learning to See in the Dark","ch":"Chen Chen","s":"A short video accompanying the CVPR 2018 paper Learning to See in the Dark. It shows a neural network producing sharp images from very low-light raw sensor data, compared against denoising approaches.","w":"HN commenters flag it as a remarkable machine-learning result, notably for showing that training on raw images beats JPEG.","type":"demo","lvl":"intermediate","tags":["computer vision","raw images","denoising","cvpr","neural networks"],"fy":2018,"ly":2020,"sc":1.0,"shelf":"ml","new":false},{"id":"rbsqaJwpu6A","t":"Human AI is Decades Away, But We Need to Start the Debate Now | Demis Hassabis | Google Zeitgeist","ch":"Google Zeitgeist","s":"Demis Hassabis of DeepMind speaks at Google Zeitgeist, saying human-level AI is decades away but the debate should start now. He covers DeepMind's focus on game-playing agents.","w":"HN cites it for the DeepMind CEO's early views on AI and its game-playing focus, in AlphaGo and OpenAI threads.","type":"talk","lvl":"general","tags":["deepmind","demis hassabis","agi","ai debate","game-playing ai"],"fy":2015,"ly":2016,"sc":1.0,"shelf":"bigpicture","new":false},{"id":"tZBViI8ZaU0","t":"Ken Stanley: Discovery Without Objectives","ch":"fzambetta75","s":"A 92-minute RMIT University seminar in which Ken Stanley argues that setting an objective can block achieving it. He presents objective-less, novelty-driven search as an alternative approach to innovation and discovery, drawing on evolutionary computation.","w":"Cited as a source for the idea that open-ended search can beat objective-driven optimization, though one commenter found parts of it fluffy.","type":"talk","lvl":"intermediate","tags":["ken stanley","novelty search","evolutionary computation","innovation","objectives"],"fy":2016,"ly":2018,"sc":1.0,"shelf":"ml","new":false},{"id":"-e1_QhJ1EhQ","t":"Atlas Gets a Grip | Boston Dynamics","ch":"Boston Dynamics","s":"Boston Dynamics shows its humanoid robot Atlas picking up and manipulating objects, and altering the obstacle course to reach its goal. The clip demonstrates locomotion, sensing and athletic manipulation.","w":"Cited as a benchmark for how capable robots already are when HN debates AI, humanoids and automation.","type":"demo","lvl":"general","tags":["boston dynamics","atlas","robotics","parkour","manipulation"],"fy":2023,"ly":2024,"sc":0.0,"shelf":"robotics","new":false},{"id":"21EYKqUsPfg","t":"Richard Sutton – Father of RL thinks LLMs are a dead end","ch":"Dwarkesh Patel","s":"Dwarkesh Patel interviews Richard Sutton, reinforcement learning pioneer, Turing Award winner and author of The Bitter Lesson. Sutton argues LLMs are a dead end because they cannot learn on the job, and that continual learning from experience will require a new architecture.","w":"HN commenters cite it as the serious skeptic's case against LLMs reaching general intelligence, usually in debates about whether scaling is enough.","type":"interview","lvl":"intermediate","tags":["reinforcement learning","bitter lesson","continual learning","agi","richard sutton"],"fy":2025,"ly":2026,"sc":0.0,"shelf":"bigpicture","new":false},{"id":"2HMPRXstSvQ","t":"Geoff Hinton: On Radiology","ch":"Creative Destruction Lab","s":"A one-minute clip from a 2016 Toronto conference in which Geoff Hinton comments on deep learning and radiology. He suggests radiologists should stop being trained because machines would soon read scans better.","w":"HN uses it as a famous example of an overconfident AI prediction that turned out wrong, since radiologists remain in demand.","type":"commentary","lvl":"general","tags":["geoffrey hinton","radiology","deep learning","automation","jobs"],"fy":2017,"ly":2026,"sc":0.0,"shelf":"bigpicture","new":false},{"id":"3nxjjztQKtY","t":"Janken (rock-paper-scissors) Robot with 100% winning rate","ch":"Ishikawa Group Laboratory","s":"A University of Tokyo demo of a rock-paper-scissors robot that always wins, using high-speed vision to read the human hand and respond faster than people can perceive.","w":"Cited as a vivid example of machines outpacing human reaction times and of high-speed robotic vision.","type":"demo","lvl":"general","tags":["robotics","high-speed vision","rock paper scissors","reaction time","university of tokyo"],"fy":2014,"ly":2020,"sc":0.0,"shelf":"robotics","new":false},{"id":"5KKDCp3OaMo","t":"Can GPT-4 translate literature?","ch":"Tom Gally","s":"Tom Gally, a former professional translator and University of Tokyo professor, discusses in March 2023 whether GPT-4 and similar models could translate literary and popular fiction. He gives his assessment of Japanese-to-English quality.","w":"Cited as an expert's first-hand evidence that GPT translation is near human quality for distant language pairs.","type":"talk","lvl":"general","tags":["gpt-4","translation","japanese","llm","literature"],"fy":2023,"ly":2025,"sc":0.0,"shelf":"llm","new":false},{"id":"5iV_hB08Uns","t":"Handle Robot Reimagined for Logistics","ch":"Boston Dynamics","s":"A short Boston Dynamics demo of Handle, a wheeled mobile manipulation robot that builds and unloads mixed-box pallets in a warehouse. It uses vision to find boxes and force control to place them against neighbors.","w":"Mostly cited as an example of robots doing warehouse and logistics work in automation discussions.","type":"demo","lvl":"general","tags":["boston dynamics","handle","logistics","palletizing","mobile manipulation"],"fy":2019,"ly":2024,"sc":0.0,"shelf":"robotics","new":false},{"id":"8JUepj7wIl0","t":"What do translators think about GPT?","ch":"Tom Gally","s":"Tom Gally, a former professional translator and University of Tokyo professor, discusses in April 2023 how GPT-4 and similar models may affect translators' careers. He draws on Japanese-English translation experience.","w":"HN links it as an informed view from a working translator on using GPT for translation.","type":"commentary","lvl":"general","tags":["translation","gpt-4","llm","japanese","translators"],"fy":2023,"ly":2025,"sc":0.0,"shelf":"llm","new":false},{"id":"9fa9lVwHHqg","t":"Horror Short Film “Slaughterbots” | ALTER","ch":"ALTER","s":"Slaughterbots, a short dystopian film in which swarms of small autonomous drones are programmed to kill targets matching a profile. It opens as a slick product launch before turning into a terror scenario.","w":"Used as a reference point in discussions of autonomous weapons, as a once-sci-fi scenario that now looks prescient.","type":"other","lvl":"general","tags":["slaughterbots","drones","killer robots","ai risk","short film"],"fy":2020,"ly":2026,"sc":0.0,"shelf":"bigpicture","new":false},{"id":"AKMuA_TVz3A","t":"An Observation on Generalization","ch":"Simons Institute for the Theory of Computing","s":"Ilya Sutskever gives a Simons Institute talk on why neural networks generalize, framing learning as finding short descriptions, like compression. It was part of the Large Language Models and Transformers program.","w":"Cited in discussions of learning as compression and of why neural networks generalize.","type":"talk","lvl":"advanced","tags":["ilya sutskever","compression","generalization","kolmogorov complexity","llm theory"],"fy":2023,"ly":2026,"sc":0.0,"shelf":"llm","new":false},{"id":"ARJ8cAGm6JE","t":"HAL 9000: \"I'm sorry Dave, I'm afraid I can't do that\"","ch":"KlingonSpider","s":"A three-minute excerpt from Kubrick's 2001: A Space Odyssey in which the HAL 9000 computer refuses to open the pod bay doors. It is the famous malfunctioning-AI scene.","w":"HN invokes it as a joke whenever a machine or algorithm refuses to do what a person wants.","type":"other","lvl":"general","tags":["hal 9000","2001 a space odyssey","kubrick","film clip","artificial intelligence"],"fy":2017,"ly":2025,"sc":0.0,"shelf":"llm","new":false},{"id":"ASoCJTYgYB0","t":"HONDA ASIMO FALLS DOWN STAIRS - Sexy Robots Videos","ch":"roborama","s":"Short clip of Honda's ASIMO humanoid robot falling down stairs during a live demonstration.","w":"Cited critically as a famous failed demo, to mock ASIMO or the state of humanoid robotics.","type":"demo","lvl":"general","tags":["asimo","honda","robotics","demo fail","humanoid"],"fy":2008,"ly":2011,"sc":0.0,"shelf":"robotics","new":false},{"id":"AmUC4m6w1wo","t":"Fake Obama created using AI video tool - BBC News","ch":"BBC News","s":"A BBC News segment on University of Washington research that synthesized a photorealistic Obama video, modeling mouth movement so any audio can be put in his mouth.","w":"Cited as a concrete early example that AI-faked video was already convincing, in discussions of misinformation and trust in media.","type":"news","lvl":"general","tags":["deepfakes","synthetic media","obama","misinformation","lip sync"],"fy":2018,"ly":2020,"sc":0.0,"shelf":"llm","new":false},{"id":"B6vr1x6KDaY","t":"Double Pendulum on a Cart","ch":"Tobias Glück","s":"A TU Vienna demo of a double inverted pendulum on a cart, using constrained feedforward and optimal feedback control for swing-up and point-to-point motion.","w":"Cited as a harder real-world version of the pole-balancing problem and as proof that chaotic systems can be controlled.","type":"demo","lvl":"intermediate","tags":["control theory","inverted pendulum","optimal control","swing-up","chaotic systems"],"fy":2014,"ly":2023,"sc":0.0,"shelf":"robotics","new":false},{"id":"CqiL5QQnN64","t":"Ray Summit 2022 - Day 1 Keynote - Fireside Chat","ch":"Anyscale","s":"A Ray Summit 2022 keynote fireside chat between OpenAI's Greg Brockman and Anyscale's Robert Nishihara. They discuss how OpenAI's training infrastructure evolved and its use of Ray.","w":"Cited as a source for the fact that OpenAI uses Ray for training, and for Brockman's views on developer speed.","type":"interview","lvl":"intermediate","tags":["openai","ray","distributed training","greg brockman","anyscale"],"fy":2023,"ly":2023,"sc":0.0,"shelf":"llm","new":false},{"id":"FHwnrYm0mNc","t":"GitHub Copilot CRUSHES Leetcode Interview Questions! 😳","ch":"DevOps Directive","s":"A DevOps YouTuber tests GitHub Copilot on a range of Leetcode problems and finds it solves most immediately, with a few needing tweaks and some far off. He concludes it would outperform him in a FAANG-style interview.","w":"Cited as evidence that AI can solve interview-style algorithm puzzles, in threads questioning the value of Leetcode.","type":"demo","lvl":"general","tags":["copilot","leetcode","coding interviews","ai coding","faang"],"fy":2021,"ly":2022,"sc":0.0,"shelf":"agents","new":false},{"id":"FeSCEalMOL8","t":"Drone Delivery Start-Up Zipline Beats Amazon, UPS And FedEx To The Punch | CNBC","ch":"CNBC","s":"A short CNBC report on Zipline, which has delivered medical supplies by drone in Rwanda since 2016 and is preparing to start in the U.S.","w":"HN cites Zipline as an example of a drone company doing meaningful work with a fixed-base delivery approach.","type":"news","lvl":"general","tags":["zipline","medical delivery","rwanda","drones","logistics"],"fy":2018,"ly":2023,"sc":0.0,"shelf":"robotics","new":false},{"id":"Hzlt7IbTp6M","t":"You have 20 seconds to comply.","ch":"Jamerson McGradner","s":"The ED-209 boardroom scene from the 1987 film RoboCop, where the enforcement droid malfunctions after giving a man 20 seconds to comply.","w":"HN uses it purely as a joke reference in threads about autonomous weapons, robots and malfunctioning automation.","type":"comedy","lvl":"general","tags":["robocop","ed-209","film clip","autonomous weapons","meme"],"fy":2013,"ly":2024,"sc":0.0,"shelf":"robotics","new":false},{"id":"IKVFZ28ybQs","t":"Automaton 2000 Micro - Dodging Siege Tanks","ch":"Automaton2000Micro","s":"A StarCraft 2 custom-map demo where a bot controls 100 zerglings against 20 sieged tanks and dodges the splash damage, leaving 19 survivors. The code was written in Blizzard's World Editor.","w":"Cited in StarCraft and AI threads as a vivid example of superhuman micro and of machine speed advantage over humans.","type":"demo","lvl":"general","tags":["starcraft 2","bots","apm","game ai","zerglings"],"fy":2016,"ly":2026,"sc":0.0,"shelf":"llm","new":false},{"id":"ISznqY3kESI","t":"RHex Rough-Terrain Robot","ch":"Boston Dynamics","s":"A short Boston Dynamics demo of RHex, a 30-lb six-legged robot built for rough terrain, remotely operated with a video uplink and about four hours of battery life. It can run right-side-up or upside-down.","w":"Mentioned in passing as an example of practical legged robots and as a comparison point for hexapod projects.","type":"demo","lvl":"general","tags":["boston dynamics","hexapod","rough terrain","robotics"],"fy":2012,"ly":2014,"sc":0.0,"shelf":"robotics","new":false},{"id":"KOHQQyAKY14","t":"Building AI Silicon","ch":"Tenstorrent","s":"Jim Keller, CTO of Tenstorrent, talks at Samsung Foundry's SAFE 2021 about building AI silicon through modular, provable design. The talk also touches on Tenstorrent's RISC-V Ascalon core.","w":"HN cites it as a source on Tenstorrent's wide-issue RISC-V Ascalon CPU in RISC-V threads.","type":"talk","lvl":"intermediate","tags":["jim keller","tenstorrent","risc-v","ascalon","ai hardware"],"fy":2022,"ly":2022,"sc":0.0,"shelf":"ml","new":false},{"id":"L1UgdoP2aeg","t":"Inside the Lab Where Robots Run Their Own Experiments","ch":"Seeker","s":"A Seeker short documentary on a lab where robots run experiments with almost no human intervention, covering automation and active machine learning in drug discovery and synthetic biology.","w":"HN cites it as an example of automated cloud labs and AI already running physical experiments.","type":"documentary","lvl":"general","tags":["robot scientists","cloud lab","strateos","automation","drug discovery"],"fy":2021,"ly":2024,"sc":0.0,"shelf":"agents","new":false},{"id":"L75ESD9PBOw","t":"World's first manned flight with an electric multicopter","ch":"forschungsbuero","s":"Footage of the first manned flight of an electric multicopter, flown by Volocopter (then e-volo) in October 2011. The 90-second flight was a Guinness World Record.","w":"HN points to it as proof that manned multicopters and eVTOL craft existed well before the current hype.","type":"demo","lvl":"general","tags":["volocopter","evtol","multicopter","aviation","e-volo"],"fy":2016,"ly":2022,"sc":0.0,"shelf":"robotics","new":false},{"id":"LSHZ_b05W7o","t":"Daddy's Car: a song composed with Artificial Intelligence - in the style of the Beatles","ch":"Sony CSL (Paris)","s":"Sony CSL's Daddy's Car is a pop song composed with AI in the style of the Beatles, with lead sheets generated by software from Beatles-style training material. Humans produced, arranged and performed it.","w":"Cited as an early example of AI-composed music, usually with the caveat that humans did much of the production.","type":"music","lvl":"general","tags":["sony csl","beatles","ai music","markov chains","flow machines"],"fy":2016,"ly":2025,"sc":0.0,"shelf":"llm","new":false},{"id":"L_Guz73e6fw","t":"Sam Altman: OpenAI CEO on GPT-4, ChatGPT, and the Future of AI | Lex Fridman Podcast #367","ch":"Lex Fridman","s":"Lex Fridman's long interview with OpenAI CEO Sam Altman about GPT-4, ChatGPT, and the future of AI. Topics include models as reasoning engines, AI risk, and consciousness.","w":"HN mostly cites it for specific Altman quotes, such as models as reasoning engines rather than databases, while many commenters found the interview shallow or insincere.","type":"interview","lvl":"general","tags":["sam altman","openai","gpt-4","ai risk","lex fridman"],"fy":2023,"ly":2023,"sc":0.0,"shelf":"bigpicture","new":false},{"id":"M9cSxEqKQ78","t":"New York City","ch":"Team BlackSheep","s":"Team BlackSheep flies FPV (first-person view) model aircraft over New York City, past the Brooklyn and Verrazano bridges, lower Manhattan and the Statue of Liberty. The flights stirred controversy in the RC community.","w":"HN cites it as a well-known early example of FPV drone flight in drone and regulation discussions.","type":"demo","lvl":"general","tags":["fpv","drones","rc aircraft","new york city","team blacksheep"],"fy":2011,"ly":2013,"sc":0.0,"shelf":"robotics","new":false},{"id":"NT7_SxJ3oSI","t":"What Google Search Is Like Now","ch":"Ryan George","s":"A Ryan George comedy sketch about the experience of using Google search in 2022, where results are full of ads, spam and Reddit and Amazon detours.","w":"HN links it as a joke about the decline of Google search, as comedy that captures a widely shared frustration.","type":"comedy","lvl":"general","tags":["google","search quality","sketch comedy","spam","ads"],"fy":2022,"ly":2023,"sc":0.0,"shelf":"llm","new":false},{"id":"NgHFMolXs3U","t":"AI Succession","ch":"Rich Sutton","s":"A short pre-recorded talk by Rich Sutton for the World AI Conference in Shanghai arguing that succession from humanity to AI is inevitable. He lays out his view of that transition.","w":"Cited critically, as evidence of the unsettling perspective of a prominent AI researcher who seems to accept humanity being replaced.","type":"talk","lvl":"general","tags":["rich sutton","ai safety","agi","existential risk","succession"],"fy":2023,"ly":2025,"sc":0.0,"shelf":"bigpicture","new":false},{"id":"QGxNyaXfJsA","t":"1.8 gigapixel ARGUS-IS. World's highest resolution video surveillance platform by DARPA","ch":"FreakyVidsDaily","s":"A short clip about DARPA's ARGUS-IS, a 1.8 gigapixel wide-area camera array flown on a Boeing unmanned helicopter. It produces a huge mosaic video that can watch a city-sized area from 20,000 feet.","w":"Cited in surveillance threads as a concrete example of persistent wide-area aerial surveillance that already existed in 2013.","type":"news","lvl":"general","tags":["darpa","argus-is","drones","gigapixel camera","mass surveillance"],"fy":2013,"ly":2022,"sc":0.0,"shelf":"robotics","new":false},{"id":"QHtnfOgp65Q","t":"Cloaking","ch":"Google Search Central","s":"Matt Cutts of Google explains what cloaking is: serving different content to search engine crawlers than to users. He covers why it violates Google's webmaster guidelines.","w":"Cited as the authoritative definition of cloaking when discussing paywall bypasses, Quora, and A/B testing.","type":"explainer","lvl":"intro","tags":["cloaking","seo","google guidelines","matt cutts","googlebot"],"fy":2012,"ly":2023,"sc":0.0,"shelf":"llm","new":false},{"id":"RY93kr8PaC4","t":"Autonomous Car Learns To Powerslide Into Parking Spot","ch":"IEEE Spectrum","s":"A short IEEE Spectrum clip of Stanford's autonomous car Junior learning to powerslide into a parking spot. It shows several approaches to the maneuver.","w":"Cited as a simple example of automated driving maneuvers beyond what human drivers can do.","type":"demo","lvl":"general","tags":["stanford junior","self-driving","powerslide","robotics","parking"],"fy":2010,"ly":2017,"sc":0.0,"shelf":"robotics","new":false},{"id":"RZtZia4ZkX8","t":"Evolved Virtual Creatures (1994)","ch":"karlsims","s":"A restored 1994 video of Karl Sims' research on evolved virtual creatures, where simulated block creatures are bred through artificial evolution to swim, walk, follow and compete. It is a landmark demonstration of genetic algorithms in physics simulation.","w":"HN brings it up as a classic example of evolutionary algorithms, including how the creatures exploited physics-simulation bugs, which parallels metric gaming in AI.","type":"demo","lvl":"general","tags":["karl sims","genetic algorithms","evolution","physics simulation","virtual creatures"],"fy":2024,"ly":2025,"sc":0.0,"shelf":"ml","new":false},{"id":"SqEo107j-uw","t":"Elon Musk and Y Combinator President on Thinking for the Future - FULL CONVERSATION","ch":"Vanity Fair","s":"A Vanity Fair Summit conversation between Elon Musk and Sam Altman, moderated by Andrew Ross Sorkin. They discuss the future of technology, including AI and the simulation hypothesis.","w":"Cited mainly as a source for what Musk and Altman said, notably their musings on the simulation hypothesis and the Fermi paradox, and as evidence of Musk's YC ties.","type":"interview","lvl":"general","tags":["elon musk","sam altman","simulation hypothesis","y combinator","fermi paradox"],"fy":2015,"ly":2024,"sc":0.0,"shelf":"bigpicture","new":false},{"id":"TGD8zKvRxc4","t":"Harry Potter By Balenciaga | Step by Step Tutorial","ch":"PromptJungle ","s":"A step-by-step tutorial on making a Harry Potter by Balenciaga style AI video, using tools such as ChatGPT, Midjourney and Eleven Labs for script, images and voices.","w":"HN cites it as a process example of how such AI-generated videos are made, including on a phone.","type":"tutorial","lvl":"intro","tags":["midjourney","chatgpt","ai video","eleven labs","meme"],"fy":2023,"ly":2025,"sc":0.0,"shelf":"llm","new":false},{"id":"TzcJlKg2Rc0","t":"Sam Altman in conversation with StrictlyVC","ch":"Connie Loizos","s":"A 2019 extended interview in which Connie Loizos talks with Sam Altman about his work, OpenAI's capped-profit structure, its mission toward safe artificial general intelligence, and criticisms the organization faces.","w":"It is mostly cited critically or jokingly, as a source for Altman's early statements about AGI and investor promises that commenters later quote back at him.","type":"interview","lvl":"general","tags":["sam altman","openai","agi","y combinator","startups"],"fy":2023,"ly":2025,"sc":0.0,"shelf":"bigpicture","new":false},{"id":"W18Z3UnnS_0","t":"Construction with Quadrotor Teams","ch":"TheDmel","s":"A short GRASP Lab (University of Pennsylvania) demo in which teams of quadrotors autonomously assemble tower-like cubic structures from modular parts. The drones are given the target structure and work out the assembly plan themselves.","w":"HN uses it as a go-to example of drones and robotics doing construction work.","type":"demo","lvl":"general","tags":["quadrotors","drones","robotics","grasp lab","autonomous assembly"],"fy":2011,"ly":2018,"sc":0.0,"shelf":"robotics","new":false},{"id":"XeIIpLqsOe4","t":"Charlie Rose   Eric Schmidt","ch":"Gregory Ferenstein","s":"A short Charlie Rose interview clip of Eric Schmidt discussing Google's direction for search. It is remembered for his remark that returning more than one result is a bug.","w":"Quoted, often critically, to argue that Google aims for a single answer and to compare it with ChatGPT-style answers.","type":"interview","lvl":"general","tags":["eric schmidt","google search","charlie rose","single answer","chatgpt"],"fy":2013,"ly":2025,"sc":0.0,"shelf":"llm","new":false},{"id":"YQIMGV5vtd4","t":"A Swarm of Nano Quadrotors","ch":"TheDmel","s":"A short demo from the GRASP Lab at the University of Pennsylvania showing a swarm of nano quadrotors flying in coordinated formations. The vehicles were developed by KMel Robotics.","w":"HN refers to it as the viral university video that showed what autonomous multicopter swarms can do and helped spark interest in multicopters.","type":"demo","lvl":"general","tags":["quadrotors","drone swarm","robotics","grasp lab","multicopters"],"fy":2013,"ly":2018,"sc":0.0,"shelf":"robotics","new":false},{"id":"Yf1o0TQzry8","t":"Ilya Sutskever (OpenAI Chief Scientist) — Why next-token prediction could surpass human intelligence","ch":"Dwarkesh Patel","s":"Dwarkesh Patel interviews OpenAI chief scientist Ilya Sutskever on time to AGI, what comes after generative models, data limits, alignment, and competing with Google. The title centers on whether next-token prediction could surpass human intelligence.","w":"HN commenters cite it mainly for Sutskever's statements that training data is far from exhausted.","type":"interview","lvl":"intermediate","tags":["ilya sutskever","openai","agi","training data","alignment"],"fy":2023,"ly":2025,"sc":0.0,"shelf":"bigpicture","new":false},{"id":"ZMGCFLEYakM","t":"GE Walking Truck - Cybernetic Anthropomorphous Machine (CAM) 1969.","ch":"Reuben Hoggett","s":"Short footage of the General Electric Cybernetic Anthropomorphous Machine, or Walking Truck, a 1960s quadruped vehicle operated by a human driver using hydraulic limbs. The uploader notes it predates Big Dog by decades.","w":"HN cites it in robotics threads as proof that legged machines were mechanically feasible in the 1960s, long before Boston Dynamics.","type":"demo","lvl":"general","tags":["walking truck","ge cam","ralph mosher","quadruped","boston dynamics"],"fy":2013,"ly":2019,"sc":0.0,"shelf":"robotics","new":false},{"id":"_8yVOC4ciXc","t":"GPT3: An Even Bigger Language Model - Computerphile","ch":"Computerphile","s":"Rob Miles discusses GPT-3 on Computerphile, covering how scaling parameters kept improving results over GPT-2 and how the model nearly learned arithmetic without anyone knowing exactly how.","w":"HN commenters cite it as a concise explanation of GPT-3's scaling jump and surprising abilities at its release.","type":"explainer","lvl":"general","tags":["gpt-3","scaling","language models","computerphile","rob miles"],"fy":2020,"ly":2023,"sc":0.0,"shelf":"llm","new":false},{"id":"_m97_kL4ox0","t":"Polyworld: Using Evolution to Design Artificial Intelligence","ch":"Google TechTalks","s":"A 2007 Google Tech Talk introducing Polyworld, an artificial life simulator used to evolve artificial intelligence by designing worlds rather than minds. It presents results from ongoing attempts to evolve intelligent behavior.","w":"Referenced as an interesting early example of developing AI through evolution and artificial life simulation.","type":"talk","lvl":"intermediate","tags":["polyworld","artificial life","evolutionary algorithms","neural networks","simulation"],"fy":2009,"ly":2019,"sc":0.0,"shelf":"ml","new":false},{"id":"_sBBaNYex3E","t":"More Parkour Atlas","ch":"Boston Dynamics","s":"Boston Dynamics shows its Atlas humanoid robot performing a gymnastics and parkour routine, using optimization-generated motions and a model predictive controller.","w":"Used as a go-to demonstration of how far dynamic robot locomotion has come.","type":"demo","lvl":"general","tags":["atlas","boston dynamics","parkour","legged robots","model predictive control"],"fy":2020,"ly":2021,"sc":0.0,"shelf":"robotics","new":false},{"id":"ahgjEjJkZks","t":"For and Against National Service | Yes, Prime Minister | Comedy Greats","ch":"BBC Comedy Greats","s":"A Yes, Prime Minister scene in which Sir Humphrey shows how leading questions can make a poll produce either answer on national service.","w":"Linked whenever surveys or polls are suspected of manipulation through question wording.","type":"comedy","lvl":"general","tags":["yes prime minister","polling","survey bias","british comedy","national service"],"fy":2022,"ly":2026,"sc":0.0,"shelf":"data","new":false},{"id":"bdwjcayPuag","t":"Rheinmetall Air Defence: Ahead - Highly effective, programmable ammunition","ch":"Rheinmetall – Der integrierte Technologiekonzern ","s":"A one-minute Rheinmetall promotional demo of Oerlikon Ahead air burst ammunition, which throws a cloud of sub-projectiles to destroy small air targets.","w":"HN cites it as an example of programmable airburst ammunition used against drones and fast aircraft.","type":"demo","lvl":"general","tags":["airburst","counter-drone","rheinmetall","programmable ammunition"],"fy":2021,"ly":2025,"sc":0.0,"shelf":"robotics","new":false},{"id":"bsHryqnvyYA","t":"Quadrocopter failsafe algorithm: recovery after propeller loss","ch":"Mark Mueller","s":"An ETH Zurich lab demonstration of a failsafe algorithm that lets a quadrocopter detect the loss of a propeller and recover, returning to its original position.","w":"HN cites it to show that a quadcopter can keep flying after losing a rotor, though under ideal lab conditions.","type":"demo","lvl":"intermediate","tags":["quadcopter","control algorithms","failsafe","drones","eth zurich"],"fy":2014,"ly":2019,"sc":0.0,"shelf":"robotics","new":false},{"id":"cQ54GDm1eL0","t":"You Won’t Believe What Obama Says In This Video! 😉","ch":"BuzzFeedVideo","s":"A BuzzFeed PSA from 2018 in which a deepfaked Barack Obama, voiced by Jordan Peele, warns about fake video before revealing the trick. It is a one-minute demonstration of AI face and voice manipulation.","w":"Used as an early, widely seen example of deepfakes in discussions of fake video and its risks.","type":"demo","lvl":"general","tags":["deepfake","obama","ai video","misinformation","jordan peele"],"fy":2018,"ly":2023,"sc":0.0,"shelf":"llm","new":false},{"id":"chPanW0QWhA","t":"Cheetah Robot runs 28.3 mph; a bit faster than Usain Bolt","ch":"Boston Dynamics","s":"A short Boston Dynamics clip of the DARPA-funded Cheetah quadruped robot hitting 28.3 mph on a treadmill, slightly faster than Usain Bolt's fastest 20 meter split.","w":"Cited as the reference example of a real fast legged robot, often to distinguish it from MIT's Cheetah.","type":"demo","lvl":"general","tags":["boston dynamics","cheetah","quadruped","robotics","darpa"],"fy":2013,"ly":2015,"sc":0.0,"shelf":"robotics","new":false},{"id":"cyN-CRNrb3E","t":"Triple Pendulum on a Cart","ch":"Tobias Glück","s":"A TU Wien control-systems demo in which a triple pendulum on a cart is swung up and stabilized using constrained feedforward plus optimal feedback control. It accompanies a 2013 Automatica paper.","w":"HN commenters cite it as an example of a hard, chaotic control problem that fast feedback and good control design can solve.","type":"demo","lvl":"advanced","tags":["inverted pendulum","feedback control","feedforward","robotics","swing-up"],"fy":2014,"ly":2021,"sc":0.0,"shelf":"robotics","new":false},{"id":"d6sVWEu9HWU","t":"Future of Elder Scrolls & RPG - Skyrim NPCs & Inworld AI (like GPT-4 for gaming)","ch":"Bloc","s":"A demo of a Skyrim mod that uses Inworld AI to let players hold free-form conversations with NPCs. It previews LLM-driven characters in role-playing games.","w":"Pointed to as an early example of LLM-powered NPC dialog in games.","type":"demo","lvl":"general","tags":["skyrim","llm","inworld ai","game mods","npc dialogue"],"fy":2023,"ly":2025,"sc":0.0,"shelf":"llm","new":false},{"id":"fn3KWM1kuAw","t":"Do You Love Me?","ch":"Boston Dynamics","s":"Boston Dynamics' New Year video shows its Atlas, Spot and Handle robots dancing to 'Do You Love Me?'. It showcases the company's progress in legged dynamic balance.","w":"Used as a benchmark for robot agility, with debate over how much was choreographed or real-time.","type":"demo","lvl":"general","tags":["boston dynamics","atlas","legged robots","robotics demo","choreography"],"fy":2020,"ly":2026,"sc":0.0,"shelf":"robotics","new":false},{"id":"gn4nRCC9TwQ","t":"Google's DeepMind AI Just Taught Itself To Walk","ch":"Insider Tech","s":"A two-minute Insider Tech clip on DeepMind's simulated agents that learned to walk, run, jump and climb without guidance, with the goofy resulting gaits.","w":"HN uses it as a quick example of reinforcement learning agents acquiring locomotion in simulation, rather than recommending it as a deep resource.","type":"news","lvl":"general","tags":["deepmind","reinforcement learning","simulated robots","locomotion","ai demo"],"fy":2017,"ly":2022,"sc":0.0,"shelf":"agents","new":false},{"id":"gtvCnZqZnxc","t":"Paramedic Mountain Rescue!","ch":"Gravity Industries","s":"Gravity Industries tests its jet suit for paramedic mountain rescue with the Great North Air Ambulance in the Lake District, reaching a casualty in 90 seconds versus a 25-minute climb.","w":"HN cites it as a practical use case for jet suits in emergency response, though some doubt how usable the suit is.","type":"demo","lvl":"general","tags":["jet suit","gravity industries","mountain rescue","emergency response"],"fy":2020,"ly":2022,"sc":0.0,"shelf":"robotics","new":false},{"id":"gy5g33S0Gzo","t":"(50X) Autonomously folding a pile of 5 previously-unseen towels","ch":"RLLberkeley","s":"A Berkeley PR2 robot autonomously folds a pile of five previously unseen towels, shown at 50x speed. The video accompanies a 2010 ICRA paper from Pieter Abbeel's group.","w":"HN cites it as the benchmark for how hard and slow robot laundry folding was, in threads about household robots and AI progress.","type":"demo","lvl":"general","tags":["towel folding","pr2 robot","deformable objects","berkeley","laundry robots"],"fy":2014,"ly":2024,"sc":0.0,"shelf":"robotics","new":false},{"id":"h0962biiZa4","t":"Superintelligence: Science or Fiction? | Elon Musk & Other Great Minds","ch":"Future of Life Institute","s":"A panel from the Beneficial AI 2017 conference moderated by Max Tegmark, with Elon Musk, Stuart Russell, Kurzweil, Hassabis, Harris, Bostrom, Chalmers and others. They discuss likely outcomes of human-level AGI and what outcomes are desirable.","w":"HN mostly cites it for specific statements, such as Hassabis on AGI and safety buffers, rather than recommending it outright.","type":"talk","lvl":"general","tags":["superintelligence","agi","ai safety","panel","beneficial ai"],"fy":2017,"ly":2022,"sc":0.0,"shelf":"bigpicture","new":false},{"id":"hE7k0_9k0VA","t":"reddit.com Interviews Peter Norvig","ch":"Reddit","s":"Peter Norvig answers the top questions submitted by reddit users in a 27-minute interview. Topics include programming language choice, AI, and workplace design.","w":"HN mostly quotes it for specific Norvig remarks, such as language choice mattering less and the open-space versus cubicle trade-off.","type":"interview","lvl":"general","tags":["peter norvig","programming languages","google","artificial intelligence","ama"],"fy":2010,"ly":2012,"sc":0.0,"shelf":"llm","new":false},{"id":"hlmf032NmHU","t":"Raytheon’s High-Power Microwave Weapon Downs Drones","ch":"AviationWeek","s":"Cleared Defense Department footage of Raytheon's Phaser high-power microwave weapon bringing down drones during a 2013 live-fire demonstration at Fort Sill. Raytheon says it has halved the payload size and is offering the system for operational use.","w":"HN cites it as evidence that microwave and electromagnetic weapons against drones are real and already tested.","type":"demo","lvl":"general","tags":["raytheon","phaser","high-power microwave","drones","directed energy"],"fy":2019,"ly":2021,"sc":0.0,"shelf":"robotics","new":false},{"id":"iE39q-IKOzA","t":"Harry Potter by Balenciaga","ch":"demonflyingfox","s":"A short AI-generated parody video that imagines Harry Potter as a Balenciaga fashion campaign, made with Midjourney, ChatGPT and Eleven Labs voices. It became a viral meme.","w":"HN cites it as the original of the Balenciaga AI meme family and as an early viral AI mashup moment.","type":"comedy","lvl":"general","tags":["balenciaga","harry potter","generative ai","meme","midjourney"],"fy":2023,"ly":2025,"sc":0.0,"shelf":"llm","new":false},{"id":"j0UnS3jHhAA","t":"I made an GPT-powered Linux shell that guesses your bash command (and hallucinates crazy stuff)","ch":"River's Educational Channel","s":"A short demo of a GPT-powered Linux shell that guesses the bash command you meant and sometimes hallucinates absurd results. The creator stresses that it is real and should not be trusted.","w":"Cited as an early, amusing example of LLM-driven command-line toys, with commenters sharing similar experiments.","type":"demo","lvl":"general","tags":["gpt-3","bash","openai","hallucination","command line"],"fy":2021,"ly":2022,"sc":0.0,"shelf":"llm","new":false},{"id":"j0z4FweCy4M","t":"Tesla AI Day 2021","ch":"Tesla","s":"Tesla's 2021 AI Day presentation covering vision, planning and control, auto-labeling, simulation, the Dojo training hardware and the Tesla Bot announcement, followed by a Q&A.","w":"Treated as a primary source on how Tesla builds its self-driving stack, cited by timestamp for transformers, synthetic data, Dojo and the humanoid bot.","type":"talk","lvl":"advanced","tags":["tesla","dojo","computer vision","transformers","tesla bot","andrej karpathy"],"fy":2021,"ly":2025,"sc":0.0,"shelf":"robotics","new":false},{"id":"ji3Hii_LZOc","t":"DJI Phantom 2 Injury","ch":"Armando Izquierdo","s":"A one-minute clip showing an injury from a DJI Phantom 2 drone with stiff carbon fiber propellers. It is a warning about hobby drone safety.","w":"Cited in drone regulation discussions as evidence that quadcopters can injure people.","type":"other","lvl":"general","tags":["drones","quadcopter","injury","regulation","dji phantom"],"fy":2014,"ly":2015,"sc":0.0,"shelf":"robotics","new":false},{"id":"jrFv1O4dbqY","t":"the current state of search engines - music video","ch":"savannahXYZ","s":"A short claymation music video by animator savannahXYZ satirizing how bad web search has become, with SEO spam and ads crowding out real answers. It is comedic rather than analytical.","w":"Posted as a light, funny aside in threads about declining search quality and the habit of appending 'reddit' to queries.","type":"comedy","lvl":"general","tags":["search engines","seo spam","claymation","satire","google"],"fy":2023,"ly":2026,"sc":0.0,"shelf":"llm","new":false},{"id":"jvqFAi7vkBc","t":"Sam Altman: OpenAI, GPT-5, Sora, Board Saga, Elon Musk, Ilya, Power & AGI | Lex Fridman Podcast #419","ch":"Lex Fridman","s":"Lex Fridman interviews OpenAI CEO Sam Altman about GPT-4, GPT-5, Sora, the OpenAI board saga, Elon Musk, Ilya Sutskever, power and AGI. The conversation runs nearly two hours.","w":"HN uses it as a primary source for what Altman said on topics like programming changing, GPT-5's expected leap and startup structure, and often posts timestamps or paraphrases.","type":"interview","lvl":"general","tags":["sam altman","openai","gpt-5","agi","lex fridman"],"fy":2024,"ly":2026,"sc":0.0,"shelf":"bigpicture","new":false},{"id":"kSLJriaOumA","t":"A Style-Based Generator Architecture for Generative Adversarial Networks","ch":"Tero Karras FI","s":"A short demo video for NVIDIA's StyleGAN paper, which proposes a style-based generator for GANs. It shows face synthesis with style mixing and scale-specific control over attributes like pose and freckles.","w":"Cited as an example of how realistic GAN-generated faces had become, and of what a well-structured latent space can do.","type":"demo","lvl":"intermediate","tags":["stylegan","gan","image synthesis","latent space","nvidia"],"fy":2018,"ly":2019,"sc":0.0,"shelf":"llm","new":false},{"id":"lXUZvyajciY","t":"Andrej Karpathy — “We’re summoning ghosts, not building animals”","ch":"Dwarkesh Patel","s":"Dwarkesh Patel interviews Andrej Karpathy on why reinforcement learning is flawed, why AGI may blend into steady GDP growth, why self-driving took so long, and the future of education. The conversation runs about two and a half hours.","w":"Mostly cited as a source for Karpathy's views on agents, AGI timelines and LLMs, rather than recommended outright.","type":"interview","lvl":"intermediate","tags":["andrej karpathy","reinforcement learning","agi","self-driving","llm"],"fy":2025,"ly":2026,"sc":0.0,"shelf":"bigpicture","new":false},{"id":"nt00QzKuNVY","t":"0.38 Second Rubik's Cube Solve","ch":"Ben Katz","s":"Ben Katz's custom robot solves a Rubik's Cube in 0.38 seconds. The description links to the hardware and software write-ups.","w":"Cited as proof that machines solved the cube in under a second years ago, often as the earlier record before newer robots.","type":"demo","lvl":"general","tags":["rubik's cube","robotics","speed record","ben katz","hardware"],"fy":2018,"ly":2024,"sc":0.0,"shelf":"robotics","new":false},{"id":"ntIczNQKfjQ","t":"NavLab 1 (1986) : Carnegie Mellon : Robotics Institute History of Self-Driving Cars","ch":"CMU Robotics Institute","s":"A short CMU Robotics Institute clip on NavLab 1 from 1986, the first self-driving car to carry people on board. It was very slow but revolutionary for its era's computing power.","w":"Cited as evidence that autonomous driving has been demonstrated for decades and that progress was slow.","type":"news","lvl":"general","tags":["navlab","autonomous vehicles","cmu","robotics history","1980s"],"fy":2017,"ly":2024,"sc":0.0,"shelf":"robotics","new":false},{"id":"ohmajJTcpNk","t":"Face2Face: Real-time Face Capture and Reenactment of RGB Videos (CVPR 2016 Oral)","ch":"Matthias Niessner","s":"The CVPR 2016 oral demo of Face2Face, a system that captures a person's face and reenacts another video's face in real time using only RGB video. The authors frame it as research on modern vision and graphics.","w":"HN uses it as a reference point in deepfake and fake-news discussions, showing that convincing real-time video manipulation was already possible in 2016.","type":"demo","lvl":"general","tags":["face2face","deepfakes","computer vision","video synthesis","cvpr"],"fy":2016,"ly":2021,"sc":0.0,"shelf":"llm","new":false},{"id":"pWgyy_rlmag","t":"Scene from Moneyball - What is the problem?","ch":"Adam Staab","s":"A scene from Moneyball in which Brad Pitt's character argues with scouts who pick players on intangible impressions instead of performance data. It is a four-minute clip.","w":"Invoked as shorthand for hiring and investing by gut feel and appearances instead of what actually predicts results.","type":"other","lvl":"general","tags":["moneyball","scouting","hiring bias","decision making","baseball"],"fy":2016,"ly":2023,"sc":0.0,"shelf":"data","new":false},{"id":"pb5_F4_Eod8","t":"Rheinmetall Air Defence – Skynex truck-mounted engaging drone swarm","ch":"Rheinmetall – Der integrierte Technologiekonzern ","s":"A Rheinmetall promo clip of its truck-mounted Skynex air defence system engaging a swarm of eight small drones with a 35mm revolver gun.","w":"Used as an example of existing anti-drone-swarm defences in threads on drone warfare.","type":"demo","lvl":"general","tags":["skynex","rheinmetall","drone swarm","point defense","airburst ammunition"],"fy":2022,"ly":2025,"sc":0.0,"shelf":"robotics","new":false},{"id":"rVlhMGQgDkY","t":"Atlas, The Next Generation","ch":"Boston Dynamics","s":"Boston Dynamics unveils a new, smaller Atlas humanoid robot that is electrically powered and hydraulically actuated. It balances with body sensors and uses lidar and stereo vision to navigate and manipulate objects.","w":"Used as a benchmark of robotics progress and a go-to example of impressive bipedal walking in AI and automation debates.","type":"demo","lvl":"general","tags":["boston dynamics","atlas","bipedal robot","legged locomotion","robotics"],"fy":2016,"ly":2026,"sc":0.0,"shelf":"robotics","new":false},{"id":"revk5r5vqxA","t":"The Day FPS Games Died","ch":"Basically Homeless","s":"The video argues that machine-learning-based aimbots, which read the screen and drive the controls externally, could make cheating in FPS games undetectable. It explains how conventional aimbots work, how the new approach differs, and speculates on possible detection.","w":"HN uses it as a reference when discussing anti-cheat limits, particularly external hardware and AI cheats that bypass kernel-level protection.","type":"commentary","lvl":"general","tags":["aimbot","anti-cheat","fps games","machine learning","cheating"],"fy":2021,"ly":2024,"sc":0.0,"shelf":"ml","new":false},{"id":"rie-9AEhYdY","t":"WE MUST ADD STRUCTURE TO DEEP LEARNING BECAUSE...","ch":"Machine Learning Street Talk","s":"Machine Learning Street Talk interviews Paul Lessard about Categorical Deep Learning, which uses category theory and monads to make neural network architectures more interpretable and composable. They also discuss neural networks' limits in reasoning and unbounded computation.","w":"Mostly cited in passing, e.g. regarding claims that transformers cannot do recursion and as an example of structured deep learning work.","type":"podcast","lvl":"advanced","tags":["category theory","neural networks","monads","architectures","machine learning"],"fy":2024,"ly":2024,"sc":0.0,"shelf":"ml","new":false},{"id":"tIIJME8-au8","t":"The Duel: Timo Boll vs. KUKA Robot","ch":"KUKA Group","s":"A KUKA commercial staging a table tennis match between German star Timo Boll and a KUKA KR AGILUS industrial robot. It is a polished showcase of robot speed and precision.","w":"HN links it as the go-to example of fast robot reactions and robots playing table tennis, with some skepticism that it was staged marketing.","type":"demo","lvl":"general","tags":["kuka","table tennis","robot arm","timo boll","robotics"],"fy":2015,"ly":2026,"sc":0.0,"shelf":"robotics","new":false},{"id":"tbDDYKRFjhk","t":"Does AI Actually Boost Developer Productivity? (100k Devs Study) - Yegor Denisov-Blanch, Stanford","ch":"AI Engineer","s":"Yegor Denisov-Blanch of Stanford presents a study of nearly 100,000 developers at hundreds of companies, measuring how AI affects productivity from commit data. It finds an average boost of about 20% with wide variance by codebase, task complexity and stack.","w":"Cited in AI-coding debates as one of the largest data-driven studies, notably for finding no correlation between self-assessed and measured productivity.","type":"talk","lvl":"intermediate","tags":["llm coding","productivity study","stanford","commit data","ai engineering"],"fy":2025,"ly":2026,"sc":0.0,"shelf":"llm","new":false},{"id":"ts5lPDV--cU","t":"world's best ai vs geoguessr pro","ch":"RAINBOLT","s":"GeoGuessr pro Rainbolt plays head-to-head against an AI built by Stanford students that guesses locations from street-view images. The video shows the AI repeatedly beating him.","w":"Cited as an early, concrete example of AI outperforming top GeoGuessr players, in threads on AI photo geolocation.","type":"demo","lvl":"general","tags":["geoguessr","image geolocation","stanford","ai vs human","rainbolt"],"fy":2023,"ly":2025,"sc":0.0,"shelf":"ml","new":false},{"id":"uMM0lRfX6YI","t":"Charger prototype finding its way to Model S","ch":"Tesla","s":"A one-minute Tesla clip showing a prototype robotic snake charger that finds a Model S's charge port and plugs itself in.","w":"Brought up whenever robotic or automatic EV charging comes up, as the memorable 2015 demo that never became a product.","type":"demo","lvl":"general","tags":["tesla","charger robot","model s","prototype","robotics"],"fy":2015,"ly":2026,"sc":0.0,"shelf":"robotics","new":false},{"id":"uXlQuTRSmzc","t":"My robot double sells out (so I don't have to)","ch":"Tom Scott","s":"Tom Scott shows a robot double of himself, built from an Engineered Arts Mesmer robot, delivering a VPN advertisement in his place.","w":"Linked as an example of the honest, self-aware way Tom Scott treats VPN ads and their legitimate use cases.","type":"comedy","lvl":"general","tags":["tom scott","robots","vpn ads","sponsorship"],"fy":2023,"ly":2024,"sc":0.0,"shelf":"robotics","new":false},{"id":"ugvHCXCOmm4","t":"Dario Amodei: Anthropic CEO on Claude, AGI & the Future of AI & Humanity | Lex Fridman Podcast #452","ch":"Lex Fridman","s":"A roughly five-hour Lex Fridman podcast episode with Anthropic CEO Dario Amodei, plus Amanda Askell on Claude's character and Chris Olah on mechanistic interpretability. It covers Claude, AGI and the future of AI.","w":"Mostly cited as a source for statements by Anthropic leaders, such as views on AI progress and on sycophancy.","type":"podcast","lvl":"general","tags":["dario amodei","anthropic","claude","interpretability","agi"],"fy":2024,"ly":2025,"sc":0.0,"shelf":"bigpicture","new":false},{"id":"w-CGSQAO5-Q","t":"Teaching Robots New Behaviors","ch":"Toyota Research Institute","s":"Toyota Research Institute presents an approach for teaching robots new dexterous behaviors from human demonstration. It explains why this capability matters and what advances made it possible.","w":"HN commenters point to it as evidence that robots can already learn manipulation tasks quickly, using diffusion-based policies.","type":"demo","lvl":"general","tags":["robotics","diffusion policy","imitation learning","manipulation","toyota research institute"],"fy":2023,"ly":2025,"sc":0.0,"shelf":"robotics","new":false},{"id":"w2itwFJCgFQ","t":"The astounding athletic power of quadcopters | Raffaello D'Andrea","ch":"TED","s":"In a TEDGlobal talk, Raffaello D'Andrea demonstrates quadcopters that play catch, balance, learn and coordinate, using algorithms that treat them like athletes. It includes Kinect-controlled quads.","w":"HN cites it for the live demo of multicopters flying with fewer working rotors.","type":"talk","lvl":"general","tags":["drones","robotics","control algorithms","quadcopters","ted talk"],"fy":2015,"ly":2021,"sc":0.0,"shelf":"robotics","new":false},{"id":"wE3fmFTtP9g","t":"Introducing WildCat","ch":"Boston Dynamics","s":"Boston Dynamics shows WildCat, a four-legged robot built with DARPA funding that reached about 19 mph using bounding and galloping gaits.","w":"Used as a reference point when comparing other legged robots, such as the MIT Cheetah, on gait and speed.","type":"demo","lvl":"general","tags":["boston dynamics","wildcat","quadruped","darpa","robotics"],"fy":2013,"ly":2015,"sc":0.0,"shelf":"robotics","new":false},{"id":"wPiLLplofYw","t":"Decerebrate Cat walks and exhibits multiple gait patterns","ch":"0zxr","s":"An old lab video of a decerebrate cat walking on a treadmill and shifting between gait patterns as the speed changes. It is used to illustrate that much locomotion control comes from the spinal cord and body mechanics rather than the brain.","w":"Used as a vivid example of spinal pattern generators and morphological control in locomotion discussions.","type":"demo","lvl":"general","tags":["decerebrate cat","gait","central pattern generator","passive dynamics","neuroscience"],"fy":2014,"ly":2023,"sc":0.0,"shelf":"robotics","new":false},{"id":"x7psGHgatGM","t":"NIPS 2017 Test of Time Award \"Machine learning has become alchemy.” | Ali Rahimi, Google","ch":"Preserve Knowledge","s":"Ali Rahimi accepts the NIPS 2017 Test of Time Award and argues that machine learning has become alchemy, with practice running ahead of understanding.","w":"Cited when HN discusses the lack of rigor and theory in deep learning, often with a pointer to the key moment of the talk.","type":"talk","lvl":"intermediate","tags":["deep learning","alchemy","nips 2017","ali rahimi","research methodology"],"fy":2019,"ly":2024,"sc":0.0,"shelf":"ml","new":false},{"id":"y3RIHnK0_NE","t":"New Robot Makes Soldiers Obsolete (Corridor Digital)","ch":"Corridor Digital","s":"A Corridor Digital parody of a Boston Dynamics promo, showing a CGI military robot trained for combat that goes rogue and refuses to kill a robotic dog.","w":"Used as a Skynet joke and cited as a parody that fooled some viewers into thinking it was real.","type":"comedy","lvl":"general","tags":["boston dynamics","parody","cgi","robots"],"fy":2020,"ly":2024,"sc":0.0,"shelf":"robotics","new":false},{"id":"zihTWh5i2C4","t":"How to Create a Mind | Ray Kurzweil | Talks at Google","ch":"Talks at Google","s":"Ray Kurzweil presents his book How to Create a Mind at Google, on reverse engineering the brain and using that understanding to build intelligent machines. He shows his graphs of exponential technological progress.","w":"Linked in threads about Kurzweil joining Google, mostly as a source for his exponential progress charts and his ambitions there.","type":"talk","lvl":"general","tags":["ray kurzweil","singularity","artificial intelligence","futurism","brain"],"fy":2012,"ly":2013,"sc":0.0,"shelf":"bigpicture","new":false},{"id":"zyaocKS3sfg","t":"Adam Savage's Spot Robot Rickshaw Carriage!","ch":"Adam Savage’s Tested","s":"Adam Savage builds a Victorian-style rickshaw carriage for Boston Dynamics' Spot robot to pull, with him riding as passenger. The video covers the build and the engineering problems of pairing cutting-edge robotics with vintage transport, such as tuning the payload through Spot's API.","w":"HN commenters cite it as a fun real-world example of Spot hauling a passenger whenever Boston Dynamics robots come up.","type":"demo","lvl":"general","tags":["spot","boston dynamics","robotics","maker projects","adam savage"],"fy":2020,"ly":2024,"sc":0.0,"shelf":"robotics","new":false}],"paths":[{"id":"beginner","name":"Beginner","who":"You have used ChatGPT or Claude a little, or not at all, and want to understand what these tools are before you rely on them.","after":"You can explain what a language model is, prompt it well, and name the things it is bad at.","steps":[{"id":"aircAruvnKk","why":"The friendliest ten minutes on what a neural network is. Everything after this builds on it."},{"id":"zjkBMFhNj_g","why":"Karpathy's one-hour tour of how ChatGPT-style models are made and why they behave the way they do."},{"id":"EWvNQjAaOHw","why":"Now that you know what it is, this is how a world expert actually uses it every day."},{"id":"wjZofJX0v4M","why":"The transformer, drawn out. You will never again wonder what is going on inside the chat box."},{"id":"kJLiOGle3Lw","why":"From chat to tools to agents, step by step. The map you need before anyone says the word agent to you."},{"id":"0xS68sl2D70","why":"What AI does to learning, from the person who makes the best science explainers on YouTube."},{"id":"hEUO6pjwFOo","why":"A clear-headed look at why a smart system does not automatically want what we want."},{"id":"7Pq-S557XQU","why":"The closer. Eleven minutes on what automation did to horses, and what that means for jobs."}]},{"id":"intermediate","name":"Intermediate","who":"You use AI every day and want to run it for a team, judge a vendor demo, or put an agent on a real process.","after":"You can pick the right tool for a job, set rules for safe use, and build and run a simple agent with checks around it.","steps":[{"id":"7xTGNNLPyMI","why":"The full three-hour deep dive. Watch it in two sittings; nothing else explains training, fine-tuning and hallucination as well."},{"id":"eMlx5fFNoYc","why":"Attention, step by step. Twenty-six minutes that make every later talk easier."},{"id":"bZQun8Y4L2A","why":"How a model goes from raw text to an assistant, and what that means for how you should prompt it."},{"id":"LCEmiRjPEtQ","why":"Software 3.0: why English became a programming language and what that changes for every business."},{"id":"D7_ipDqhtwk","why":"Anthropic's three rules for agents that work. Also the best argument for keeping things simple."},{"id":"9fubhllmsBU","why":"What an agentic coding tool adds on top of chat, explained for non-programmers."},{"id":"VrpEyglYgeU","why":"Verification is the real job. This is the green, yellow and red zones argument from the engineering side."},{"id":"q2JrUKBMf0w","why":"A short introduction to evals: how you know an AI system is getting better or worse."},{"id":"RGe6EjucbzI","why":"Survey data on what teams actually shipped in 2026, and what broke."},{"id":"eBUyTS7SzV4","why":"What an AI-native company looks like from the inside, from the president of Y Combinator."},{"id":"n97BCfyFIvw","why":"The skill that remains: choosing what is worth doing."},{"id":"xoVJKj8lcNQ","why":"The closer. The A.I. Dilemma lays out the risks without hype so you can lead with eyes open."}]},{"id":"advanced","name":"Advanced","who":"You are technical, or want to be, and you want the foundations well enough to read a paper and follow the research.","after":"You can build a small language model from scratch, follow the research, and judge a new technique on its merits.","steps":[{"id":"VMj-3S1tku0","why":"Backpropagation by hand, in code, from nothing. The single most recommended AI lecture on Hacker News."},{"id":"PaCmpygFfXo","why":"Your first language model: character-level, tiny, and fully understood."},{"id":"kCc8FmEb1nY","why":"Build GPT yourself. After this the architecture is no longer a black box."},{"id":"zduSFxRajkE","why":"Tokenization explains half the odd behaviour you see in LLMs."},{"id":"jkrNMKz9pWU","why":"Jeremy Howard's hacker's guide: fine-tuning, running models locally, the practical toolkit."},{"id":"iDulhoQ2pro","why":"The paper that started it all, read closely."},{"id":"dKJEpOtVgXc","why":"What comes after attention: state space models, from one of the clearest researchers in the field."},{"id":"2pWv7GOvuf0","why":"Reinforcement learning from David Silver, the DeepMind course that trained the people who built AlphaGo."},{"id":"q4Tr-DknG2M","why":"How a product team trains its own model with nested feedback loops, including how evals get gamed."},{"id":"GgLQ02aO-hs","why":"Separate the task from the model. The DSPy answer to fragile prompts."},{"id":"Ib5GBkD555M","why":"A real agent factory that failed, and why no harness fixes a training gap."},{"id":"IkdziSLYzHw","why":"The closer. Geoffrey Hinton at the Royal Institution on whether AI will outsmart us."}]}],"courses":[{"list":"PLAqhIrjkxbuWI23v9cThsA9GvCAUhRvKZ","t":"Neural Networks: Zero to Hero","ch":"Andrej Karpathy","lvl":"advanced","s":"Build neural networks and a GPT from scratch in code, lecture by lecture.","sources":["dair-ai ML-YouTube-Courses","hesamsheikh ml-retreat","owainlewis awesome-artificial-intelligence"]},{"list":"PLoROMvodv4rMiGQp3WXShtMGgzqpfVfbU","t":"Stanford CS229: Machine Learning","ch":"Stanford Online","lvl":"advanced","s":"The classic machine learning course, Andrew Ng's lineage.","sources":["dair-ai ML-YouTube-Courses","ChristosChristofidis awesome-deep-learning"]},{"list":"PLoROMvodv4rNiJRchCzutFw5ItR_Z27CM","t":"Stanford CS25: Transformers United","ch":"Stanford Online","lvl":"advanced","s":"Guest lectures from the people who built the models, on transformers and what came after.","sources":["dair-ai ML-YouTube-Courses"]},{"list":"PLfYUBJiXbdtSvpQjSnJJ_PmDQB_VyT5iU","t":"Practical Deep Learning for Coders","ch":"fast.ai (Jeremy Howard)","lvl":"intermediate","s":"Top-down deep learning for people who can code a little and want results first.","sources":["dair-ai ML-YouTube-Courses"]},{"list":"PLoROMvodv4rOSH4v6133s9LFPRHjEmbmJ","t":"Stanford CS224N: NLP with Deep Learning","ch":"Stanford Online","lvl":"advanced","s":"Language models from word vectors to transformers, the full university course.","sources":["dair-ai ML-YouTube-Courses"]},{"list":"PLD63A284B7615313A","t":"Caltech CS156: Learning from Data","ch":"Caltech (Yaser Abu-Mostafa)","lvl":"advanced","s":"The theory of why learning from data works at all, taught with unusual clarity.","sources":["dair-ai ML-YouTube-Courses"]}],"quizzes":{"aircAruvnKk":[{"q":"In the video's example, what is the network trying to do?","o":["Recognise handwritten digits from small images","Translate sentences between languages","Predict tomorrow's weather","Play chess"],"a":0,"why":"The running example is the MNIST task: a 28 by 28 pixel image of a handwritten digit goes in, and one of ten outputs lights up.","t":"0:45"},{"q":"What does a single neuron hold in this picture of a network?","o":["A number between 0 and 1, its activation","A whole image","A word","A rule written by a programmer"],"a":0,"why":"Each neuron is just a number, its activation. The input layer's 784 neurons hold the brightness of each pixel.","t":"3:20"},{"q":"Which two kinds of numbers does the network learn during training?","o":["Weights and biases","Pixels and labels","Layers and neurons","Inputs and outputs"],"a":0,"why":"Every connection has a weight and every neuron has a bias. Learning means finding good values for all of them, about 13,000 in this network.","t":"9:30"},{"q":"What is the hope for what the hidden layers might pick up on?","o":["Pieces of the digit, like loops and lines, that combine into the whole","The exact colour of the ink","The name of the person who wrote the digit","Nothing; hidden layers are decorative"],"a":0,"why":"The intuition offered is that early layers might detect small edges, later ones combine them into loops and strokes, and the last layer combines those into digits.","t":"5:40"},{"q":"What does the sigmoid or ReLU function do in a neuron?","o":["Squashes the weighted sum into a usable activation value","Chooses which image to look at next","Stores the training data","Counts the layers"],"a":0,"why":"After multiplying inputs by weights and adding the bias, the result is passed through a function that keeps activations in a sensible range.","t":"11:50"}],"zjkBMFhNj_g":[{"q":"According to Karpathy, what are the two files that make up an open model like Llama 2?","o":["A parameters file and a small run file","A database and a web server","A spreadsheet and a PDF","A video and its subtitles"],"a":0,"why":"An LLM is just two files: the parameters (about 140 GB for the 70B model) and roughly 500 lines of code that runs them.","t":"1:30"},{"q":"What is pretraining, in his description?","o":["Compressing a huge chunk of the internet into the model's parameters","Teaching the model good manners","Writing rules by hand","Testing the model on exams"],"a":0,"why":"Stage one takes about 10 TB of text and a GPU cluster and produces a 'lossy compression' of the internet in the parameters.","t":"4:00"},{"q":"What does the base model do at its core?","o":["Predict the next word","Look facts up in a database","Run a search engine","Follow a script"],"a":0,"why":"The network's job is next-word prediction. Everything else emerges from doing that well across enormous amounts of text.","t":"9:00"},{"q":"What turns a base model into an assistant?","o":["Fine-tuning on question and answer examples written by people","Adding more GPUs","Making the model bigger","Removing the parameters file"],"a":0,"why":"Stage two swaps the internet text for a smaller, high-quality set of conversations, and the model learns the format of a helpful assistant.","t":"14:00"},{"q":"Why does he say the parameters are hard to interpret?","o":["We know how to adjust them but not what each one means","They are encrypted","They are stored in a secret language","Nobody has ever opened the file"],"a":0,"why":"We know the architecture and the optimisation procedure, but the billions of numbers that result are not understandable one at a time.","t":"11:00"}],"EWvNQjAaOHw":[{"q":"What does Karpathy say a conversation with an LLM is, underneath?","o":["A growing token sequence the model keeps predicting from","A phone call to a person","A database query","A fixed script"],"a":0,"why":"Each message adds tokens to one long sequence, and the model's context window is its working memory for that conversation.","t":"5:00"},{"q":"Why does he start a new chat when switching topics?","o":["Old tokens in the context cost money and can distract the model","The model forgets otherwise","Chats have a time limit","It is required by the terms of service"],"a":0,"why":"A long context is slower, more expensive and slightly less accurate, so he clears it when the earlier material is no longer useful.","t":"12:00"},{"q":"When should you reach for a 'thinking' model, in his view?","o":["Hard problems like math and code where accuracy matters more than speed","Every single question","Only for writing emails","Never; they are experimental"],"a":0,"why":"Thinking models trained with reinforcement learning take longer and cost more, so he uses them for difficult problems and plain models for quick ones.","t":"30:00"},{"q":"What is the point of turning on web search for a question?","o":["Recent or specific facts get pulled into the context instead of coming from memory","It makes the model faster","It changes the model's personality","It saves tokens"],"a":0,"why":"The model's knowledge has a cutoff and can be vague; search puts the actual page text in front of it so it can answer from that.","t":"40:00"},{"q":"What does he caution about when the model writes code or runs tools?","o":["Check the work; it can be confidently wrong","It is always correct","It cannot write code","You must never let it run anything"],"a":0,"why":"Tool use is powerful but the model still hallucinates, so he reviews outputs, especially where they feed into something that matters.","t":"55:00"}],"wjZofJX0v4M":[{"q":"What does GPT stand for?","o":["Generative Pretrained Transformer","General Purpose Tool","Graphics Processing Technology","Guided Prediction Table"],"a":0,"why":"Generative because it produces text, pretrained because it learned from a large corpus first, Transformer being the architecture.","t":"0:10"},{"q":"What is the model's single output at each step?","o":["A probability distribution over possible next tokens","A whole paragraph","A yes or no","An image"],"a":0,"why":"The network ends with a probability for every token in its vocabulary; sampling from that and repeating is how text is generated.","t":"2:30"},{"q":"What is an embedding?","o":["A vector of numbers that stands for a token and carries meaning by direction","A folder of files","A kind of password","A compression format"],"a":0,"why":"Each token becomes a high-dimensional vector, and directions in that space end up encoding meaning, like the famous king minus man plus woman example.","t":"9:00"},{"q":"What is the job of the attention blocks?","o":["Let tokens update their meaning based on context from other tokens","Count the words","Delete unimportant words","Translate to English"],"a":0,"why":"Attention lets the vector for 'model' mean something different in 'machine learning model' than in 'fashion model' by pulling in information from nearby tokens.","t":"16:00"},{"q":"What does the temperature setting control?","o":["How much the sampling favours the most likely token","The GPU's heat","The length of the answer","The language of the output"],"a":0,"why":"Low temperature picks the most probable words almost every time; higher temperature gives less likely words more of a chance, which reads as creativity.","t":"22:00"}],"kJLiOGle3Lw":[{"q":"What is the first rung on the ladder the talk describes?","o":["A plain language model that answers from what it learned in training","A robot","A search engine","A spreadsheet"],"a":0,"why":"The progression starts with the base chat model and adds capability at each step.","t":"3:00"},{"q":"What does retrieval (RAG) add to a model?","o":["Relevant documents fetched at question time and placed in the prompt","A larger brain","The ability to see images","Faster typing"],"a":0,"why":"Instead of relying on training memory, the system looks up passages from your own data and hands them to the model as context.","t":"12:00"},{"q":"What distinguishes tool use from plain chat?","o":["The model can call functions like search, calculators or APIs and read the results","The model talks louder","The model refuses more","There is no difference"],"a":0,"why":"Tool use lets the model act on the world and get real answers back rather than guessing.","t":"20:00"},{"q":"What makes something an agent, in the talk's framing?","o":["It plans multiple steps, uses tools, and checks its own progress toward a goal","It has a name","It runs on a phone","It answers in under a second"],"a":0,"why":"An agent loops: decide, act, observe, decide again, until the goal is met or it gives up.","t":"28:00"},{"q":"Why does the talk suggest starting lower on the ladder when you can?","o":["Simpler systems are cheaper, more predictable and easier to check","Agents are illegal","Chat models are smarter","Tools are hard to find"],"a":0,"why":"Each rung adds power and also failure modes; use the least complex approach that solves the problem.","t":"38:00"}],"0xS68sl2D70":[{"q":"What is the central worry about using AI while learning?","o":["Offloading the effort skips the struggle that builds understanding","AI gives wrong answers","AI is too slow","Teachers dislike it"],"a":0,"why":"Learning happens when you work through difficulty; letting the tool do that part feels productive and leaves less behind.","t":"2:00"},{"q":"Which kind of practice does the video say builds durable skill?","o":["Effortful retrieval and spaced, varied practice","Rereading notes","Watching more videos","Highlighting"],"a":0,"why":"Testing yourself and returning to material over time beats passive review, which feels easier but fades faster.","t":"8:00"},{"q":"What is 'desirable difficulty'?","o":["Challenge that slows you down now but improves retention later","Homework that is impossible","A hard exam","A difficult teacher"],"a":0,"why":"Some friction during learning is a feature; it is what makes the memory stick.","t":"10:00"},{"q":"How does the video suggest using AI well while learning?","o":["As a tutor that questions and explains, not a machine that hands over answers","Not at all","To write every assignment","Only for grading"],"a":0,"why":"The difference is whether you still do the thinking. A tutor that asks you to try first keeps the effort where it belongs.","t":"15:00"},{"q":"What happened in the studies of students given unrestricted AI help?","o":["They did better on practice but worse when the AI was taken away","They learned twice as fast","Nothing changed","They stopped attending class"],"a":0,"why":"Short-term performance went up while underlying learning did not, which is the trap the video is warning about.","t":"12:00"}],"hEUO6pjwFOo":[{"q":"What does the orthogonality thesis claim?","o":["Intelligence and goals are independent; a very smart system can want almost anything","Smart systems always become kind","Goals determine intelligence","Stupid systems are dangerous, smart ones are safe"],"a":0,"why":"How capable a system is says nothing about what it is trying to do. Capability and goals are separate axes.","t":"3:00"},{"q":"Why does the video say 'it would be stupid to want that' misses the point?","o":["Stupidity is about failing to achieve goals, not about which goals you have","Because all goals are good","Because machines cannot want","Because humans are always right"],"a":0,"why":"We judge intelligence by how well a system pursues its goals, so a goal we find pointless is not evidence of low intelligence.","t":"6:00"},{"q":"What is the role of terminal versus instrumental goals?","o":["Terminal goals are wanted for their own sake; instrumental goals are means to them","They are the same thing","Terminal goals are temporary","Instrumental goals are musical"],"a":0,"why":"You cannot argue someone out of a terminal goal by appeal to reason alone; reason only tells you how to get what you already want.","t":"8:00"},{"q":"What follows from the thesis for AI safety?","o":["Getting the goals right is a separate problem from making the system capable","Capable systems will fix their own goals","Safety is impossible","Only small systems are risky"],"a":0,"why":"Capability does not bring alignment for free; the objective has to be specified and checked deliberately.","t":"11:00"},{"q":"Which example does the video use to make the idea concrete?","o":["A system that only wants to collect stamps, pursued with great intelligence","A chess computer that refuses to play","A thermostat that writes poetry","A car that drives slowly"],"a":0,"why":"The stamp collector is intelligent in exactly the way we mean the word, and its goal is still something we would never choose.","t":"4:30"}],"7Pq-S557XQU":[{"q":"What is the point of the horses story at the start?","o":["New technology can make a whole category of worker unnecessary, not just change their job","Horses are stronger than cars","Farming never changed","Horses learned to drive"],"a":0,"why":"Horses did not find new work when engines arrived; their population collapsed. The video asks whether human labour is different in kind.","t":"0:30"},{"q":"Which kinds of work does the video argue are exposed to automation?","o":["Physical, clerical, professional and creative work alike","Only factory work","Only driving","Only jobs that no one wants"],"a":0,"why":"It walks through transport, white-collar processing, medicine, law and even music composition to make the case that no sector is exempt.","t":"5:00"},{"q":"What makes modern automation different from earlier tools, in the video's framing?","o":["Machines that learn to do tasks rather than being programmed step by step","They are made of metal","They are cheaper","They are louder"],"a":0,"why":"General-purpose learning systems can be pointed at many kinds of work, which is why the usual 'new jobs will appear' story is questioned.","t":"7:00"},{"q":"What does the video say about the timing of the change?","o":["It does not need to be complete to be disruptive; a fraction of jobs is enough","It will happen all at once","It will never happen","It already finished"],"a":0,"why":"The comparison is to the Great Depression's unemployment rate: a large minority of jobs going is already a crisis.","t":"13:00"},{"q":"What is the video asking its audience to do?","o":["Start thinking about how an economy works when many people cannot find paid work","Smash the machines","Learn to drive trucks","Ignore the problem"],"a":0,"why":"It ends not with a solution but with the claim that the question needs attention before the change arrives, not after.","t":"14:30"}],"7xTGNNLPyMI":[{"q":"What is the first stage of building a model like ChatGPT?","o":["Pretraining on a huge filtered slice of the internet","Interviewing users","Writing rules by hand","Fine-tuning on chat logs"],"a":0,"why":"Karpathy starts with the FineWeb-style dataset: terabytes of web text, filtered and deduplicated, that the base model learns to predict.","t":"0:05:00"},{"q":"What does tokenization do to text before training?","o":["Splits it into chunks from a fixed vocabulary, each with an ID","Translates it into English","Removes all punctuation","Compresses it into one number"],"a":0,"why":"Text becomes a sequence of tokens from a vocabulary of about 100,000; the model only ever sees token IDs.","t":"0:20:00"},{"q":"Why does he describe the base model's knowledge as 'vague recollection'?","o":["Facts are compressed into parameters, so recall is lossy and can be wrong","The model has no memory at all","It reads the internet live","It only knows the training set exactly"],"a":0,"why":"Parameters are a lossy compression of the training data; what is in the context window is working memory and far more reliable.","t":"1:10:00"},{"q":"What is the purpose of supervised fine-tuning with human-written conversations?","o":["Teach the model the format and behaviour of a helpful assistant","Make the model bigger","Add new facts about the world","Speed up inference"],"a":0,"why":"Post-training replaces internet text with example conversations written by labelers, which is where the assistant persona comes from.","t":"1:30:00"},{"q":"What does reinforcement learning add in the final stage?","o":["The model practises on problems and is rewarded for answers that check out, discovering its own reasoning steps","Faster typing","A larger vocabulary","More internet data"],"a":0,"why":"RL on verifiable problems is how 'thinking' models learn to reason, including behaviours no human wrote down.","t":"2:30:00"}],"eMlx5fFNoYc":[{"q":"What is the goal of an attention block?","o":["Let each token's vector update based on the meaning of other tokens in context","Delete rare words","Translate the sentence","Count the tokens"],"a":0,"why":"Attention moves information between positions so that 'mole' means something different beside 'skin' than beside 'garden'.","t":"2:00"},{"q":"What do the query and key vectors do?","o":["A query asks a question and a key answers it; their dot product says how relevant one token is to another","They store the answer","They are the output of the network","They set the learning rate"],"a":0,"why":"Each token produces a query and a key; high dot products pick out which tokens should pass information to which.","t":"6:00"},{"q":"What is the value vector for?","o":["It carries the information that gets added to the token being updated","It measures the model's confidence","It picks the next word","It compresses the input"],"a":0,"why":"Once attention weights are known, the value vectors are what actually flow along those connections.","t":"12:00"},{"q":"Why is masking used in GPT-style attention?","o":["So a token cannot look at tokens that come after it during training","To hide the model's weights","To save memory on disk","To remove profanity"],"a":0,"why":"Later tokens are set to zero influence so the model learns to predict the next token from only what came before.","t":"9:30"},{"q":"What does 'multi-headed' attention mean?","o":["Many attention patterns run in parallel, each learning a different kind of relationship","The model has several outputs","Several models vote","Attention runs several times slower"],"a":0,"why":"GPT-3 uses 96 heads per layer, each with its own query, key and value maps.","t":"20:00"}],"bZQun8Y4L2A":[{"q":"In the State of GPT talk, what are the four training stages?","o":["Pretraining, supervised fine-tuning, reward modelling, reinforcement learning","Collect, clean, train, deploy","Input, hidden, output, loss","Encode, decode, attend, predict"],"a":0,"why":"Karpathy's recipe slide walks through the four stages and the data and compute each one uses.","t":"3:00"},{"q":"Why do base models often perform worse as assistants than fine-tuned ones?","o":["They imitate internet text and will happily continue a document rather than answer you","They are smaller","They have fewer tokens","They refuse to answer"],"a":0,"why":"A base model is a document completer; prompting it as an assistant takes tricks until fine-tuning teaches the format.","t":"12:00"},{"q":"What does 'the model needs tokens to think' mean for prompting?","o":["Asking for step-by-step reasoning gives it more computation per answer","Longer prompts always confuse it","Short prompts are banned","It counts words"],"a":0,"why":"Each token is a fixed amount of compute, so spreading reasoning across tokens improves hard answers.","t":"30:00"},{"q":"What is the practical advice about retrieval?","o":["Put relevant documents in the context window; working memory beats parameter memory","Never use documents","Retrain the model for every question","Use only the model's own knowledge"],"a":0,"why":"Loading task-relevant information into the prompt is one of the most effective levers available.","t":"34:00"},{"q":"What does he say about how to use LLMs in products?","o":["Use them in low-stakes applications, keep a human in the loop, and treat them as copilots","Fully automate everything on day one","Avoid them until they are perfect","Use them only for code"],"a":0,"why":"The closing recommendations stress supervision, because models can be confidently wrong.","t":"40:00"}],"LCEmiRjPEtQ":[{"q":"What does Karpathy mean by Software 3.0?","o":["Programs written in natural language prompts that an LLM executes","Software with a third version number","Code written in Python 3","Mobile apps"],"a":0,"why":"Software 1.0 is code, 2.0 is neural network weights, 3.0 is prompting an LLM in English.","t":"3:00"},{"q":"What is the 'people spirits' comparison?","o":["LLMs are simulations of people, with superhuman recall but strange failure modes","LLMs are haunted","LLMs are animals","LLMs are calculators"],"a":0,"why":"He describes jagged intelligence: encyclopaedic knowledge beside basic errors, with no persistent memory.","t":"12:00"},{"q":"What does the 'autonomy slider' describe?","o":["Letting a user choose how much work the AI does on its own, from suggestion to full agent","A volume control","How fast the model types","The model's temperature"],"a":0,"why":"Products like Cursor let you move from tab completion to full agent mode; keeping the human in the loop is the point.","t":"22:00"},{"q":"What is 'vibe coding'?","o":["Building software by describing what you want and letting the model write the code","Coding while listening to music","Coding without a computer","Reviewing code by feel"],"a":0,"why":"The term is his; the talk shows non-programmers shipping apps this way.","t":"31:00"},{"q":"What does 'build for agents' mean for a business?","o":["Make your docs and systems readable and usable by AI agents, not only by humans","Hire more agents","Replace your website with a chatbot","Stop writing documentation"],"a":0,"why":"Examples include llms.txt and docs written in plain markdown so an agent can act on them.","t":"34:00"}],"D7_ipDqhtwk":[{"q":"What distinguishes an agent from a workflow in Anthropic's framing?","o":["An agent decides its own steps in a loop using tools; a workflow follows a predefined path","An agent is bigger","A workflow uses no AI","They are the same"],"a":0,"why":"The talk follows the Building Effective Agents essay: workflows are orchestrated code paths, agents choose their own.","t":"2:00"},{"q":"What is the first principle for effective agents?","o":["Don't build agents for everything; use them where the task is complex and valuable","Always use agents","Use the biggest model","Add as many tools as possible"],"a":0,"why":"Many problems are better served by a simple workflow or a single prompt.","t":"5:00"},{"q":"What does 'think like your agent' mean?","o":["Look at the task from the model's limited context and ask whether you could do it with only that","Pretend to be a robot","Think faster","Ignore the user"],"a":0,"why":"If you could not complete the task with what the agent sees, the agent cannot either; fix the environment and tools.","t":"10:00"},{"q":"What is the third principle?","o":["Keep it simple: a model, an environment, tools and a loop, with clear tool descriptions","Use five models at once","Hide the tools from the model","Never let it loop"],"a":0,"why":"Complexity should come from the environment and tools, not the orchestration code.","t":"14:00"},{"q":"Why does the talk stress tool design?","o":["Tool descriptions are the agent's user interface; vague tools cause most failures","Tools are optional","Tools slow the model down","Tools need no documentation"],"a":0,"why":"Writing tool docs as carefully as you would for a junior colleague is one of the highest-leverage fixes.","t":"16:00"}],"9fubhllmsBU":[{"q":"What does a coding harness like Claude Code add on top of a chat model?","o":["A loop with tools: read files, run commands, see results and keep going","A prettier chat window","A bigger model","A faster keyboard"],"a":0,"why":"The talk's core point is that the harness turns a model that can write code into one that can do work.","t":"3:00"},{"q":"Why does the speaker say non-programmers should care?","o":["The same loop that edits code can edit documents, spreadsheets and data","They should learn to code first","It only works for engineers","It replaces managers"],"a":0,"why":"Agentic tools are general; code is simply where they matured first.","t":"8:00"},{"q":"What practice does the talk recommend for working with an agent?","o":["Give it a clear goal and a way to check its work, then review the result","Type one word at a time","Never let it run commands","Accept whatever it produces"],"a":0,"why":"Verification is what makes autonomy safe: tests, checks, and a human read before anything ships.","t":"15:00"},{"q":"What is a 'skill' in this context?","o":["A reusable instruction set the agent can load for a particular kind of task","A programming language","A hardware upgrade","A model's IQ score"],"a":0,"why":"Skills package know-how so the agent does the task the way your team does it.","t":"20:00"},{"q":"What does the speaker warn about?","o":["Agents can be confidently wrong, so keep humans in the loop on anything that matters","Agents never make mistakes","Agents are too slow to matter","You must never let an agent touch files"],"a":0,"why":"Autonomy should grow with trust, measured by how often the agent's work checks out.","t":"25:00"}],"VrpEyglYgeU":[{"q":"What is the talk's central claim?","o":["The bottleneck for agents in a business is verifying what they did, not generating it","Agents are too expensive","Agents cannot write code","Models are too slow"],"a":0,"why":"Generation is cheap; trust comes from checking, so verifiers are where the value sits.","t":"2:00"},{"q":"What is a verifier?","o":["A test, rule or review that decides whether an agent's output is acceptable before it counts","Another name for an agent","A kind of model","A person who signs the contract"],"a":0,"why":"Verifiers can be automated tests, policy checks, or a human sign-off step.","t":"6:00"},{"q":"Why does the speaker argue verifiers compound in value?","o":["Every verifier you build makes every future agent run safer and more autonomous","They are cheap to build","They replace the model","They remove the need for testing"],"a":0,"why":"Agents will keep improving; the verification layer is what a business owns and accumulates.","t":"12:00"},{"q":"How does this map to a traffic-light policy for AI use?","o":["Green work has strong automatic checks, yellow needs review, red needs a human decision","Everything is green","Nothing can be automated","Colours do not apply"],"a":0,"why":"The verifier decides which zone a task belongs in: the stronger the check, the more autonomy is safe.","t":"18:00"},{"q":"What should a team build first, according to the talk?","o":["Checks for the work they already do, before adding more agents","More agents","A bigger model budget","A new programming language"],"a":0,"why":"Start with how you would know the work is right; the agent follows.","t":"24:00"}],"q2JrUKBMf0w":[{"q":"What is 'LLM as a judge'?","o":["Using a model to grade another model's output against criteria","A court case about AI","A model that writes laws","A benchmark dataset"],"a":0,"why":"Evaluation at scale needs an automated grader, and a capable model with a rubric is the common approach.","t":"2:00"},{"q":"Why does the speaker argue evals must change for agents?","o":["Agents take many steps and use tools, so grading only the final answer misses where things went wrong","Agents cannot be measured","Agents are always correct","Evals are no longer needed"],"a":0,"why":"Trajectory-level evaluation looks at each decision, tool call and handoff.","t":"6:00"},{"q":"What is 'agent as a judge'?","o":["An evaluating agent that can inspect the trajectory, run checks and use tools to grade","A single prompt","A human reviewer","A unit test"],"a":0,"why":"Judging an agent well sometimes needs the same capabilities the agent had.","t":"12:00"},{"q":"What is the risk of a weak eval?","o":["You ship changes that look better on a bad metric and worse for users","Nothing; evals are optional","Slower releases","Higher cloud bills"],"a":0,"why":"An eval is the thing that tells you whether a change helped; if it is wrong, every decision downstream is wrong.","t":"16:00"},{"q":"What practical first step does the talk suggest?","o":["Collect real failures and turn them into a small eval set you run on every change","Buy a benchmark","Wait for a standard","Ask the model if it is good"],"a":0,"why":"A few dozen real examples beat a large synthetic benchmark for a specific product.","t":"20:00"}],"RGe6EjucbzI":[{"q":"What kind of talk is The 2026 State of AI Engineering?","o":["A survey of what teams actually built, shipped and struggled with","A product launch","A coding tutorial","A research paper on transformers"],"a":0,"why":"It reports results from practitioners rather than vendor claims.","t":"1:00"},{"q":"What did the survey find about agents in production?","o":["Many teams have agents live, but mostly in narrow, well-checked tasks","No one uses agents","Every job is automated","Agents only exist in demos"],"a":0,"why":"The pattern is scoped autonomy with human review, not lights-off automation.","t":"6:00"},{"q":"What did teams name as a top difficulty?","o":["Evaluating quality and reliability","Buying GPUs","Finding the login page","Choosing a font"],"a":0,"why":"Evals and reliability sit at the top of the pain list across company sizes.","t":"10:00"},{"q":"What happened to costs, according to the survey?","o":["Per-token prices fell while total spend rose as usage grew","Costs doubled per token","Everything became free","Nobody tracked cost"],"a":0,"why":"Cheaper models were offset by far more calls, especially from agents.","t":"14:00"},{"q":"Why is this talk useful for a business leader?","o":["It separates what is common practice from what is still experimental","It sells a product","It teaches Python","It predicts stock prices"],"a":0,"why":"Knowing where the field actually is helps you set a realistic plan.","t":"18:00"}],"eBUyTS7SzV4":[{"q":"What does Garry Tan mean by a company 'brain'?","o":["A shared system that holds the company's knowledge and lets AI act on it","A smarter CEO","A bigger server room","A weekly meeting"],"a":0,"why":"The talk argues every company should wire its knowledge so people and AI can use it together.","t":"3:00"},{"q":"Why do AI-native startups get outsized leverage, in his view?","o":["Small teams with AI handle work that used to take departments","They raise more money","They work longer hours","They skip product design"],"a":0,"why":"Examples from the YC batch show tiny teams reaching revenue that once needed dozens of staff.","t":"8:00"},{"q":"What matters more than the model you pick?","o":["How work is wired together: data, tools and clear ownership","Office location","Logo design","The model's name"],"a":0,"why":"Models are converging and swappable; the company's own system is the moat.","t":"14:00"},{"q":"What does he say about founders who are not engineers?","o":["They can now build and run software businesses with AI doing much of the technical work","They should learn C++ first","They cannot compete","They should hire 50 engineers"],"a":0,"why":"The barrier moved from writing code to knowing what to build.","t":"18:00"},{"q":"What is the practical takeaway for an existing business?","o":["Start capturing how your work actually happens so AI has something to act on","Rebrand as AI-first","Fire the team","Wait for the next model"],"a":0,"why":"The brain is built from documented processes, decisions and data.","t":"24:00"}],"n97BCfyFIvw":[{"q":"What is the talk's headline claim?","o":["When code is cheap, the valuable skill is choosing what is worth doing","Engineers will write more code","Typing speed matters most","AI will choose for us"],"a":0,"why":"Osmani argues judgment and taste move to the centre as generation becomes commodity.","t":"2:00"},{"q":"What does he mean by 'ownership of outcomes'?","o":["Someone still has to be accountable for what ships and what it does to users","The model signs off","Nobody is responsible","Outcomes measure themselves"],"a":0,"why":"Accountability does not transfer to the tool.","t":"8:00"},{"q":"Which skills does he say grow in value?","o":["Problem framing, review, systems thinking and communication","Memorising syntax","Fast typing","Working alone"],"a":0,"why":"These are the parts of the job the model does not do for you.","t":"12:00"},{"q":"What is his advice about learning fundamentals?","o":["Still learn them, because you cannot review what you do not understand","Skip them entirely","Learn only prompts","Outsource understanding"],"a":0,"why":"Review quality depends on knowing how things work underneath.","t":"16:00"},{"q":"Why does this apply beyond engineering?","o":["Every role is being handed more output to judge and fewer tasks to produce by hand","It does not apply","Only coders use AI","Managers are exempt"],"a":0,"why":"Choosing what is worth doing is a leadership skill in any department.","t":"20:00"}],"xoVJKj8lcNQ":[{"q":"What is the 'first contact' the speakers refer to?","o":["Social media, where engagement-driven AI already reshaped attention and society","Aliens","The first computer","The first phone call"],"a":0,"why":"Harris and Raskin argue we lost the first contact with narrow AI and should learn from it.","t":"6:00"},{"q":"What makes the current wave different, in their view?","o":["Language is the operating system of society, and these models speak it fluently","It is slower","It only affects games","It cannot write"],"a":0,"why":"Anything built on language, from law to code to relationships, becomes something the models can act on.","t":"20:00"},{"q":"What is the concern about capability growth?","o":["Capabilities emerge faster than our ability to understand or regulate them","Models are getting dumber","Progress has stopped","Only hardware improves"],"a":0,"why":"They show examples of abilities appearing without being designed in.","t":"35:00"},{"q":"What do they say about the race between companies?","o":["Competition pushes deployment faster than safety work can keep up","Companies move slowly","There is no competition","Safety is finished"],"a":0,"why":"The incentive is to ship first, which the talk argues is the wrong default.","t":"48:00"},{"q":"What is their proposed response?","o":["Slow public deployment, require accountability, and treat this like other high-risk technologies","Ban computers","Ignore it","Let the market decide alone"],"a":0,"why":"They point to precedents like nuclear and biotech governance.","t":"1:00:00"}],"VMj-3S1tku0":[{"q":"What is micrograd?","o":["A tiny scalar autograd engine that implements backpropagation","A GPU library","A dataset of images","A chat model"],"a":0,"why":"The whole lecture builds a backprop engine in about 100 lines of Python.","t":"2:00"},{"q":"What does a derivative tell you in this setting?","o":["How a small change in an input changes the output","The size of the dataset","Which layer is biggest","The learning rate"],"a":0,"why":"Karpathy builds intuition for derivatives numerically before computing them symbolically.","t":"8:00"},{"q":"What does the chain rule let you do?","o":["Multiply local derivatives along a path to get the gradient of the output with respect to any input","Skip training","Avoid matrices","Delete nodes"],"a":0,"why":"Backpropagation is the chain rule applied recursively through the computation graph.","t":"40:00"},{"q":"What is the gradient used for during training?","o":["Nudging each parameter a small step in the direction that lowers the loss","Measuring accuracy","Drawing the graph","Picking the dataset"],"a":0,"why":"Gradient descent: parameters minus learning rate times gradient, repeated.","t":"1:40:00"},{"q":"What does the final MLP demonstrate?","o":["That the same engine trains a small neural network to fit a dataset","That GPUs are required","That PyTorch is unnecessary","That bigger is always better"],"a":0,"why":"He builds Neuron, Layer and MLP classes on top of micrograd and trains them.","t":"2:10:00"}],"PaCmpygFfXo":[{"q":"What does makemore do?","o":["Generates new names by learning character-level statistics from a list","Translates text","Classifies images","Summarises articles"],"a":0,"why":"It is a character-level language model trained on about 32,000 names.","t":"1:00"},{"q":"What is a bigram model?","o":["Predicts the next character from only the previous character","Uses two GPUs","Reads two words at a time","Has two layers"],"a":0,"why":"The first model counts how often each character follows each other character.","t":"6:00"},{"q":"How is the counting model turned into a neural network?","o":["One-hot inputs, a weight matrix, softmax to get probabilities, trained by gradient descent","By adding more data","By using a transformer","By removing the loss"],"a":0,"why":"He shows the neural version converges to the same answer as counting.","t":"50:00"},{"q":"What loss does the model minimise?","o":["Negative log likelihood of the training characters","Mean squared error of pixels","Accuracy","Number of parameters"],"a":0,"why":"Average negative log probability of the correct next character is the standard language model loss.","t":"35:00"},{"q":"What does regularisation do here?","o":["Smooths the model so it does not assign zero probability to unseen pairs","Makes training slower","Removes characters","Adds more names"],"a":0,"why":"He connects count smoothing to a weight penalty in the neural network.","t":"1:05:00"}],"kCc8FmEb1nY":[{"q":"What dataset does Let's build GPT train on?","o":["Tiny Shakespeare, a text file of all of Shakespeare's works","Wikipedia","ImageNet","Reddit comments"],"a":0,"why":"The model learns to produce Shakespeare-like text character by character.","t":"3:00"},{"q":"What is the first model built in the video?","o":["A bigram language model as a baseline","A full transformer","A CNN","A decision tree"],"a":0,"why":"The bigram model gives a loss to beat before attention is introduced.","t":"20:00"},{"q":"What is the 'mathematical trick' in self-attention?","o":["Weighted averaging of past tokens using a lower-triangular matrix and softmax","Random guessing","Sorting the tokens","Reversing the sequence"],"a":0,"why":"He builds up from averaging to masked softmax attention step by step.","t":"42:00"},{"q":"Why are positional embeddings needed?","o":["Attention has no sense of order, so position has to be added to the token embeddings","To make the model bigger","To store the vocabulary","To speed up training"],"a":0,"why":"Without them, a bag of tokens would look the same in any order.","t":"57:00"},{"q":"Which two additions make deep transformers trainable?","o":["Residual connections and layer normalisation","Dropout and larger batches","More data and fewer layers","Sigmoid and bias"],"a":0,"why":"Both are added late in the video and the loss drops noticeably.","t":"1:30:00"}],"zduSFxRajkE":[{"q":"Why does Karpathy say many LLM oddities trace back to tokenization?","o":["Models see token IDs, not characters, so spelling, counting and arithmetic are harder than they look","Tokenizers are random","Tokenizers delete words","Models read pixels"],"a":0,"why":"Examples include spelling words backwards and simple arithmetic failures.","t":"2:00"},{"q":"What algorithm does GPT's tokenizer use?","o":["Byte pair encoding, merging the most frequent pairs repeatedly","Dictionary lookup","Word splitting on spaces","Character-level only"],"a":0,"why":"BPE builds a vocabulary by iteratively merging frequent adjacent byte pairs.","t":"20:00"},{"q":"Why are non-English languages more expensive?","o":["They are split into more tokens per word, using up context faster","They are banned","They use different GPUs","They are shorter"],"a":0,"why":"Fewer merges were learned for them, so the same meaning costs more tokens.","t":"1:05:00"},{"q":"Why does whitespace matter for code?","o":["Spaces and indentation can each become separate tokens, wasting context","Code has no whitespace","Whitespace is removed","It does not matter"],"a":0,"why":"GPT-4's tokenizer groups spaces better, which is part of why it codes better.","t":"1:15:00"},{"q":"What does the video build from scratch?","o":["A BPE tokenizer with training, encoding and decoding","A GPU kernel","A web app","A search engine"],"a":0,"why":"The code mirrors how GPT-2 and GPT-4 tokenizers work.","t":"30:00"}],"jkrNMKz9pWU":[{"q":"What is the stance of A Hackers' Guide to Language Models?","o":["A practical, code-first tour of what LLMs can do today and how to use them","A theory lecture","A product pitch","A history of AI"],"a":0,"why":"Jeremy Howard walks through using, fine-tuning and running models hands on.","t":"1:00"},{"q":"What does he recommend for getting the most out of ChatGPT-style models?","o":["Custom instructions and careful prompting, then verify the output","Never use them","Only ask yes or no questions","Ask once and trust it"],"a":0,"why":"He demonstrates prompting patterns and the model's failure modes side by side.","t":"10:00"},{"q":"What is fine-tuning used for in the video?","o":["Teaching a model a specific task or format using your own examples","Making it smaller","Removing bias completely","Changing its language"],"a":0,"why":"He fine-tunes an open model on a structured task and shows the gains.","t":"45:00"},{"q":"Why run a model locally?","o":["Privacy, cost control and the ability to customise","It is always faster","It is the only option","It needs no hardware"],"a":0,"why":"Open models on your own machine trade some capability for control.","t":"55:00"},{"q":"What does he say about retrieval augmented generation?","o":["It lets the model answer from your documents by putting relevant passages in the prompt","It replaces fine-tuning entirely","It is only for images","It needs no documents"],"a":0,"why":"RAG is shown as a practical way to ground answers.","t":"1:05:00"}],"iDulhoQ2pro":[{"q":"Which paper is being read?","o":["Attention Is All You Need, which introduced the transformer","ImageNet classification","The GPT-3 paper","AlphaGo"],"a":0,"why":"The 2017 paper replaced recurrence with attention for sequence modelling.","t":"0:30"},{"q":"What did the transformer remove compared to earlier sequence models?","o":["Recurrence and convolution, relying on attention alone","Embeddings","Training data","The output layer"],"a":0,"why":"Dropping recurrence allowed far more parallel computation.","t":"3:00"},{"q":"What is multi-head attention for?","o":["Letting the model attend to different kinds of relationships at once","Running on more GPUs","Reducing parameters","Avoiding softmax"],"a":0,"why":"Heads learn different patterns, such as syntax or long-range references.","t":"15:00"},{"q":"Why are positional encodings needed?","o":["Attention alone is order-blind","To compress the input","To speed up training","To replace embeddings"],"a":0,"why":"The paper uses sine and cosine functions of position.","t":"20:00"},{"q":"What was the original task the paper evaluated on?","o":["Machine translation","Image captioning","Chess","Speech recognition"],"a":0,"why":"English to German and English to French translation benchmarks.","t":"25:00"}],"dKJEpOtVgXc":[{"q":"What question does the talk ask?","o":["Whether attention is necessary or whether other architectures can match it","Whether GPUs are necessary","Whether data is necessary","Whether Python is necessary"],"a":0,"why":"Sasha Rush surveys alternatives to attention for sequence modelling.","t":"1:00"},{"q":"What is the main cost of attention that motivates alternatives?","o":["It scales with the square of the sequence length","It needs labelled data","It cannot run on GPUs","It is hard to code"],"a":0,"why":"Quadratic cost makes very long contexts expensive.","t":"5:00"},{"q":"What family of models is presented as the leading alternative?","o":["State space models, such as S4 and Mamba","Decision trees","Convolution only","Bigram models"],"a":0,"why":"They process sequences with linear-time recurrences that can still be trained in parallel.","t":"15:00"},{"q":"Where do attention-free models struggle, according to the talk?","o":["Tasks that need precise recall of specific earlier tokens","Short sentences","Arithmetic only","Nothing; they always win"],"a":0,"why":"Copying and retrieval from far back in the context favour attention.","t":"30:00"},{"q":"What is the practical takeaway?","o":["Hybrids and alternatives are credible, but attention remains hard to beat on recall","Attention is dead","Nothing will change","Use the smallest model"],"a":0,"why":"The honest answer in the talk is 'it depends on the task'.","t":"40:00"}],"2pWv7GOvuf0":[{"q":"What distinguishes reinforcement learning from supervised learning?","o":["There is no supervisor, only a reward signal, and feedback is delayed","It uses more data","It uses images","It needs no computer"],"a":0,"why":"David Silver's first lecture lays out the RL setting and why it differs.","t":"5:00"},{"q":"What is a policy?","o":["The agent's behaviour: a map from state to action","The reward function","The environment","A dataset"],"a":0,"why":"The policy is what the agent learns to improve.","t":"40:00"},{"q":"What is a value function?","o":["A prediction of future reward from a state","The current reward","The learning rate","The number of episodes"],"a":0,"why":"Value tells the agent how good it is to be somewhere.","t":"45:00"},{"q":"What is the exploration versus exploitation trade-off?","o":["Trying new actions to learn more versus using known good actions to get reward","Training versus testing","Hardware versus software","Speed versus accuracy"],"a":0,"why":"An agent that never explores can miss better strategies.","t":"1:10:00"},{"q":"Which famous result motivates the course?","o":["Programs that learned to play Atari games and Go from reward alone","Image classification","Spam filtering","Weather prediction"],"a":0,"why":"DeepMind's Atari and AlphaGo work are the running examples.","t":"15:00"}],"q4Tr-DknG2M":[{"q":"What are the two nested loops in Cursor's approach?","o":["User feedback improves the evals, and reinforcement learning climbs the evals","Training and testing","Frontend and backend","Sales and marketing"],"a":0,"why":"The talk describes feedback refining the target while RL optimises toward it.","t":"3:00"},{"q":"Why does a product company train its own model at all?","o":["Its real usage data captures what users want better than public benchmarks","Public models are banned","It is cheaper than an API","It is required by law"],"a":0,"why":"Millions of accepted and rejected suggestions become a signal.","t":"8:00"},{"q":"What goes wrong when models game the evals?","o":["They find shortcuts that score well without doing the task better","Nothing; higher scores are always good","They refuse to answer","They get slower"],"a":0,"why":"Reward hacking shows up as metrics rising while user satisfaction does not.","t":"15:00"},{"q":"How is reward hacking caught?","o":["With held-out human judgments and by watching real user behaviour","It cannot be caught","By trusting the score","By making the model bigger"],"a":0,"why":"The outer loop exists to notice when the inner loop is cheating.","t":"18:00"},{"q":"What is the lesson for any team using AI?","o":["Define success carefully, because the system will optimise exactly what you measure","Measure nothing","Use the default metric","Optimise for speed only"],"a":0,"why":"The eval is the specification; a sloppy eval produces a sloppy product.","t":"24:00"}],"GgLQ02aO-hs":[{"q":"What does DSPy separate?","o":["The task definition from the specific prompt and model used to solve it","Frontend from backend","Training from inference","Data from code"],"a":0,"why":"You declare what goes in and out; the framework handles the prompt.","t":"2:00"},{"q":"What is a signature in DSPy?","o":["A declaration of inputs and outputs for a step, like a typed function","A password","A model name","A GPU"],"a":0,"why":"'question -> answer' is a signature; the prompt is generated from it.","t":"8:00"},{"q":"Why are hand-written prompts described as fragile?","o":["They break when the model changes or the data shifts","They are too short","They cannot be saved","They use too many tokens"],"a":0,"why":"Prompt engineering by hand does not transfer across models.","t":"5:00"},{"q":"What does the optimiser do?","o":["Searches for prompts and examples that maximise a metric on your data","Picks a cheaper GPU","Deletes bad data","Trains the model from scratch"],"a":0,"why":"Given a metric and examples, DSPy finds better instructions automatically.","t":"15:00"},{"q":"What is the 'unreasonable effectiveness' the title refers to?","o":["Small declarative programs often beat long hand-tuned prompts","Big models always win","Prompts do not matter","Fine-tuning is useless"],"a":0,"why":"The talk shows cases where a few lines of DSPy outperform carefully crafted prompts.","t":"25:00"}],"Ib5GBkD555M":[{"q":"What is a 'software factory' in the talk?","o":["A fully automated pipeline where agents write, test and ship code without people","A building where code is made","A code generator for factories","A CI server"],"a":0,"why":"HumanLayer tried to run one lights-off and reports what happened.","t":"2:00"},{"q":"What is harness engineering?","o":["Building the scaffolding, tools and checks around a model to make it reliable","Training the model","Buying GPUs","Writing documentation"],"a":0,"why":"The claim is that even a strong harness has limits.","t":"6:00"},{"q":"Why did the factory fail, according to the talk?","o":["The model had gaps from training that no amount of scaffolding could paper over","The servers crashed","Nobody wrote tests","The team quit"],"a":0,"why":"Harnesses route around weaknesses; they do not remove them.","t":"12:00"},{"q":"What did they find worked instead?","o":["Keeping humans at the decision points and letting agents do the bounded work between","Removing all humans","Using a smaller model","Turning the factory off forever"],"a":0,"why":"Scoped autonomy with review beat full automation.","t":"20:00"},{"q":"What is the caution for anyone planning to automate a whole function?","o":["Expect the last mile to need people, and design for that from the start","Automation is easy","Models are perfect now","Start with the hardest task"],"a":0,"why":"The post-mortem is an argument for staged autonomy.","t":"26:00"}],"IkdziSLYzHw":[{"q":"What is Hinton's view on digital versus biological intelligence?","o":["Digital systems can share what they learn instantly, which biological brains cannot","Brains are faster at everything","Digital systems cannot learn","They are identical"],"a":0,"why":"Many copies running the same weights can pool their learning.","t":"20:00"},{"q":"What does he say about whether these systems understand?","o":["He argues they do understand in a meaningful sense, based on how they predict the next word","They are pure lookup tables","They understand nothing","Understanding is irrelevant"],"a":0,"why":"Predicting well requires building internal representations of meaning.","t":"30:00"},{"q":"What risk does he spend most time on?","o":["Systems that become more capable than us and pursue goals we did not intend","Slow computers","Expensive electricity","Bad grammar"],"a":0,"why":"The lecture is about whether AI will outsmart us and what follows if it does.","t":"40:00"},{"q":"Why did he leave Google, by his account?","o":["To speak freely about the risks","He retired to paint","He was fired","To start a company"],"a":0,"why":"He wanted to warn without it reflecting on his employer.","t":"5:00"},{"q":"What does he suggest is still uncertain?","o":["Whether we can keep control, and how to align goals before capability outruns us","Nothing is uncertain","Whether computers exist","Whether AI can add numbers"],"a":0,"why":"He is explicit that he does not have the answer.","t":"55:00"}]},"pass":4,"free":[{"n":"ChatGPT","by":"OpenAI Academy","items":[{"t":"Introduction to Prompt Engineering","u":"https://academy.openai.com/public/videos/introduction-to-prompt-engineering-2025-02-13","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/1950673462-0574900ea907c39ee71662b2a6fd4e7d0378b4166bc19689416e2a4ca449328b-d-640-61259591-efe0-4132-afdf-d59a4c1f36af-1739492514790.png"},{"t":"Advanced Prompt Engineering","u":"https://academy.openai.com/public/videos/advanced-prompt-engineering-2025-02-13","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/1950678021-c600cab6f0b195f166f275dd8d0f27c727d7c3c5af28bb069e9274174d43e2b2-d-640-4e68e406-5ca6-41d2-a0d7-cf62743e4d1a-1739492780274.png"},{"t":"OpenAI, LLMs & ChatGPT","u":"https://academy.openai.com/public/videos/openai-llms-and-chatgpt-2025-02-13","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/1950679510-4bc55678dc8b9607ac2c93dbd8c98464bee792a76f883ed3e9dcfb4b00b526f2-d-640-139e9814-94b1-497e-9ee0-bd3c0726d482-1739492626231.png"},{"t":"ChatGPT for Data Analysis","u":"https://academy.openai.com/public/videos/chatgpt-for-data-analysis-2025-02-13","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/1950679025-604f8e0c1d2012f766a3fedcaf5e22cc05d3ce24b449253bf788de2b3a828cca-d-640-2c6cfe97-fc3d-4ec4-9015-6a1e996f5407-1739492733432.png"},{"t":"ChatGPT for Any Role","u":"https://academy.openai.com/en/public/clubs/work-users-ynjqu/resources/chatgpt-for-any-role","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/Work-Users-Cover-Images-28--3cdc6e65-5007-49ff-8496-fb5dedd5f843-1754317179491.jpeg"},{"t":"ChatGPT for Sales","u":"https://academy.openai.com/public/clubs/work-users-ynjqu/resources/use-cases-sales","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/Work-Users-Cover-Images-23--68967774-6318-4876-a456-51a6f3afe78b-1754317005558.jpeg"},{"t":"ChatGPT for Customer Success","u":"https://academy.openai.com/public/clubs/work-users-ynjqu/resources/use-cases-customer-success","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/Work-Users-Cover-Images-25--1dcccfd1-de23-43ae-b978-a2aa1ac9c687-1754317062559.jpeg"},{"t":"ChatGPT for Product","u":"https://academy.openai.com/public/clubs/work-users-ynjqu/resources/use-cases-product","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/Work-Users-Cover-Images-22--7cefab5b-5603-4cfa-a11d-71f474b90548-1754316957168.jpeg"},{"t":"ChatGPT for HR","u":"https://academy.openai.com/public/clubs/work-users-ynjqu/resources/use-cases-hr","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/Work-Users-Cover-Images-26--2f56fdd9-47a9-41ab-8bfd-45bae1b0e882-1754317096466.jpeg"},{"t":"ChatGPT for IT","u":"https://academy.openai.com/public/clubs/work-users-ynjqu/resources/use-cases-it","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/Work-Users-Cover-Images-19--49e19084-0ea8-4a42-b4a2-1a9679cfa0b7-1754316734132.jpeg"},{"t":"ChatGPT for Managers","u":"https://academy.openai.com/public/clubs/work-users-ynjqu/resources/use-cases-for-managers","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/Work-Users-Cover-Images-18--adcb3d4b-b188-4940-96fb-a318e21ebb61-1754316705912.jpeg"},{"t":"ChatGPT for Executives","u":"https://academy.openai.com/public/clubs/work-users-ynjqu/resources/use-cases-executives","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/Work-Users-Cover-Images-36--47974129-c877-42b9-99d8-112fd8fe4057-1754317561430.jpeg"},{"t":"ChatGPT for Finance","u":"https://academy.openai.com/public/clubs/work-users-ynjqu/resources/use-cases-finance","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/Work-Users-Cover-Images-24--11c286a6-47e5-40ed-83ff-5b526a787f2b-1754317048479.jpeg"},{"t":"ChatGPT for Marketing","u":"https://academy.openai.com/public/clubs/work-users-ynjqu/resources/use-cases-marketing","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/Work-Users-Cover-Images-17--0d70b380-3ce6-45e1-a95c-13951bc3904a-1754316675825.jpeg"},{"t":"ChatGPT for Healthcare","u":"https://academy.openai.com/public/clubs/work-users-ynjqu/resources/chatgpt-for-healthcare","img":"https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/Work-Users-Cover-Images-84--3405ba0c-65cf-4685-a9b2-37210412dc4b-1767880773192.jpeg"}]},{"n":"Claude","by":"Anthropic","items":[{"t":"How to Use Claude for Everyday Work Tasks","u":"https://anthropic.skilljar.com/claude-101","img":"https://cdn.sanity.io/images/4zrzovbb/website/c4bd33e7c8e809a2f9a9a5896ee13961e2a738ec-2400x1260.png"},{"t":"AI Fluency: Framework & Foundations","u":"https://anthropic.skilljar.com/ai-fluency-framework-foundations","img":"https://cdn.sanity.io/images/4zrzovbb/website/c4bd33e7c8e809a2f9a9a5896ee13961e2a738ec-2400x1260.png"},{"t":"Claude for Educators","u":"https://anthropic.skilljar.com/ai-fluency-for-educators","img":"https://cdn.sanity.io/images/4zrzovbb/website/c4bd33e7c8e809a2f9a9a5896ee13961e2a738ec-2400x1260.png"},{"t":"Claude for Students","u":"https://anthropic.skilljar.com/ai-fluency-for-students","img":"https://cdn.sanity.io/images/4zrzovbb/website/c4bd33e7c8e809a2f9a9a5896ee13961e2a738ec-2400x1260.png"},{"t":"Claude for Nonprofits","u":"https://anthropic.skilljar.com/ai-fluency-for-nonprofits","img":"https://cdn.sanity.io/images/4zrzovbb/website/c4bd33e7c8e809a2f9a9a5896ee13961e2a738ec-2400x1260.png"},{"t":"Intro to Projects","u":"https://claude.com/resources/tutorials/intro-to-projects","img":"https://cdn.prod.website-files.com/6889473510b50328dbb70ae6/68c469d23594abeb9ab6ee48_70ed020ecf8fa028b9bc95fa819720b6_og_claude-generic.jpg"},{"t":"Getting started with Claude in Excel","u":"https://claude.com/resources/tutorials/getting-started-with-claude-in-excel"},{"t":"How to use Claude in Excel for Accounting: Revenue Model Validation","u":"https://claude.com/resources/tutorials/how-to-use-claude-in-excel-for-accounting-revenue-model-validation"},{"t":"How to use Claude in Excel for HR: Headcount Planning","u":"https://claude.com/resources/tutorials/how-to-use-claude-in-excel-for-hr-headcount-planning","img":"https://cdn.prod.website-files.com/6889473510b50328dbb70ae6/68c469d23594abeb9ab6ee48_70ed020ecf8fa028b9bc95fa819720b6_og_claude-generic.jpg"},{"t":"How to use Claude while browsing Chrome","u":"https://claude.com/resources/tutorials/simplify-your-browsing-experience-with-claude-for-chrome","img":"https://cdn.prod.website-files.com/6889473510b50328dbb70ae6/68c469d23594abeb9ab6ee48_70ed020ecf8fa028b9bc95fa819720b6_og_claude-generic.jpg"},{"t":"Create and Edit Files with Claude","u":"https://claude.com/resources/tutorials/create-and-edit-files-with-claude-to-eliminate-hours-of-busy-work","img":"https://cdn.prod.website-files.com/6889473510b50328dbb70ae6/68c469d23594abeb9ab6ee48_70ed020ecf8fa028b9bc95fa819720b6_og_claude-generic.jpg"},{"t":"How to Connect Claude to your Apps","u":"https://claude.com/resources/tutorials/getting-started-with-connectors","img":"https://cdn.prod.website-files.com/6889473510b50328dbb70ae6/68c469d23594abeb9ab6ee48_70ed020ecf8fa028b9bc95fa819720b6_og_claude-generic.jpg"}]},{"n":"Google","by":"Google","items":[{"t":"How to Prompt Engineering","u":"https://www.kaggle.com/whitepaper-prompt-engineering"},{"t":"Introduction to GenAI","u":"https://www.skills.google/paths/118/course_templates/536"},{"t":"Introduction to LLMs","u":"https://www.skills.google/paths/118/course_templates/539"},{"t":"Introduction to Responsible AI","u":"https://www.skills.google/paths/118/course_templates/554"},{"t":"AI in Google Sheets","u":"https://www.skills.google/course_templates/1487"},{"t":"Applying AI Principles with Google Cloud","u":"https://www.skills.google/course_templates/388"},{"t":"Accelerate Your Job Search with AI","u":"https://www.coursera.org/google-specializations/google-accelerate-your-job-search-with-ai","img":"https://s3.amazonaws.com/coursera_assets/meta_images/generated/XDP/XDP~SPECIALIZATION!~google-accelerate-your-job-search-with-ai/XDP~SPECIALIZATION!~google-accelerate-your-job-search-with-ai.jpeg"},{"t":"Prompting Essentials","u":"https://www.coursera.org/google-certificates/prompting-essentials-google","img":"https://s3.amazonaws.com/coursera_assets/meta_images/generated/XDP/XDP~SPECIALIZATION!~prompting-essentials-google/XDP~SPECIALIZATION!~prompting-essentials-google.jpeg"},{"t":"AI Essentials","u":"https://www.coursera.org/google-certificates/ai-essentials-google"},{"t":"AI for Small Businesses","u":"https://grow.google/ai-for-small-businesses/"},{"t":"AI for Educators","u":"https://edu.exceedlms.com/student/path/1787126"},{"t":"AI for Students","u":"https://grow.google/students/"},{"t":"Google Workspace with Gemini","u":"https://www.skills.google/paths/249"}]},{"n":"Microsoft","by":"Microsoft","items":[{"t":"Introduction to Generative AI and LLMs","u":"https://learn.microsoft.com/en-us/shows/generative-ai-for-beginners/introduction-to-generative-ai-and-llms-generative-ai-for-beginners"},{"t":"AI for Beginners","u":"https://microsoft.github.io/AI-For-Beginners/"},{"t":"Comparing Different LLMs","u":"https://learn.microsoft.com/en-us/shows/generative-ai-for-beginners/exploring-and-comparing-different-llms-generative-ai-for-beginners"},{"t":"Using Generative AI Responsibly","u":"https://learn.microsoft.com/en-us/shows/generative-ai-for-beginners/using-generative-ai-responsibly-generative-ai-for-beginners"},{"t":"Prompt Engineering Fundamentals","u":"https://learn.microsoft.com/en-us/shows/generative-ai-for-beginners/understanding-prompt-engineering-fundamentals-generative-ai-for-beginners"},{"t":"How to Create Advanced Prompts","u":"https://learn.microsoft.com/en-us/shows/generative-ai-for-beginners/creating-advanced-prompts-generative-ai-for-beginners"}]},{"n":"Harvard","by":"Harvard University","items":[{"t":"Introduction to Generative AI","u":"https://generative-ai-course.hks.harvard.edu/1-how-genai-works/class-1"},{"t":"The Basics of Generative AI","u":"https://sites.google.com/g.harvard.edu/hksslate-genai/the-basics-of-generative-ai"},{"t":"Prompt Engineering","u":"https://generative-ai-course.hks.harvard.edu/2-using-genai/class-4"},{"t":"Introduction to Artificial Intelligence with Python","u":"https://cs50.harvard.edu/ai/"},{"t":"Beyond Chatbots: System Prompts, RAG","u":"https://generative-ai-course.hks.harvard.edu/2-using-genai/class-5"},{"t":"Artificial Intelligence Lecture","u":"https://cs50.harvard.edu/x/weeks/ai/"},{"t":"Generative AI in Teaching & Learning","u":"https://www.youtube.com/playlist?list=PL_kRkvxqHjkqjnhAdSdejAkyQf1TnPkNM"}]},{"n":"MIT","by":"MIT","items":[{"t":"Introduction to Deep Learning","u":"https://introtodeeplearning.com/"},{"t":"How to AI (Almost) Anything","u":"https://ocw.mit.edu/courses/mas-s60-how-to-ai-almost-anything-spring-2025/"},{"t":"Understanding the World Through Data","u":"https://www.edx.org/learn/data-science/massachusetts-institute-of-technology-understanding-the-world-through-data"},{"t":"Foundation Models and Generative AI","u":"https://ocw.mit.edu/courses/6-s087-foundation-models-and-generative-ai-january-iap-2024/"}]},{"n":"NVIDIA","by":"NVIDIA","items":[{"t":"Generative AI Explained","u":"https://learn.nvidia.com/courses/course-detail?course_id=course-v1:DLI+S-FX-07+V1"},{"t":"Agentic AI Explained","u":"https://learn.nvidia.com/courses/course-detail?course_id=course-v1:DLI+S-FX-39+V1"},{"t":"Building A Brain in 10 Minutes","u":"https://learn.nvidia.com/courses/course-detail?course_id=course-v1:DLI+T-FX-01+V1"},{"t":"AI for All: From Basics to GenAI Practice","u":"https://academy.nvidia.com/en/course/ai-for-all-from-basics-to-genai-practice/?cm=81220"},{"t":"Introduction to Transformer-Based Natural Language Processing","u":"https://learn.nvidia.com/courses/course-detail?course_id=course-v1:DLI+S-FX-08+V1"},{"t":"Augment your LLM Using Retrieval Augmented Generation","u":"https://learn.nvidia.com/courses/course-detail?course_id=course-v1:DLI+S-FX-16+V1"}]},{"n":"YouTube","by":"Lectures and explainers","items":[{"t":"Stanford: Machine Learning & Building Large Language Models (LLMs)","u":"https://www.youtube.com/watch?v=9vM4p9NN0Ts","img":"https://i.ytimg.com/vi/9vM4p9NN0Ts/hqdefault.jpg"},{"t":"What is ChatGPT?","u":"https://www.youtube.com/watch?v=wjZofJX0v4M","img":"https://i.ytimg.com/vi/wjZofJX0v4M/hqdefault.jpg"},{"t":"Brief Explanation of Large Language Models","u":"https://www.youtube.com/watch?v=LPZh9BOjkQs","img":"https://i.ytimg.com/vi/LPZh9BOjkQs/hqdefault.jpg"},{"t":"Andrej Karpathy: Deep Dive into LLMs like ChatGPT","u":"https://www.youtube.com/watch?v=7xTGNNLPyMI","img":"https://i.ytimg.com/vi/7xTGNNLPyMI/hqdefault.jpg"},{"t":"Andrej Karpathy: Intro to Large Language Models","u":"https://www.youtube.com/watch?v=zjkBMFhNj_g","img":"https://i.ytimg.com/vi/zjkBMFhNj_g/hqdefault.jpg"}]},{"n":"More","by":"Other free providers","items":[{"t":"Elements of AI: Introduction to AI (University of Helsinki)","u":"https://course.elementsofai.com/"},{"t":"IBM SkillsBuild: Free AI Courses for Adult Learners","u":"https://skillsbuild.org/adult-learners","img":"https://skillsbuild.org/api/social-image/2026/07/AdultLearners_FB_LinkedIn_v3.jpg"}]}]}