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.
Topics › AI, Machine Learning & Data
The best machine learning & deep learning videos, according to Hacker News
130 videos about machine learning & deep learning that Hacker News commenters keep pointing each other to, ranked by how many separate months they were recommended. Part of AI, Machine Learning & Data.
HN treats it as the standard first stop for learning neural networks, praised for making the math intuitive.
HN commenters call it a moving, well-made human story about AI and Go that works for non-technical viewers and does not date.
HN treats it as the go-to visual introduction to transformers, and many say it finally made attention and embeddings click.
HN considers it a classic, hilarious and sobering talk on ML inscrutability and tech hubris, often rewatched.
Recommended as the rigorous, math-heavy original course, deeper than the Coursera version, for learning machine learning.
The go-to accessible introduction to neuroevolution and NEAT, praised for crediting prior research.
HN's most frequent recommendation for a clear, approachable foundation in how LLMs work without heavy math.
Loved on HN as a hilarious, brilliant example of hacker-academic work and cited whenever game-playing AI or fuzzing comes up.
Frequently recommended as an excellent, clear introductory RL course from the lead of AlphaGo, pairing well with Sutton and Barto.
It is the go-to recommendation for learning how transformers and LLMs actually work, praised for its clarity and Karpathy's teaching.
HN commenters recommended it for years as a simple, exciting overview of the deep learning wave.
Repeatedly named the best practical tutorial for using LLMs, with a well-liked fine-tuning section.
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.
HN recommends it as the dense, approachable starting point for understanding LLMs and often cites it for specific claims about scaling and training.
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.
Repeatedly called an excellent, practical introduction that covers a lot in two hours and explains issues academic textbooks skip.
Recommended as a simple, elegant explanation of dropout and its connection to how the brain works.
The go-to link in any anti-cheat thread for a concrete, well-delivered account of machine-learning cheat detection in production.
Recommended to people who find it inspiring or perspective-changing on pursuing goals, open-ended search and AI beyond optimization.
Recommended repeatedly in 'where do I start with ML' threads as a clear, math-aware introduction.
Treated as the accessible reference on analog computing for neural nets, especially the Mythic AI segment.
Tom7 is an HN favorite, and this video is praised as hilarious, educational silly ML that keeps getting better.
Recommended as the best deep dive on tokenizers, worth the two hours for anyone working with LLMs.
HN commenters recommend it as an approachable first step into machine learning for programmers, with an apples-vs-oranges classifier.
HN treats it as the go-to visual explanation of attention, recommended to people who find text-based transformer explainers hard to follow.
Recommended as a clear explanation of why RNNs fall short and how transformers solve it.
Commenters recommend it as an entertaining, wide-ranging interview, and cite it for Hotz's views on end-to-end neural-net driving.
HN recommends it as an accessible, clear walkthrough of the transformer paper for people with some AI background.
Cited as the more accessible companion to the paper, and as the most interesting treatment of AD via compiling to categories.
Recommended as the most accessible introduction to backprop for people who struggle with the pure math.
Seen as the clearest introduction to the idea behind RLHF, and also borrowed by commenters as a metaphor for human motivation.
HN points to it as the sharpest, most-quoted critique of deep learning practice, and cites it whenever ML reliability or understanding is debated.
Cited as an early, striking demo of neural networks acting as game engines and world models.
HN recommends it as a clear argument for why deep and cheap learning succeeds, often alongside the associated paper.
HN commenters use it as the canonical demo of deep reinforcement learning, noting that watching it teaches more than an article does.
HN recommends the series as a very accessible way to understand the math behind machine learning and LLMs.
HN readers recommend it as an intuitive explanation of neural networks that shifts how people think about AI and language.
HN commenters point to it as a candid, contrarian view of self-driving approaches and the real limits of the business.
Recommended as one of the best free courses for learning modern deep learning for NLP.
Recommended as a fun, semi-technical intro to machine learning that holds kids' attention.
Recommended as the clearest overview of how LLM assistants are built and how to use them well.
HN commenters point beginners to this series as one of the clearest short walkthroughs of how neural networks and backpropagation work.
Cited for Karpathy's first-hand explanations of camera-only self-driving, transformers and ML trends.
HN commenters recommend it for building intuition about state space models and the tradeoffs between transformer and SSM layers.
HN commenters cite it in Tesla threads as the primary-source explanation of why Tesla removed radar and how its vision depth estimation works.
HN recommends it as the clear explanation of why Google's statistical, data-first approach works.
Recommended as a favorite tech talk showing how huge data availability changes how problems are approached.
Recommended as an enjoyable, clear introduction to information theory that pairs well with MacKay's free textbook.
HN points to it as a vivid demo that plain A* search can beat fancier machine-learning approaches in game AI.
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.
HN recommends it to newcomers as a clear explanation of attention, including the query/key/value intuition.
HN recommends it as a long but worthwhile way to demystify how language models work technically.
Cited both as a primary source on Tesla's technical autonomy approach and, later, for Musk's unmet robotaxi predictions.
Cited in floating-point threads as the main introduction to posits and unums as an alternative to IEEE floats.
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.
HN recommends it as a quick, non-dumbed-down way to get up to speed on how LLMs are built.
Pointed to as an accessible early explanation of transformers and attention for newcomers.
HN people send others here to hear Hawkins explain his theory in his own words, including reference frames and the old versus new brain.
HN considers the course one of the best hands-on ways into deep learning, praising the code-first teaching style.
Pointed to during self-driving news threads as an insider view of Waymo's progress, deep learning's role and its limits.
HN uses it as primary-source evidence on how Tesla's vision-based driving stack and training pipeline work.
HN points to it as an accessible explainer of AlphaFold and as an example of narrow, specialized ML delivering real scientific breakthroughs.
HN recommends it as a striking, well-explained example of AlphaZero's play and of zugzwang.
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.
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.
Recommended as a well-made, digestible introduction to scaling laws and why models cluster along a similar efficiency frontier.
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.
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.
HN mentions it as a fun and slightly worrying example of AI systems behaving in unintended ways when optimizing.
Used as a go-to example of neural networks enhancing game rendering, referenced in later AI-graphics discussions.
HN recommends it for Hinton's research history and the two pioneers reflecting on a decade of progress.
HN cites it as a fun, tangible way to teach machine learning, including to kids.
Cited as a watchable early introduction to deep learning from its pioneer, and remembered as a mind-blowing moment in the field.
Cited as an accessible explanation of AI denoising, which underpins real-time ray tracing on RTX hardware.
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.
Recommended in Elixir and machine learning threads as the introduction to what Nx makes possible.
Recommended in LLM threads as an authoritative explanation from a ChatGPT architect of why models hallucinate and why imitation is not enough.
HN commenters cite it for NVIDIA's perspective on AI hardware efficiency, including logarithmic number formats and energy use.
Cited as footage of AlphaStar beating a top human StarCraft 2 player, and for accessible game analysis.
Recommended as a more nuanced look at Hinton's views than headlines, especially his argument that LLMs are more than auto-complete.
Pointed to as LeCun's argument for the limits of LLMs and the need for world models.
Recommended as Hinton's own explanation of capsule networks and his critique of pooling in CNNs.
Recommended as an accessible, visual way to understand backpropagation and neural networks.
HN recommends the CS231n lectures as a solid, applied deep learning course for self-study.
HN commenters point to it as an interesting conversation with the creator of Swift and LLVM, especially for compilers and ML compilation.
Recommended as a historical look at the AI winter debate and a demolition of the expert systems hype.
Recommended as an approachable explanation of how reinforcement learning works in practice, and cited for the constraints needed to stop exploits.
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.
HN lists it as a free, brilliant introduction to deep learning and computer vision, often within ML self-study plans.
HN found it a nifty demo and cites it as a primitive but effective learning tracker that predates modern deep-learning trackers.
Recommended as Hawkins' own accessible statement of his brain-inspired theory of intelligence and its software implementation.
HN points to it as an entertaining analysis of AlphaZero's play in threads about chess engines.
Cited in hardware threads as a Keller talk on the different ways of doing parallel processing.
Cited as a fun, unconventional example of neural networks built from floating point error.
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.
HN cites it as an early impressive real-time speech translation demo, often in discussions of voice-preserving translation and later AI demos.
HN people point to it as the canonical explanation from the method's author whenever t-SNE comes up.
Cited as a vivid example of what Deep Dream and feature visualization look like, resembling psychedelic visuals.
HN cites it as a source for how Valve uses machine learning on player behavior to detect cheating.
HN commenters point to it as the talk by the author of Neanderthal and the Clojure GPU libraries, including comparisons against the JVM and CPU.
It is cited as an early example of neural networks learning physics, and recalled during later discussions of AI video generation.
HN uses it as a clear explainer for diffusion mechanics, such as negative prompts and faster sampling, and for joint text-image encoders.
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.
HN commenters flag it as a remarkable machine-learning result, notably for showing that training on raw images beats JPEG.
HN cites it for the DeepMind CEO's early views on AI and its game-playing focus, in AlphaGo and OpenAI threads.
Cited as the source for Google's Swift for TensorFlow effort, later remembered as a once-hyped ML language.
Cited as a source for the idea that open-ended search can beat objective-driven optimization, though one commenter found parts of it fluffy.
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.
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.
Pointed to as evidence that AI misalignment is an empirically demonstrated problem, for people who dismiss it.
HN commenters cite it as the serious skeptic's case against LLMs reaching general intelligence, usually in debates about whether scaling is enough.
HN uses it as a famous example of an overconfident AI prediction that turned out wrong, since radiologists remain in demand.
Cited in discussions of learning as compression and of why neural networks generalize.
Cited as a source for the fact that OpenAI uses Ray for training, and for Brockman's views on developer speed.
HN cites it as evidence that cheap, capable humanoid robots from Chinese companies are arriving, often in contrast to Tesla's Optimus.
HN cites it as a source on Tenstorrent's wide-issue RISC-V Ascalon CPU in RISC-V threads.
Mostly cited as the source of Hotz's AMD driver rant, which some commenters found validating and others found unflattering.
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.
HN commenters cite it as a concise explanation of GPT-3's scaling jump and surprising abilities at its release.
Which of these have you already seen?
Connect YouTube and HackerTube AI hides what you've watched and ranks the rest for you.