some youtube channels I recommend for those interested in understanding current capability trends; separate comments for votability. Please open each one synchronously as it catches your eye, then come back and vote on it. downvote means not mission critical, plenty of good stuff down there too.
I’m subscribed to every single channel on this list (this is actually about 10% of my youtube subscription list), and I mostly find videos from these channels by letting the youtube recommender give them to me and pushing myself to watch them at least somewhat to give the cute little obsessive recommender the reward it seeks for showing me stuff. definitely I’d recommend subscribing to everything.
Let me know which if any of these are useful, and please forward the good ones to folks—this short form thread won’t get seen by that many people!
Yannic Kilcher: paper explanations, capability news. Yannic is the machine learning youtuber. 129k subscribers, every one of whom has published 200 papers on machine learning (I kid). Has some of the most in depth and also broad paper explanations, with detailed drawings of his understanding of the paper. Great for getting a sense of how to read a machine learning paper. his paper choices are top notch and his ML news videos have really great capabilities news. https://www.youtube.com/channel/UCZHmQk67mSJgfCCTn7xBfew
Valence Discovery: graph NNs, advanced chem models. Valence Discovery is a research group focusing on advanced chemical modeling. We don’t have full strength general agent AI to plug into this quite yet, and certainly not safe reinforcement learning, but work like theirs has thoroughly eclipsed human capabilities in understanding chemicals. as long as we can use narrow ai to prevent general AI from destroying the cooperation network between beings, I think work like this has the potential to give the world every single goal of transhumanism: post scarcity, molecular assemblers, life extension, full bodily autonomy and morphological freedom, the full lot should be accessible. It’ll take a bit longer to get to that level, but the research trajectory continues to look promising and these models haven’t been scaled as much as language models. https://www.youtube.com/channel/UC3ew3t5al4sN-Zk01DGVKlg
Steve Brunton: fancy visual lectures on nonlinear control systems & ML. has some of the best educational content I’ve ever seen, just barely beating Mutual Information for explanation quality while going into much more advanced topics. Focuses on control theory, nonlinear control, dynamical systems, etc. https://www.youtube.com/channel/UCm5mt-A4w61lknZ9lCsZtBw
It’s several college courses worth of material—it really depends what you want out of it. I personally am extremely curiosity-driven; without assessing what you already know I don’t feel able to give strong recommendations of where to start, which is in fact why I posted so many links here in the first place. if you want to work through Brunton’s content sequentially, I’d suggest picking the course playlist that interests you: https://www.youtube.com/c/Eigensteve/playlists
If your interests are mostly unprimed, I’d suggest checking out the physics-informed ML and sparsity playlists, maybe also skip around the fluid dynamics playlist to get a sense of what’s going on there. Alternately, skim a few videos to get a sense of which ones are relevant to your interests (2x speed with heavy jumping around), then queue the playlist that seems appropriate to you. If you really find it useful you might benefit from actually doing it like a course—I generally underpractice compared to ideal practice amount.
The simons institute: very best wide variety, especially ai safety and game theory.
The simons institute for theoretical computer science at UC Berkeley is a contender for my #1 recommendation from this whole list. Banger talk after banger talk after banger talk there. Several recent workshops with kickass ai safety focus. https://www.youtube.com/user/SimonsInstitute
they have a number of “boot camp” lessons that appear to be meant for an interdisciplinary advanced audience as well. the current focus of talks is on causality and games, and they also have some banger talks on “how not to run a forecasting competition”, “the invisible hand of prediction”, “communicating with anecdotes”, “the challenge of understanding what users want”, and my personal favorite due to its fundamental reframing of what game theory even is, “in praise of game dynamics”: https://www.youtube.com/watch?v=lCDy7XcZsSI
Schwartz Reisman Institute is a multi-agent safety discussion group, one of the very best ai safety sources I’ve seen anywhere. a few interesting videos include, for example, this one, which I think is on the cutting edge in terms of where AI safety will eventually end up (potentially multi-agent safety that comes into existence after humanity dies, if we don’t get there fast enough to prevent darwinist AIs that don’t love us from literally eating us, as yudkowsky describes with the words “does not love you, does not hate you, made out of atoms that can be used for something else”):
SRI’s weekly Seminar Series welcomes Richard Watson, associate professor in the Agents, Interaction and Complexity group at the University of Southampton’s School of Electronics and Computer Science. Watson has over 80 publications on topics spanning artificial life, robotics, evolutionary computation, population genetics, neural networks, evolutionary theory and computational biology, and is the author of Compositional Evolution: The Impact of Sex, Symbiosis, and Modularity on the Gradualist Framework of Evolution (MIT Press, 2006).
Watson’s research seeks to deepen our understanding of biological evolution by expanding the formal equivalence of learning and evolution—in particular, using connectionist models of cognition and learning. In this talk, he will introduce the concept of “natural induction” as a critique of the ideas of Universal Darwinism that focuses on the evolutionary potential of how organisms develop relationships by working together
Mutual Information: visual explanations of ML fundamentals. Mutual Information is one of the absolute best tutorial-and-explanation videos about the visual math of basic (small-model) machine learning. includes things like gaussian processes, which, it turns out, neural networks are a special case of. This means that neural networks are actually equivalent to non-parametric models, the weights are simply a reprojection of the training data (kinda obvious in retrospect), and understanding gaussian processes is not optional in understanding how neural networks interpolate between their training data. His video on gaussian processes is wonderful. https://www.youtube.com/watch?v=UBDgSHPxVME—lots of other interesting videos as well https://www.youtube.com/channel/UCCcrR0XBH0aWbdffktUBEdw
Machine Learning Street Talk: Industry professionals giving talks meant for youtube. is one of the most interesting interview series-es (seriesen? serii?) on youtube. Discusses stuff like gflownets with yoshua bengio, geometric deep learning, thousand brains theory—all the stuff you really, really need to understand if you want to have any sense at all of where machine learning is going. (no, it’s not hitting a wall.) https://www.youtube.com/channel/UCMLtBahI5DMrt0NPvDSoIRQ
IPAM at UCLA: academic talks; Math, quantum, ML, game theory, ai safety, misc. is one of the most notable channels on this list; lots of hard math topics, but also quite a few extremely interesting ML topics, including an absolute banger talk series on distributed computation and collective intelligence. They also discuss extremely interesting topics about advanced physics which is way above my head as a self-taught ML nerd, but very interesting to attempt to absorb. https://www.youtube.com/c/IPAMUCLA/videos
IARAI: cutting-edge academic ML talks. “The Institute of Advanced Research in Artificial Intelligence” is not messing around with their name. The recent discussion of “Neural diffusion PDEs, differential geometry, and graph neural networks” seems to me to be a major next direction in ai capabilities, refining the issues with transformers with fundamental mathematics of graph curvature. “How GNNs and Symmetries can help solve PDEs” is also promising, though I haven’t watched all the way through yet. https://www.youtube.com/channel/UClC7A82p47Nnj8ttU_COYeA/videos
CPAIOR: formal verification in general, including on deep learning. Has a number of interesting videos on formal verification, how it works, and some that apply it to machine learning, eg “Safety in AI Systems—SMT-Based Verification of Deep Neural Networks”; “Formal Reasoning Methods in Machine Learning Explainability”; “Reasoning About the Probabilistic Behavior of Classifiers”; “Certified Artificial Intelligence”; “Explaining Machine Learning Predictions”; a few others. https://www.youtube.com/channel/UCUBpU4mSYdIn-QzhORFHcHQ/videos
William Spaniel is a textbook writer and youtube video author on game theory. Probably not as relevant to an advanced audience, but has nice if slightly janky intros to the concepts.
edit: since I posted this, he’s gotten into detailed descriptions of war incentives and as a result became quite popular.
https://www.youtube.com/user/JimBobJenkins
Edan Meyer makes mid-level paper explanations. Not quite as good as yannic kilcher yet, but getting there. Has discussed a number of notable papers Yannic hasn’t gotten to yet, such as the deepmind scaling laws paper. One of the higher production-quality, on-the-edge channels I’ve encountered for its level of beginner-friendliness, though. https://www.youtube.com/c/EdanMeyer/videos
Emergent Garden is a fairly new channel, but has a great video on why even a simple feedforward network is already a very powerful general function approximator. Compare Art Of The Problem. https://www.youtube.com/watch?v=0QczhVg5HaI
“Web IR / NLP Group at NUS” has talks, many from google research, about information retrieval, which is looking more and more likely to be a core component of any superintelligence (what a surprise, given the size of the internet, right? except also, information retrieval and interpolation is all that neural networks do anyway, see work on Neural Tangent Kernel) https://www.youtube.com/channel/UCK8KLoKYvow7X6pe_di-Gvw/videos
udiprod makes animated explainer videos about advanced computer science, including some fun quantum computer science. also has a visualization of, eg, an SVM. https://www.youtube.com/c/udiprod/videos
The AI Epiphany is a solid paper explanations channel, and his choices of paper to discuss are often telling in terms of upcoming big-deal directions. Not quite as good as Yannic IMO, but imo worth at least subscribing to. https://www.youtube.com/c/TheAIEpiphany/videos
Stanford MLSys Seminars is where talks from the Hazy Research group at stanford get posted, and their work has been some of the most eye-catching for me in the past two years. In particular, the S4 sequence model seems to me to represent a major capability bump in next-step-after-transformers models, due to its unusually stable learning. I might just be taken in by a shiny toy, but S4 is the next thing I’m going to play with capabilities wise. https://www.youtube.com/c/StanfordMLSysSeminars
Robert Miles makes kickass AI safety videos. Y’all probably already know about him. He has repeated many opinions I don’t think hold that came from less wrong, but if reading the archives here isn’t your jam, watching the archives on his channel might be better. https://www.youtube.com/channel/UCLB7AzTwc6VFZrBsO2ucBMg
Reducible creates absolutely kickass computer science explanation videos, including one on why jpeg is so effective, another on the interesting information routing in the fast fourier transform. https://www.youtube.com/channel/UCK8XIGR5kRidIw2fWqwyHRA
another slightly-off-topic one, Paul Beckwith discusses large-scale climate science, and hooo boy it really isn’t looking good at all if his estimates are remotely on target. We’re going to need that weather superintelligence you published a few steps towards, deepmind! https://www.youtube.com/user/PaulHBeckwith
Oxford VGG continues to be one of the most cutting edge vision research groups, and their presentations on generative models of images, 3d neural rendering, etc seem very promising in fixing the 3d reasoning gap that is still present in powerful models like DALL-E 2. https://www.youtube.com/channel/UCFXBh2WNhGDXFNafOrOwZEQ/videos
nPlan: paper discussion group—they’re a research group of some kind or other that does great paper-discussion meetups and posts them to youtube. Paper-discussion with multiple confused researchers is in general more to my preference than paper-explanation with one confused researcher explaining it to the audience, because having multiple folks makes sure more questions come up. Competitive with Yannic for “best papers-summary channel on youtube” (as far as I’ve found, anyway) because of the format difference. https://www.youtube.com/c/nPlan/videos
Normalized Nerd is another overviews channel with good overviews of various basic small-model ml approaches. Not as good as Mutual Information, but mostly they don’t overlap. https://www.youtube.com/c/NormalizedNerd/featured
Neuroscientifically Challenged makes great quick-intro 2-minute videos on neuroscience topics. Not the most important in understanding machine learning at this point since the stuff about the brain that is still likely to usefully generalize is rather advanced details of neuron behaviors and is likely not as useful as the general research direction towards [conservation laws, symmetries, continuous space&time, etc] research track, but relevant to generalizing machine learning knowledge to the brain, and relevant to general understanding of the brain. https://www.youtube.com/c/Neuroscientificallychallenged/videos
MIT Embodied Intelligence: industry professionals giving academic talks. Is a channel (and presumably org of some kind) that posts talks with major industry and research folks. Recent talks include “Recent advances in deep equilibrium models”, “The deep learning toolbox: from alphafold to alphacode”, and “the past, present, and future of SLAM”. https://www.youtube.com/channel/UCnXGbvgu9071i3koFooncAw/videos
Mind under Matter is a pop-explanations channel about neuroscience, which I absolutely love, she really goes over the top making it fun and playful and imo hits it out of the park. Definitely upper intro level, but a great recommendation if that’s an interesting topic to you. https://www.youtube.com/c/MindUnderMatter/videos
Justin Solomon has a number of video topics on his channel, but notably a class he taught on Shape Analysis in 2021, which covers a number of interesting subtopics. I added the whole class to my watch later and have occasionally been speedwatching it when it comes up on shuffle. https://www.youtube.com/c/justinmsolomon/featured
Jordan Harrod is an ML person who is also a popsci-ML video creator. She has lots of great stuff on things like “how I self-study”, “is it too late to get into machine learning”, “productivity tools I tried and didn’t like”, etc. not as information dense as the talks channels, but a good subscription-without-bell on youtube, and I occasionally love her stuff. https://www.youtube.com/c/JordanHarrod/videos
Joint Mathematics Meetings has quite a number of interesting videos on math, but the one where I found their channel was this one, Daniel Spielman on “Miracles of Algebraic Graph Theory”. Presents, among other things, a demonstration of why the first eigenvectors of some graph representation or other (I have to rewatch it every damn time to remember exactly which one) end up being an analytical solution to force-directed graph drawing. https://www.youtube.com/watch?v=CDMQR422LGM—https://www.youtube.com/channel/UCKxjz1WXZOKcAh9T9CBfJoA
[1] tangent: as long as ML doesn’t suddenly smash the “defect against other life” button really really hard like yudkowsky is terrified its totally gonna (I think he’s just given himself a paranoia disorder and is unable to evaluate algorithms without pascals-mugging himself out of the steps of the reasoning process, but that’s another thread)
GAMMA UMD posts paper summary videos, thought they’re not the most industry-changing they can be interesting. topics like Automatic Excavactor [sic], Speech2AffectiveGestures, Text2Gestures, etc. https://www.youtube.com/c/gammaunc/videos
Fancy Fueko is an intro level programming-and-AI channel. She makes great stuff and makes it look shiny and neon—I occasionally reference her stuff when feeling mentally diffuse and need a reminder. Same category as Daniel Bourke. https://www.youtube.com/c/fancyfueko/videos
“DeepMind ELLIS UCL CSML Seminar Series” (what a mouthful) appears to be a sponsored-by-deepmind series at a school, one of those acronyms is probably the school name. UCL? has a bunch of interesting topics, but I haven’t found it to be as cutting edge as some other channels, maybe I haven’t watched the right videos. https://www.youtube.com/channel/UCiCXRD_NcvVjkLCE39GkwVQ/videos
Conference on Robot Learning has many great talks and is sponsored by a number of serious industry groups. Examples include “Safe Reinforcement Learning”, “A fabrics perspective on nonlinear behavior representation”, “walking the boundary of learning and interaction”, “integrating planning and learning for scalable robot decision making”, etc. https://www.youtube.com/c/ConferenceonRobotLearning
Conference on Computer-Aided Verification has a number of interesting talks on how to do verified neuro-symbolic ML. recent videos include “modular synthesis of reactive programs”, “neuro-symbolic program synthesis from natural language and demonstrations”, “gradient descent over metagrammars for syntax guided synthesis”. I think transformers are more powerful than any of these techniques, but they provide interesting comparison for what a model (eg transformers) must be able to learn in order to succeed. https://www.youtube.com/channel/UCe3M4Hc2hCeNGk54Dcbrbpw/videos
CMU AI Seminar: Paper presentations by authors. Has some great talks on various projects, such as one that I think is significantly beyond SOTA in learning efficiency, DreamCoder: https://www.youtube.com/watch?v=KykcFYDkAHo
AIPursuit archives talks they find notable, including many from major conferences. a quick browse is necessary to find what you seek in this archive. Links to several related channels they also run with subtopics, such as RL. https://www.youtube.com/c/AIPursuit/featured
sentdex makes lots of fun tutorial and livecoding videos, including some recent ones about building neural networks completely from scratch in order to understand the computation steps exactly. https://www.youtube.com/user/sentdex
the Institute of Advanced Study has many remarkable videos, but they are on a wide variety of mathematical topics. A recent interesting-and-on-topic one is “Multi-group fairness, loss minimization and indistinguishability”. https://www.youtube.com/channel/UC8aRaZ6_0weiS50pvCmo0pw
Huggingface post videos to youtube about their python library, nothing terribly fancy but can be convenient to have it pop up in my recommender between in-depth videos. https://www.youtube.com/c/HuggingFace
Henry AI Labs is a research group (I think?) that also have a podcast, and they often advertise ML products on it. They’ve advertised weaviate several times, which does look like a fairly nice ready-to-use vector+trad search database, though I haven’t actually tried it yet. They also have discussions about APIs, causal inference, misc other stuff. https://www.youtube.com/channel/UCHB9VepY6kYvZjj0Bgxnpbw/videos
Cyrill Stachniss makes various video summaries of ML topics, especially focusing on applied topics like plant phenotyping, self-driving-car perception, etc. includes interviews, etc. https://www.youtube.com/c/CyrillStachniss/videos
Andreas Geiger is a vision researcher who posts vision research to youtube. Vision has some major steps left before completion, and his work seems like a promising direction in that process to me. includes NeRF stuff. https://www.youtube.com/user/cvlibs
Alex Smola makes lecture-style ~30 minute videos on various machine learning topics, including some recent ones on shapley values, fairness, graph neural networks, etc. https://www.youtube.com/c/smolix/videos
Oxford Mathematics is a widely ranging math channel that I don’t strongly recommend, but which passed my inclusion criteria of quality and may be worth checking out. Has an interesting video series on math with machine learning. https://www.youtube.com/channel/UCLnGGRG__uGSPLBLzyhg8dQ
Prof. Nando de Freitas is a machine learning researcher/teacher who has an old class on deep learning on youtube—reasonable, but imo insufficiently concise and out of date. Don’t recommend, included for completeness. Watch to get the youtube recommender to give you old stuff like it, if you feel like. https://www.youtube.com/user/ProfNandoDF
Hausdorff Center for Mathematics is focused on hard math, and I haven’t found it super interesting. Including for completeness since I found it originally while watching lots of math videos. https://www.youtube.com/c/HausdorffCenterforMathematics
slightly less on-topic, “Fluid Mechanics 101” goes through a number of interesting topics on fluids and the math behind them. As usual with any large-scale physics, it ends up being another example of tensor programming, just like machine learning. I wonder if there’s some connection? /s
https://www.youtube.com/channel/UCcqQi9LT0ETkRoUu8eYaEkg
CIS 522 Deep Learning is a class at some university or other. Lots of interesting discussion, including one, “Lyle Ungar’s Personal Meeting Room”, which discusses ethics in what imo is a solid way. not that trad lesswrongers are going to agree with me on that. https://www.youtube.com/channel/UCT1ejuxsdomILyc5I2EdzYg/videos
“GraphXD: Graphs Across Domains” is an archive of a talk series on graph theory, including eg “A History of Spectral Graph Theory”, “Linear Regression with Graph Constraints”, “Graph Clustering Algorithms”. including for completeness, seems outdated. https://www.youtube.com/channel/UCzee-ohKJciqFvxnIT1sYpg/videos
some youtube channels I recommend for those interested in understanding current capability trends; separate comments for votability. Please open each one synchronously as it catches your eye, then come back and vote on it. downvote means not mission critical, plenty of good stuff down there too.
I’m subscribed to every single channel on this list (this is actually about 10% of my youtube subscription list), and I mostly find videos from these channels by letting the youtube recommender give them to me and pushing myself to watch them at least somewhat to give the cute little obsessive recommender the reward it seeks for showing me stuff. definitely I’d recommend subscribing to everything.
Let me know which if any of these are useful, and please forward the good ones to folks—this short form thread won’t get seen by that many people!
edit: some folks have posted some youtube playlists for ai safety as well.
Yannic Kilcher: paper explanations, capability news. Yannic is the machine learning youtuber. 129k subscribers, every one of whom has published 200 papers on machine learning (I kid). Has some of the most in depth and also broad paper explanations, with detailed drawings of his understanding of the paper. Great for getting a sense of how to read a machine learning paper. his paper choices are top notch and his ML news videos have really great capabilities news. https://www.youtube.com/channel/UCZHmQk67mSJgfCCTn7xBfew
Valence Discovery: graph NNs, advanced chem models. Valence Discovery is a research group focusing on advanced chemical modeling. We don’t have full strength general agent AI to plug into this quite yet, and certainly not safe reinforcement learning, but work like theirs has thoroughly eclipsed human capabilities in understanding chemicals. as long as we can use narrow ai to prevent general AI from destroying the cooperation network between beings, I think work like this has the potential to give the world every single goal of transhumanism: post scarcity, molecular assemblers, life extension, full bodily autonomy and morphological freedom, the full lot should be accessible. It’ll take a bit longer to get to that level, but the research trajectory continues to look promising and these models haven’t been scaled as much as language models. https://www.youtube.com/channel/UC3ew3t5al4sN-Zk01DGVKlg
The Alan Turing Institute: variety, lately quite a bit of ai safety. eg: https://www.youtube.com/channel/UCcr5vuAH5TPlYox-QLj4ySw
they have a playlist of recent ai safety videos, many of which look like they plausibly include information not heavily discussed, or at least not well indexed, on less wrong https://www.youtube.com/watch?v=ApGusxR7JAc&list=PLuD_SqLtxSdXVSrXneEPkZtzTTQMT4hQ8
They discuss social issues, including stuff like who gets to decide a non-explosive ai’s targets https://www.youtube.com/watch?v=4Txa7pAOHZQ&list=PLuD_SqLtxSdVy8meO_ezV9l89Q9Gg8q6p
quite a few more interesting playlists on safety and security of ai in the playlists section https://www.youtube.com/c/TheAlanTuringInstituteUK/playlists
lots of discussion of complex systems
in particular, I love their video on social network analysis and I recommend it often https://www.youtube.com/watch?v=2ZHuj8uBinM&list=PLuD_SqLtxSdWcl2vx4K-0mSflRRLyfwlJ&index=9
Steve Brunton: fancy visual lectures on nonlinear control systems & ML. has some of the best educational content I’ve ever seen, just barely beating Mutual Information for explanation quality while going into much more advanced topics. Focuses on control theory, nonlinear control, dynamical systems, etc. https://www.youtube.com/channel/UCm5mt-A4w61lknZ9lCsZtBw
Where do I start with this channel? Oldest video first?
It’s several college courses worth of material—it really depends what you want out of it. I personally am extremely curiosity-driven; without assessing what you already know I don’t feel able to give strong recommendations of where to start, which is in fact why I posted so many links here in the first place. if you want to work through Brunton’s content sequentially, I’d suggest picking the course playlist that interests you: https://www.youtube.com/c/Eigensteve/playlists
If your interests are mostly unprimed, I’d suggest checking out the physics-informed ML and sparsity playlists, maybe also skip around the fluid dynamics playlist to get a sense of what’s going on there. Alternately, skim a few videos to get a sense of which ones are relevant to your interests (2x speed with heavy jumping around), then queue the playlist that seems appropriate to you. If you really find it useful you might benefit from actually doing it like a course—I generally underpractice compared to ideal practice amount.
The simons institute: very best wide variety, especially ai safety and game theory. The simons institute for theoretical computer science at UC Berkeley is a contender for my #1 recommendation from this whole list. Banger talk after banger talk after banger talk there. Several recent workshops with kickass ai safety focus. https://www.youtube.com/user/SimonsInstitute
A notable recent workshop is “learning in the presence of strategic behavior”: https://www.youtube.com/watch?v=6Uq1VeB4h3w&list=PLgKuh-lKre101UQlQu5mKDjXDmH7uQ_4T
another fun one is “learning and games”: https://www.youtube.com/watch?v=hkh23K3-EKw&list=PLgKuh-lKre13FSdUuEerIxW9zgzsa9GK9
they have a number of “boot camp” lessons that appear to be meant for an interdisciplinary advanced audience as well. the current focus of talks is on causality and games, and they also have some banger talks on “how not to run a forecasting competition”, “the invisible hand of prediction”, “communicating with anecdotes”, “the challenge of understanding what users want”, and my personal favorite due to its fundamental reframing of what game theory even is, “in praise of game dynamics”: https://www.youtube.com/watch?v=lCDy7XcZsSI
Schwartz Reisman Institute is a multi-agent safety discussion group, one of the very best ai safety sources I’ve seen anywhere. a few interesting videos include, for example, this one, which I think is on the cutting edge in terms of where AI safety will eventually end up (potentially multi-agent safety that comes into existence after humanity dies, if we don’t get there fast enough to prevent darwinist AIs that don’t love us from literally eating us, as yudkowsky describes with the words “does not love you, does not hate you, made out of atoms that can be used for something else”):
“An antidote to Universal Darwinism”—https://www.youtube.com/watch?v=ENpdhwYoF5g
as well as this kickass video on “whose intelligence, whose ethics” https://www.youtube.com/watch?v=ReSbgRSJ4WY
https://www.youtube.com/channel/UCSq8_q4SCU3rYFwnA2bDxyQ
Mutual Information: visual explanations of ML fundamentals. Mutual Information is one of the absolute best tutorial-and-explanation videos about the visual math of basic (small-model) machine learning. includes things like gaussian processes, which, it turns out, neural networks are a special case of. This means that neural networks are actually equivalent to non-parametric models, the weights are simply a reprojection of the training data (kinda obvious in retrospect), and understanding gaussian processes is not optional in understanding how neural networks interpolate between their training data. His video on gaussian processes is wonderful. https://www.youtube.com/watch?v=UBDgSHPxVME—lots of other interesting videos as well https://www.youtube.com/channel/UCCcrR0XBH0aWbdffktUBEdw
Machine Learning Street Talk: Industry professionals giving talks meant for youtube. is one of the most interesting interview series-es (seriesen? serii?) on youtube. Discusses stuff like gflownets with yoshua bengio, geometric deep learning, thousand brains theory—all the stuff you really, really need to understand if you want to have any sense at all of where machine learning is going. (no, it’s not hitting a wall.) https://www.youtube.com/channel/UCMLtBahI5DMrt0NPvDSoIRQ
IPAM at UCLA: academic talks; Math, quantum, ML, game theory, ai safety, misc. is one of the most notable channels on this list; lots of hard math topics, but also quite a few extremely interesting ML topics, including an absolute banger talk series on distributed computation and collective intelligence. They also discuss extremely interesting topics about advanced physics which is way above my head as a self-taught ML nerd, but very interesting to attempt to absorb. https://www.youtube.com/c/IPAMUCLA/videos
The collective intelligence workshop playlist: https://www.youtube.com/watch?v=qhjho576fms&list=PLHyI3Fbmv0SfY5Ft43_TbsslNDk93G6jJ
IARAI: cutting-edge academic ML talks. “The Institute of Advanced Research in Artificial Intelligence” is not messing around with their name. The recent discussion of “Neural diffusion PDEs, differential geometry, and graph neural networks” seems to me to be a major next direction in ai capabilities, refining the issues with transformers with fundamental mathematics of graph curvature. “How GNNs and Symmetries can help solve PDEs” is also promising, though I haven’t watched all the way through yet. https://www.youtube.com/channel/UClC7A82p47Nnj8ttU_COYeA/videos
CPAIOR: formal verification in general, including on deep learning. Has a number of interesting videos on formal verification, how it works, and some that apply it to machine learning, eg “Safety in AI Systems—SMT-Based Verification of Deep Neural Networks”; “Formal Reasoning Methods in Machine Learning Explainability”; “Reasoning About the Probabilistic Behavior of Classifiers”; “Certified Artificial Intelligence”; “Explaining Machine Learning Predictions”; a few others. https://www.youtube.com/channel/UCUBpU4mSYdIn-QzhORFHcHQ/videos
William Spaniel is a textbook writer and youtube video author on game theory. Probably not as relevant to an advanced audience, but has nice if slightly janky intros to the concepts. edit: since I posted this, he’s gotten into detailed descriptions of war incentives and as a result became quite popular. https://www.youtube.com/user/JimBobJenkins
The National Socio-Environmental Synthesis Center has a number of topics that felt a bit scientifically offbeat to me, but in particular, talks on knowledge integration across disciplines I found remarkably interesting. https://www.youtube.com/playlist?list=PLIGFwrZq94y-rj8CKOaVzBXGD5OTmeelc
https://www.youtube.com/c/TheNationalSocioEnvironmentalSynthesisCenter
The Berkman Klein Center for Internet and Society has some interesting discussion content that gets into ai safety: https://www.youtube.com/playlist?list=PL68azUN8PTNjTUsspsam0m0KmmUZ6l1Sh
https://www.youtube.com/c/BKCHarvard
Edan Meyer makes mid-level paper explanations. Not quite as good as yannic kilcher yet, but getting there. Has discussed a number of notable papers Yannic hasn’t gotten to yet, such as the deepmind scaling laws paper. One of the higher production-quality, on-the-edge channels I’ve encountered for its level of beginner-friendliness, though. https://www.youtube.com/c/EdanMeyer/videos
Emergent Garden is a fairly new channel, but has a great video on why even a simple feedforward network is already a very powerful general function approximator. Compare Art Of The Problem. https://www.youtube.com/watch?v=0QczhVg5HaI
ACM SIGPLan is a special interest group on programming languages. Talks, discussions, presentations, long videos. https://www.youtube.com/channel/UCwG9512Wm7jSS6Iqshz4Dpg
“Welcome AI Overlords” is a popsci ML-intros channel with high quality explanations of things like Graph Attention Networks: https://www.youtube.com/watch?v=SnRfBfXwLuY and an author interview with Equivariant Subgraph Aggregation Networks: https://www.youtube.com/watch?v=VYZog7kbXks https://www.youtube.com/channel/UCxw9_WYmLqlj5PyXu2AWU_g
“Web IR / NLP Group at NUS” has talks, many from google research, about information retrieval, which is looking more and more likely to be a core component of any superintelligence (what a surprise, given the size of the internet, right? except also, information retrieval and interpolation is all that neural networks do anyway, see work on Neural Tangent Kernel) https://www.youtube.com/channel/UCK8KLoKYvow7X6pe_di-Gvw/videos
UMich-CURLY is a research group and associated youtube channel discussing Simultaneous Localization And Mapping (SLAM) with neural networks. a recent overview talk was particularly interesting: https://www.youtube.com/watch?v=TUOCMevmbOg—https://www.youtube.com/channel/UCZ7Up19hdIWuCSuuATlzlbw/videos
udiprod makes animated explainer videos about advanced computer science, including some fun quantum computer science. also has a visualization of, eg, an SVM. https://www.youtube.com/c/udiprod/videos
The AI Epiphany is a solid paper explanations channel, and his choices of paper to discuss are often telling in terms of upcoming big-deal directions. Not quite as good as Yannic IMO, but imo worth at least subscribing to. https://www.youtube.com/c/TheAIEpiphany/videos
Stanford MLSys Seminars is where talks from the Hazy Research group at stanford get posted, and their work has been some of the most eye-catching for me in the past two years. In particular, the S4 sequence model seems to me to represent a major capability bump in next-step-after-transformers models, due to its unusually stable learning. I might just be taken in by a shiny toy, but S4 is the next thing I’m going to play with capabilities wise. https://www.youtube.com/c/StanfordMLSysSeminars
Robert Miles makes kickass AI safety videos. Y’all probably already know about him. He has repeated many opinions I don’t think hold that came from less wrong, but if reading the archives here isn’t your jam, watching the archives on his channel might be better. https://www.youtube.com/channel/UCLB7AzTwc6VFZrBsO2ucBMg
Reducible creates absolutely kickass computer science explanation videos, including one on why jpeg is so effective, another on the interesting information routing in the fast fourier transform. https://www.youtube.com/channel/UCK8XIGR5kRidIw2fWqwyHRA
A few more programming languages channels I don’t think are worth their own votable comments:
PLISS—programming language implementation summer school—https://www.youtube.com/channel/UCofC5zis7rPvXxWQRDnrTqA/videos
POPL 2019 - https://www.youtube.com/channel/UCe0bH8tWBjH_Fpqs3veiIzg
another slightly-off-topic one, Paul Beckwith discusses large-scale climate science, and hooo boy it really isn’t looking good at all if his estimates are remotely on target. We’re going to need that weather superintelligence you published a few steps towards, deepmind! https://www.youtube.com/user/PaulHBeckwith
Oxford VGG continues to be one of the most cutting edge vision research groups, and their presentations on generative models of images, 3d neural rendering, etc seem very promising in fixing the 3d reasoning gap that is still present in powerful models like DALL-E 2. https://www.youtube.com/channel/UCFXBh2WNhGDXFNafOrOwZEQ/videos
One World Theoretical Machine Learning is a paper-discussions channel I’ve watched nearly none of but which looks very interesting. https://www.youtube.com/channel/UCz7WlgXs20CzugkfxhFCNFg/videos
nPlan: paper discussion group—they’re a research group of some kind or other that does great paper-discussion meetups and posts them to youtube. Paper-discussion with multiple confused researchers is in general more to my preference than paper-explanation with one confused researcher explaining it to the audience, because having multiple folks makes sure more questions come up. Competitive with Yannic for “best papers-summary channel on youtube” (as far as I’ve found, anyway) because of the format difference. https://www.youtube.com/c/nPlan/videos
Normalized Nerd is another overviews channel with good overviews of various basic small-model ml approaches. Not as good as Mutual Information, but mostly they don’t overlap. https://www.youtube.com/c/NormalizedNerd/featured
Neuroscientifically Challenged makes great quick-intro 2-minute videos on neuroscience topics. Not the most important in understanding machine learning at this point since the stuff about the brain that is still likely to usefully generalize is rather advanced details of neuron behaviors and is likely not as useful as the general research direction towards [conservation laws, symmetries, continuous space&time, etc] research track, but relevant to generalizing machine learning knowledge to the brain, and relevant to general understanding of the brain. https://www.youtube.com/c/Neuroscientificallychallenged/videos
MIT Embodied Intelligence: industry professionals giving academic talks. Is a channel (and presumably org of some kind) that posts talks with major industry and research folks. Recent talks include “Recent advances in deep equilibrium models”, “The deep learning toolbox: from alphafold to alphacode”, and “the past, present, and future of SLAM”. https://www.youtube.com/channel/UCnXGbvgu9071i3koFooncAw/videos
Mind under Matter is a pop-explanations channel about neuroscience, which I absolutely love, she really goes over the top making it fun and playful and imo hits it out of the park. Definitely upper intro level, but a great recommendation if that’s an interesting topic to you. https://www.youtube.com/c/MindUnderMatter/videos
Justin Solomon has a number of video topics on his channel, but notably a class he taught on Shape Analysis in 2021, which covers a number of interesting subtopics. I added the whole class to my watch later and have occasionally been speedwatching it when it comes up on shuffle. https://www.youtube.com/c/justinmsolomon/featured
Jordan Harrod is an ML person who is also a popsci-ML video creator. She has lots of great stuff on things like “how I self-study”, “is it too late to get into machine learning”, “productivity tools I tried and didn’t like”, etc. not as information dense as the talks channels, but a good subscription-without-bell on youtube, and I occasionally love her stuff. https://www.youtube.com/c/JordanHarrod/videos
Joint Mathematics Meetings has quite a number of interesting videos on math, but the one where I found their channel was this one, Daniel Spielman on “Miracles of Algebraic Graph Theory”. Presents, among other things, a demonstration of why the first eigenvectors of some graph representation or other (I have to rewatch it every damn time to remember exactly which one) end up being an analytical solution to force-directed graph drawing. https://www.youtube.com/watch?v=CDMQR422LGM—https://www.youtube.com/channel/UCKxjz1WXZOKcAh9T9CBfJoA
Interpretable Machine Learning is an archive of some discussions about interpretability from a NeurIPS 2017. Great talks, definitely worth some speedwatching if interpretability is of interest. https://www.youtube.com/channel/UCv0AwnKZkSk2sU1mkETYfIw/videos
“Intelligent Systems Lab” appears to be a university class focused on intro to ML. Not my first recommendation for the topic, but solid, above 50% percentile on this list IMO. https://www.youtube.com/channel/UC7qFYa4HVoufKcz-2q3pr7A/videos
Hugo Larochelle is a deep learning researcher who has also made a number of interesting talks and discussion videos, including this interesting playlist from the TechAide AI4Good conference-and-hackathon in 2020. https://www.youtube.com/watch?v=jFRnvtiPpL8&list=PL6Xpj9I5qXYFTaKnvgyfFFkxrOb4Ss_-J
Harvard Medical AI: ML for medical science, cutting edge academic talks. They publish talks on machine learning for medical science, probably the most important use of machine learning IMO[1] - includes eg this interesting discussion of geometric deep learning, one of the most promising next directions for ML in my opinion. https://www.youtube.com/watch?v=oz3vaxFleh4 - https://www.youtube.com/channel/UCld99fdpOgqW80TW-oOvltA/videos
[1] tangent: as long as ML doesn’t suddenly smash the “defect against other life” button really really hard like yudkowsky is terrified its totally gonna (I think he’s just given himself a paranoia disorder and is unable to evaluate algorithms without pascals-mugging himself out of the steps of the reasoning process, but that’s another thread)
GAMMA UMD posts paper summary videos, thought they’re not the most industry-changing they can be interesting. topics like Automatic Excavactor [sic], Speech2AffectiveGestures, Text2Gestures, etc. https://www.youtube.com/c/gammaunc/videos
Fancy Fueko is an intro level programming-and-AI channel. She makes great stuff and makes it look shiny and neon—I occasionally reference her stuff when feeling mentally diffuse and need a reminder. Same category as Daniel Bourke. https://www.youtube.com/c/fancyfueko/videos
“DeepMind ELLIS UCL CSML Seminar Series” (what a mouthful) appears to be a sponsored-by-deepmind series at a school, one of those acronyms is probably the school name. UCL? has a bunch of interesting topics, but I haven’t found it to be as cutting edge as some other channels, maybe I haven’t watched the right videos. https://www.youtube.com/channel/UCiCXRD_NcvVjkLCE39GkwVQ/videos
Conference on Robot Learning has many great talks and is sponsored by a number of serious industry groups. Examples include “Safe Reinforcement Learning”, “A fabrics perspective on nonlinear behavior representation”, “walking the boundary of learning and interaction”, “integrating planning and learning for scalable robot decision making”, etc. https://www.youtube.com/c/ConferenceonRobotLearning
Conference on Computer-Aided Verification has a number of interesting talks on how to do verified neuro-symbolic ML. recent videos include “modular synthesis of reactive programs”, “neuro-symbolic program synthesis from natural language and demonstrations”, “gradient descent over metagrammars for syntax guided synthesis”. I think transformers are more powerful than any of these techniques, but they provide interesting comparison for what a model (eg transformers) must be able to learn in order to succeed. https://www.youtube.com/channel/UCe3M4Hc2hCeNGk54Dcbrbpw/videos
CMU Robotics has a number of interesting talks, including some about ethics of ai robotics and robust human-robot interaction. https://www.youtube.com/user/cmurobotics/videos
CMU AI Seminar: Paper presentations by authors. Has some great talks on various projects, such as one that I think is significantly beyond SOTA in learning efficiency, DreamCoder: https://www.youtube.com/watch?v=KykcFYDkAHo
Art of the Problem makes explainer videos that are unusually high quality among explainer videos I’ve encountered, especially among ones on deep learning. https://www.youtube.com/playlist?list=PLbg3ZX2pWlgKV8K6bFJr5dhM7oOClExUJ
AIPursuit archives talks they find notable, including many from major conferences. a quick browse is necessary to find what you seek in this archive. Links to several related channels they also run with subtopics, such as RL. https://www.youtube.com/c/AIPursuit/featured
“What’s AI” is a popsci-only channel about ai, but the content doesn’t seem completely off base, just popular-audience focused https://www.youtube.com/channel/UCUzGQrN-lyyc0BWTYoJM_Sg
“Visual Inference” is a channel with misc paper presentation videos. Doesn’t seem like the most remarkable paper presentation videos channel ever, but it’s interesting. https://www.youtube.com/channel/UCBk6WGWfm7mjqftlHzJOt5Q/videos
TUM-DAML is a research group that posts discussions of their papers. A recent interesting one is “Ab-initio Potential Energy Surfaces by Pairing GNNs with Neural Wave Functions”. https://www.youtube.com/channel/UC0sPhfmHXhNE7lOv5J3wteg
The Royal Institution is a bit like popsci for scientists. in depth talks, not always my first choice but pretty solid and recommendable. https://www.youtube.com/user/TheRoyalInstitution
Stanford MedAI’s youtube talks aren’t quite as kickass as the harvard medical channel, but they’re pretty solid https://www.youtube.com/channel/UCOkkljs06NPPkjNysCdQV4w/videos
sentdex makes lots of fun tutorial and livecoding videos, including some recent ones about building neural networks completely from scratch in order to understand the computation steps exactly. https://www.youtube.com/user/sentdex
DrSaradaHerke made a couple of classes on graph theory and discrete maths a few years ago. Solid content. https://www.youtube.com/user/DrSaradaHerke
Jeremy Mann makes tutorial videos on topics like Homological Algebra. https://www.youtube.com/user/jmann277/videos
jbstatistics is a fairly solid statistics intro class, with nice animated explanations. not the best I’ve ever seen, but solid. https://www.youtube.com/user/jbstatistics/videos
the Institute for Neural Computation has some of the most interesting hard-neuroscience talks I’ve found on youtube yet, such as this one about basis vectors of the central nervous system. https://www.youtube.com/watch?v=xQX4GIDh_pI—https://www.youtube.com/channel/UCV1SrkEl2-UI60GZlXy5gLA/videos
the Institute of Advanced Study has many remarkable videos, but they are on a wide variety of mathematical topics. A recent interesting-and-on-topic one is “Multi-group fairness, loss minimization and indistinguishability”. https://www.youtube.com/channel/UC8aRaZ6_0weiS50pvCmo0pw
Huggingface post videos to youtube about their python library, nothing terribly fancy but can be convenient to have it pop up in my recommender between in-depth videos. https://www.youtube.com/c/HuggingFace
Henry AI Labs is a research group (I think?) that also have a podcast, and they often advertise ML products on it. They’ve advertised weaviate several times, which does look like a fairly nice ready-to-use vector+trad search database, though I haven’t actually tried it yet. They also have discussions about APIs, causal inference, misc other stuff. https://www.youtube.com/channel/UCHB9VepY6kYvZjj0Bgxnpbw/videos
Eye on AI is a podcast-style discussion channel. eg, here’s a discussion about protein labeling. https://www.youtube.com/watch?v=90ymin29K7g—https://www.youtube.com/channel/UC-o9u9QL4zXzBwjvT1gmzNg
Deeplizard makes entry-level and glossary M-Anim videos about various machine learning topics. https://www.youtube.com/c/deeplizard/videos
Cyrill Stachniss makes various video summaries of ML topics, especially focusing on applied topics like plant phenotyping, self-driving-car perception, etc. includes interviews, etc. https://www.youtube.com/c/CyrillStachniss/videos
Andreas Geiger is a vision researcher who posts vision research to youtube. Vision has some major steps left before completion, and his work seems like a promising direction in that process to me. includes NeRF stuff. https://www.youtube.com/user/cvlibs
Alfredo Canziani makes long, in-depth videos about cutting edge topics, often inviting experts such as Yann LeCun. https://www.youtube.com/c/AlfredoCanziani/videos
Alex Smola makes lecture-style ~30 minute videos on various machine learning topics, including some recent ones on shapley values, fairness, graph neural networks, etc. https://www.youtube.com/c/smolix/videos
AI Coffee break with Latita is a mid-level beginner ai techniques youtuber-production-value channel. https://www.youtube.com/channel/UCobqgqE4i5Kf7wrxRxhToQA
Vision Learning is a misc talks channel with mostly intro level content and discussion of applied robotics. Mediocre compared to most stuff on this list, but worth a mention. https://www.youtube.com/channel/UCmct-3iP5w66oZzN_V5dAMg/videos
“Vector Podcast”: Podcast on vector search engines. unremarkable compared to most of the stuff I’ve linked. https://www.youtube.com/c/VectorPodcast/videos
The bibites is a fun life simulation channel that demonstrates some of the stuff that comes up in evobio and game theory from the other channels I’ve recommended today https://www.youtube.com/channel/UCjJEUMnBFHOP2zpBc7vCnsA
Oxford Mathematics is a widely ranging math channel that I don’t strongly recommend, but which passed my inclusion criteria of quality and may be worth checking out. Has an interesting video series on math with machine learning. https://www.youtube.com/channel/UCLnGGRG__uGSPLBLzyhg8dQ
Prof. Nando de Freitas is a machine learning researcher/teacher who has an old class on deep learning on youtube—reasonable, but imo insufficiently concise and out of date. Don’t recommend, included for completeness. Watch to get the youtube recommender to give you old stuff like it, if you feel like. https://www.youtube.com/user/ProfNandoDF
Missing Semester is a little off-topic, but is an MIT (after-hours?) course on misc tools one needs in computer science work. https://www.youtube.com/channel/UCuXy5tCgEninup9cGplbiFw
Jeremy Howard made fast.ai and has various misc intro content on youtube. definitely not my first recommendation, but if fast.ai seems shiny then this is one place on youtube you can learn about it. https://www.youtube.com/user/howardjeremyp
Hausdorff Center for Mathematics is focused on hard math, and I haven’t found it super interesting. Including for completeness since I found it originally while watching lots of math videos. https://www.youtube.com/c/HausdorffCenterforMathematics
slightly less on-topic, “Fluid Mechanics 101” goes through a number of interesting topics on fluids and the math behind them. As usual with any large-scale physics, it ends up being another example of tensor programming, just like machine learning. I wonder if there’s some connection? /s https://www.youtube.com/channel/UCcqQi9LT0ETkRoUu8eYaEkg
Fancy Manifold is a bit of a stretch, but they have a whole bunch of really good pinned channels as well as a couple of M-Anim videos on physics manifolds. https://www.youtube.com/c/fancymanifold/featured
Daniel Bourke makes entry-level programming videos, with a focus on AI. https://www.youtube.com/channel/UCr8O8l5cCX85Oem1d18EezQ/videos
CIS 522 Deep Learning is a class at some university or other. Lots of interesting discussion, including one, “Lyle Ungar’s Personal Meeting Room”, which discusses ethics in what imo is a solid way. not that trad lesswrongers are going to agree with me on that. https://www.youtube.com/channel/UCT1ejuxsdomILyc5I2EdzYg/videos
anucvml posts their paper overviews, such as recent ICCV papers on image retrieval, smooth pose sequences, spatially conditioned graphs for detecting human object interactions, etc. https://www.youtube.com/channel/UC36k2pZk3TmEweWFt6sIlqw/featured
2d3d.ai is a channel discussing 3d data in neural networks. talks, discussions, presentations. https://www.youtube.com/channel/UCHObHaxTXKFyI_EI8HiQ5xw
“GraphXD: Graphs Across Domains” is an archive of a talk series on graph theory, including eg “A History of Spectral Graph Theory”, “Linear Regression with Graph Constraints”, “Graph Clustering Algorithms”. including for completeness, seems outdated. https://www.youtube.com/channel/UCzee-ohKJciqFvxnIT1sYpg/videos