AI Resources
Free tools, archives, and reading that we recommend to our participants, for use during the masterclass and long after it.
Every entry links to an external website, which opens in a new tab.
Learn by doing in the browser
A small neural network that trains live in your browser on four toy datasets, with every hidden neuron, its weights, and the loss curve drawn as it learns.
playground.tensorflow.org CNN ExplainerA convolutional network classifying real images, with every convolution, activation, and pooling step shown with its numbers.
poloclub.github.io Transformer ExplainerA GPT-2 model running in the browser: tokens, embeddings, the query, key, and value projections, the attention heat maps, and the next-token probabilities.
poloclub.github.io LLM VisualizationA three-dimensional walk through a small language model, one matrix at a time, from the input tokens to the output distribution.
bbycroft.net Embedding ProjectorWord vectors for 10,000 words and the MNIST digits in three dimensions by PCA or t-SNE, with nearest neighbours by cosine distance.
projector.tensorflow.org PathFinding.jsA grid on which breadth-first search, Dijkstra's algorithm, and A* run step by step, with the explored cells shown.
qiao.github.io Seeing TheoryBrown University's visual introduction to probability and statistics, with interactive chapters on conditional probability, Bayesian inference, and regression.
seeing-theory.brown.edu TiktokenizerType a sentence and see how the tokenizers of current language models split it, and how many tokens the same sentence costs in different languages.
tiktokenizer.vercel.appNotebooks, libraries, and datasets
A free Jupyter notebook that runs Python in the browser on Google's machines, with a GPU on request; the format every AI course and paper repository uses.
colab.research.google.com PyTorch tutorialsThe official tutorials of the library most research groups train with, from a first tensor to a full training loop, runnable in Colab.
docs.pytorch.org scikit-learnThe standard Python library for classical machine learning, with a user guide that explains each method before it shows the code.
scikit-learn.org Kaggle datasetsThe largest public collection of datasets and notebooks, with competitions whose leaderboards show what a good test error looks like; browsing needs no account.
kaggle.com UCI Machine Learning RepositoryThe archive of classic datasets, run at the University of California, Irvine since 1987, with Iris, the Wisconsin breast cancer data, and hundreds more.
archive.ics.uci.edu OpenMLThousands of datasets with their tasks and the published results of many methods on each, so that you can compare your own result with the field's.
openml.org Hugging Face HubThe public repository of open-weight models and datasets, with model cards and a dataset viewer that lets you read preference data row by row.
huggingface.co GymnasiumThe standard library of reinforcement learning environments, from the frozen lake gridworld to Atari and robot simulators, with every state, action, and reward documented.
gymnasium.farama.org NetronOpen a saved model file and see its layers, shapes, and connections drawn as a graph, in the browser.
netron.appRead the research
Where almost every machine learning paper appears first, free to read. New papers every weekday.
arxiv.org arXiv, artificial intelligenceThe listing for search, planning, reasoning, agents, and the rest of the field beyond learning.
arxiv.org Semantic ScholarA search engine of the scientific literature with citation counts, who cited a paper, and a one-sentence summary of each.
semanticscholar.org Hugging Face PapersThe papers the community is reading this week, each linked to its models, datasets, and code.
huggingface.co Journal of Machine Learning ResearchThe field's oldest open-access journal, free to read since 2000, with the survey on automatic differentiation that Unit 1 presents.
jmlr.org OpenReviewThe submissions and the reviewers' discussions of ICLR and other conferences, which show how a paper is judged before it is accepted.
openreview.net DistillInteractive explanations of research, among them the articles on feature visualization and circuits that started modern interpretability. The journal is on hiatus and its archive stays open.
distill.pub Transformer Circuits ThreadThe interpretability research of one laboratory, published as long articles with interactive figures, on features, superposition, and circuits in language models.
transformer-circuits.pubFree textbooks and courses
Simon Prince's textbook on deep learning, free as PDF, with notebooks for every chapter; the reference the masterclass uses for Units 6 to 11.
udlbook.github.io Reinforcement Learning: An IntroductionSutton and Barto's textbook, free as PDF from the authors; the source of the notation Unit 12 uses.
incompleteideas.net Deep LearningGoodfellow, Bengio, and Courville's textbook of 2016, free to read online chapter by chapter.
deeplearningbook.org Dive into Deep LearningAn interactive textbook in which every method comes with runnable code in PyTorch, JAX, or TensorFlow.
d2l.ai Mathematics for Machine LearningDeisenroth, Faisal, and Ong's book on the linear algebra, calculus, and probability that machine learning needs, free as PDF.
mml-book.github.io Probabilistic Machine LearningKevin Murphy's two volumes on machine learning from the probabilistic point of view, free as PDF drafts.
probml.github.io An Introduction to Statistical LearningThe classic introduction to regression, classification, and resampling, with editions in R and Python and free lecture videos.
statlearning.com 3Blue1Brown, neural networksGrant Sanderson's animated videos on neural networks, gradient descent, backpropagation, and transformers.
3blue1brown.com CS231n, StanfordStanford's course on convolutional networks for computer vision, with lecture notes and assignments open to everyone.
cs231n.stanford.eduThe field in numbers
Charts of the field's history from public data: training compute since 1950, parameter counts, benchmark results, investment, and public opinion, each with its source.
ourworldindata.org Epoch AI data hubA maintained database of notable models with their parameter counts, training tokens, compute, hardware, and cost estimates, with the uncertain numbers marked.
epoch.ai Stanford AI IndexThe annual report on the state of the field: research output, technical performance, economy, education, and policy, with the underlying data.
hai.stanford.edu SWE-benchThe benchmark of real software issues that an agent must fix, with the leaderboard of systems, their scores, and the task descriptions.
swebench.com NeuronpediaA public explorer of interpretability data: the features of open models with the texts that activate each one, searchable by text.
neuronpedia.org AlphaFold Protein Structure DatabasePredicted structures for more than 200 million proteins, searchable by name or sequence, with the confidence of every residue shown.
alphafold.ebi.ac.uk Common CrawlThe open web crawl that the training data of most language models starts from, with the size of every monthly archive.
commoncrawl.orgCompete and study
The olympiad in AI for high school students, held since 2024, with the syllabus, the past tasks, and the list of accredited countries and their selection rounds.
ioai-official.org International Olympiad in InformaticsThe olympiad in algorithms and programming for high school students, held since 1989, with the tasks of every edition.
ioinformatics.org Kaggle competitionsOpen machine learning competitions for anyone, with public leaderboards, shared notebooks, and beginner competitions that never close.
kaggle.com Bundeswettbewerb Künstliche IntelligenzThe German AI project competition for school students, organized from Tübingen, with a final each autumn; open to students from Austria and Switzerland as well.
bw-ki.de Regeneron International Science and Engineering FairThe largest pre-college science competition, reached through affiliated fairs in more than 60 countries; machine learning projects compete in several categories.
societyforscience.org Research Experiences for UndergraduatesPaid summer research placements at universities in the United States, funded by the National Science Foundation, including many in computer science and machine learning.
nsf.gov Eastern European Machine Learning Summer SchoolA one-week summer school with lectures and tutorials by researchers from leading laboratories, held each year in a different city of the region, with applications open to students worldwide.
eeml.eu Deep Learning IndabaThe annual gathering of the African machine learning community, with lectures, practicals, and mentorship, and the local IndabaX events across the continent.
deeplearningindaba.comPhotographs on this website are cropped and, where text sits on them, darkened. Mark I Perceptron, 1958: National Museum of the U.S. Navy (Public domain). Go: Dietmar Rabich (CC BY-SA 4.0). ATI RV670 die shot: Martijn Boer (Public domain). Tensor Processing Unit 3.0: Zinskauf (CC BY-SA 4.0). Rear of rack at NERSC data center: Derrick Coetzee (CC0). Bletchley Park Bombe: Mike Peel (CC BY-SA 4.0). SRI Shakey robot, 1969: The wub (CC BY-SA 4.0). Say hello to Deep Blue: Erik Pitti (CC BY 2.0). Analytical Engine: Marcin Wichary (CC BY 2.0). Punch cards, industrial loom: Sunfox (CC BY-SA 2.0). Line drawing of the retina: Santiago Ramón y Cajal, Wellcome Collection (CC BY 4.0). t-SNE embedding of MNIST: Kyle McDonald (CC BY 2.0). Unit photographs are credited in the units. University wordmarks are trademarks of their owners and are shown for identification only.