Resources
Where to go next. 57 textbooks, courses, explainers, papers, and reports chosen for accuracy and clarity, each with a note on who it suits. 55 of them are free to read or watch.
Textbooks
Complete, carefully edited books. Most can be read free online.
Understanding Deep Learning
Simon J. D. Prince
Readers who want a clear, well-illustrated path from first principles to transformers and diffusion models.
MIT Press, 2023. The author keeps the full PDF free and updated (latest February 2026), with 68 Python notebook exercises.
udlbook.github.io
Dive into Deep Learning
Aston Zhang, Zachary C. Lipton, Mu Li, and Alexander J. Smola
Programmers who learn best by running code: every section is an executable notebook in PyTorch, JAX, TensorFlow, or MXNet.
Print edition from Cambridge University Press.
d2l.ai
Deep Learning
Ian Goodfellow, Yoshua Bengio, and Aaron Courville
Students who want the mathematical foundations of deep learning, from linear algebra and probability to optimization and regularization.
MIT Press. Free to read as HTML; the publisher's contract rules out a PDF. Predates transformers.
deeplearningbook.org
Deep Learning: Foundations and Concepts
Christopher M. Bishop and Hugh Bishop
Upper-level students and practitioners who want a rigorous, self-contained course in deep learning, including an introduction to probability theory.
Springer. The authors' site has a free-to-use online version; hardback and eBook are sold.
bishopbook.com
Mathematics for Machine Learning
Marc Peter Deisenroth, A. Aldo Faisal, and Cheng Soon Ong
Anyone whose linear algebra, calculus, or probability is rusty and who wants exactly the math that machine learning uses.
Cambridge University Press. The authors keep the PDF freely available.
mml-book.github.io
An Introduction to Statistical Learning
Gareth James, Daniela Witten, Trevor Hastie, Robert Tibshirani, and Jonathan Taylor
Readers who want the classical methods taught here, such as regression, classification, trees and ensembles, and clustering, with little math beyond algebra.
Python edition (ISLP), Springer, 2023; the 2021 second edition uses R. The authors' site offers PDFs of both.
statlearning.com
Probabilistic Machine Learning: An Introduction
Kevin P. Murphy
Readers who want machine learning framed through probability, from linear models to deep networks, as a reference to return to.
MIT Press. Free draft PDF under a CC BY-NC-ND license.
probml.github.io
Probabilistic Machine Learning: Advanced Topics
Kevin P. Murphy
Graduate students and researchers going deeper into inference, generative models, and decision making under uncertainty.
MIT Press. Free draft PDF under a CC BY-NC-ND license.
probml.github.io
Reinforcement Learning: An Introduction
Richard S. Sutton and Andrew G. Barto
Anyone who wants reinforcement learning from the ground up: bandits, Markov decision processes, temporal-difference learning, Q-learning, and policy gradients.
Second edition, MIT Press, 2018. The authors offer the full PDF free on their own site, incompleteideas.net.
mitpress.mit.edu
Speech and Language Processing
Daniel Jurafsky and James H. Martin
Anyone studying language technology, from tokens and n-grams to transformers and large language models.
Third edition draft, released chapter by chapter; the latest full draft is dated August 19, 2026.
web.stanford.edu
Artificial Intelligence: A Modern Approach
Stuart Russell and Peter Norvig
Readers who want the whole field, including search, logic, planning, and probabilistic reasoning, not only machine learning.
Fourth edition, Pearson. The site offers the contents, preface, exercises, code, and figures.
aima.cs.berkeley.edu
Courses and lecture series
University courses and video series with a syllabus, from first principles to research.
Neural networks
Grant Sanderson (3Blue1Brown)
Newcomers who want to see what a network, gradient descent, backpropagation, and a transformer are doing before reading any equations.
Animated lessons with written versions, starting from a network that reads handwritten digits.
3blue1brown.com
Practical Deep Learning for Coders
Jeremy Howard (fast.ai)
People with some coding experience who want to train useful models first and learn the theory as they go.
Video lessons and notebooks; no university math required.
course.fast.ai
Neural Networks: Zero to Hero
Andrej Karpathy
Programmers who want to build backpropagation and then a GPT-style language model from scratch, one line of code at a time.
Assumes solid Python and intro-level math, such as derivatives and the Gaussian distribution.
karpathy.ai
Machine Learning Crash Course
Google
Beginners who want a short, practical tour of models, data, neural networks, and LLMs, with exercises in the browser.
Refreshed in 2024 with modules on large language models and AutoML.
developers.google.com
MIT 6.S191 Introduction to Deep Learning
Alexander Amini and Ava Amini (MIT)
Learners who want a fast-paced lecture series on deep learning methods and applications, with software labs.
Lectures on YouTube; slides and lab code are open to everyone.
introtodeeplearning.com
LLM Course
Hugging Face
Developers who want hands-on practice using, fine-tuning, and sharing language models with open source libraries.
About 6 to 8 hours per chapter; code runs in Colab or SageMaker Studio Lab.
huggingface.co
CS229: Machine Learning
Stanford University
Students comfortable with linear algebra and probability who want the mathematical foundations of classical machine learning.
Lecture notes are on the course site; Andrew Ng's Autumn 2018 lectures are on Stanford's YouTube channel.
cs229.stanford.edu
CS231n: Deep Learning for Computer Vision
Stanford University
Learners who want to understand and implement neural networks for images, with assignments built from scratch.
Course materials on the course site; lecture videos from past offerings on YouTube.
cs231n.stanford.edu
CS224N: Natural Language Processing with Deep Learning
Stanford University
Learners ready for a university course covering word vectors, neural network foundations, self-attention, and transformers.
Slides and assignments are posted online; lectures from the Spring 2024 offering are on YouTube.
web.stanford.edu
Deep Reinforcement Learning Course
Hugging Face
Learners who want to train reinforcement learning agents themselves, from the basics to state-of-the-art methods, with Stable Baselines3 and CleanRL.
Self-paced, with a free certificate of completion for finishing 80% of the assignments.
huggingface.co
Large Language Model Agents (MOOC)
Dawn Song (UC Berkeley) and Xinyun Chen (Google DeepMind)
Readers who want a lecture series on how language model agents reason, plan, use tools, and are evaluated, with talks by researchers from many labs.
Fall 2024 lectures with slides; a Spring 2025 sequel, Advanced Large Language Model Agents, is linked from the page.
llmagents-learning.org
AI Agents Course
Hugging Face
Developers who want to build agents that call tools, from the fundamentals to the smolagents, LlamaIndex, and LangGraph frameworks.
Self-paced, about a week per chapter, with free certificates.
huggingface.co
Introduction to Flow Matching and Diffusion Models (MIT 6.S184)
Peter Holderrieth and Ezra Erives
Readers comfortable with calculus and probability who want the math of diffusion and flow models from first principles, after the diffusion lessons here.
IAP 2026 offering: course notes, lecture slides, and three Colab labs, ending with a diffusion transformer.
diffusion.csail.mit.edu
Audio Course
Hugging Face
Developers who want to work with speech and audio: audio data, transformer architectures for audio, speech recognition, and text to speech.
All materials free and open source; units were released in June and July 2023.
huggingface.co
Interactive explainers
Pages you can play with, in the spirit of this site.
A Neural Network Playground
Daniel Smilkov and Shan Carter
Anyone who wants to add layers and neurons to a small network and watch it learn to separate points in real time.
Open source; the neural networks track here has a similar lab.
playground.tensorflow.org
Transformer Explainer
Aeree Cho, Grace C. Kim, and colleagues (Georgia Institute of Technology)
Learners who want to follow their own prompt through a real GPT-2 model, step by step, from tokens to the next-token probabilities.
Runs GPT-2 (small, 124 million parameters) in the browser.
poloclub.github.io
LLM Visualization
Brendan Bycroft
Visual learners who want to see every matrix in a GPT-style network in 3D while a guided tour explains each step.
The walkthrough uses a tiny network from Andrej Karpathy's minGPT that sorts letters.
bbycroft.net
CNN Explainer
Zijie J. Wang and colleagues (Georgia Institute of Technology)
Beginners who want to see how convolution, activation, and pooling layers turn an image into a prediction.
Presented at IEEE VIS 2020.
poloclub.github.io
The Illustrated Transformer
Jay Alammar
Readers who want the original transformer explained with step-by-step diagrams before reading the paper.
A diagram-heavy article rather than an interactive page.
jalammar.github.io
Distill
Distill editors and authors
Readers who want deep, interactive articles on how models work, such as momentum, feature visualization, and circuits.
On hiatus since July 2021; the published articles remain online.
distill.pub
Seeing Theory
Daniel Kunin and colleagues (Brown University)
Anyone who wants an intuitive, visual grounding in probability and statistics before studying machine learning.
seeing-theory.brown.edu
Interpretability
Research and tools for finding out what is happening inside trained models.
Transformer Circuits Thread
Anthropic interpretability team
Readers who want to see how researchers reverse engineer what happens inside transformers, from toy models to production language models.
Good starting points: the 2021 mathematical framework and the 2022 induction heads paper.
transformer-circuits.pub
Zoom In: An Introduction to Circuits
Chris Olah, Nick Cammarata, Ludwig Schubert, Gabriel Goh, Michael Petrov, and Shan Carter
Readers new to interpretability who want the case for studying networks neuron by neuron and connection by connection.
Published in Distill, March 2020, using an image classification network.
distill.pub
Neuronpedia
Johnny Lin and contributors
Learners who want to browse features found inside real language models and try steering a model by hand.
An open source interpretability platform.
neuronpedia.org
TransformerLens
Neel Nanda and contributors
Python users who want to load an open model, cache and edit its internal activations, and run their own interpretability experiments.
Library documentation, with a getting started guide to mechanistic interpretability.
transformerlensorg.github.io
ARENA curriculum
Alignment Research Engineer Accelerator (ARENA)
Programmers who want structured exercises in PyTorch fundamentals, transformer interpretability, reinforcement learning, and LLM evaluations.
Materials are free online; ARENA also runs in-person programs.
arena.education
Seminal papers
Primary sources behind ideas taught here. Each also appears on the history timeline.
Learning representations by back-propagating errors
David E. Rumelhart, Geoffrey E. Hinton, and Ronald J. Williams
Anyone curious how backpropagation was presented to a wide audience: four pages, with networks that learn their own features.
Subscription required at Nature; many libraries provide access.
nature.com
ImageNet Classification with Deep Convolutional Neural Networks
Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton
Readers who want to see the design choices (ReLU, dropout, GPUs, data augmentation) behind the 2012 ImageNet result.
proceedings.neurips.cc
Playing Atari with Deep Reinforcement Learning
Volodymyr Mnih and colleagues (DeepMind)
Readers who have finished the reinforcement learning track and want the paper that combined Q-learning with a convolutional network learning from pixels.
arxiv.org
Adam: A Method for Stochastic Optimization
Diederik P. Kingma and Jimmy Ba
Readers who have met momentum and RMSProp and want the short paper behind the most common default optimizer.
arxiv.org
Deep Residual Learning for Image Recognition
Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun
Readers who want to know why adding a layer's input to its output made very deep networks trainable.
arxiv.org
Attention Is All You Need
Ashish Vaswani et al.
Readers who have finished the transformer lessons here and are ready for the original architecture paper.
arxiv.org
Scaling Laws for Neural Language Models
Jared Kaplan et al.
Readers who want the evidence that language model loss falls as a power law in parameters, data, and compute.
arxiv.org
Language Models are Few-Shot Learners
Tom B. Brown et al.
Readers who want to see how prompting with a few examples, instead of fine-tuning, emerged as models grew to 175 billion parameters.
arxiv.org
Denoising Diffusion Probabilistic Models
Jonathan Ho, Ajay Jain, and Pieter Abbeel
Readers who have tried the diffusion lab here and want the training objective and sampling procedure in full.
arxiv.org
Training language models to follow instructions with human feedback
Long Ouyang et al.
Readers who want the recipe that turns a pretrained model into an assistant: demonstrations, a reward model, and reinforcement learning.
arxiv.org
ReAct: Synergizing Reasoning and Acting in Language Models
Shunyu Yao and colleagues (Princeton University and Google)
Readers of the agent lessons who want the paper behind the loop of reasoning, tool call, and observation.
arxiv.org
Robust Speech Recognition via Large-Scale Weak Supervision
Alec Radford, Jong Wook Kim, and colleagues (OpenAI)
Readers of the speech-to-text lesson who want to know how Whisper was trained on 680,000 hours of audio and how well it transfers.
arxiv.org
Safety, policy, and the state of the field
Reports and frameworks from public institutions and research groups.
International AI Safety Report
Expert panel led by Yoshua Bengio, nominated by over 30 countries and international organizations
Anyone who wants an independent, evidence-based summary of what general-purpose AI can do, the risks it poses, and how they can be mitigated.
Second edition published February 3, 2026, with a summary for policymakers.
internationalaisafetyreport.org
Introduction to AI Safety, Ethics, and Society
Dan Hendrycks (Center for AI Safety)
Readers from any background who want a structured introduction to AI risk as a societal problem, not only a technical one.
Free to read online; print edition from Taylor and Francis; a virtual course is offered.
aisafetybook.com
The Future of AI
BlueDot Impact
Newcomers who want a short, self-paced overview of frontier AI and how people work on making it go well.
About 2 hours, no application; BlueDot also runs longer cohort courses.
bluedot.org
AI Index Report
Stanford Institute for Human-Centered Artificial Intelligence
Policymakers, journalists, researchers, and anyone else who wants vetted yearly data on AI capabilities, investment, adoption, and governance.
hai.stanford.edu
AI Risk Management Framework (AI RMF 1.0)
National Institute of Standards and Technology (NIST)
Teams that build or deploy AI and want a voluntary framework for identifying, measuring, and managing its risks.
Released January 26, 2023, with a companion playbook and later profiles.
nist.gov
AI Act
European Commission
Readers who want to understand the EU's risk-based law on AI: what it bans, what it requires of high-risk systems, and when rules apply.
Regulation (EU) 2024/1689. Some dates were changed by a 2026 amendment; the page tracks the current timeline.
digital-strategy.ec.europa.eu
OECD AI Principles
Organisation for Economic Co-operation and Development (OECD)
Readers who want the principles for trustworthy AI agreed by governments through the OECD, now with 47 adherents.
The OECD calls them the first intergovernmental standard on AI. Adopted May 2019, updated May 2024.
oecd.ai
AI Security Institute
UK government
Readers who want to follow government research on evaluating the capabilities and risks of advanced AI models.
A UK government research organization; publishes research and funds grants.
aisi.gov.uk
Links and details (editions, years, and whether a resource is free) were checked on the date this list was last updated. Prices and availability change; the publisher's page is the authority. The history timeline cites the primary sources behind each milestone.