📚 An Illustrated Guide to AI Agents Artificial intelligence is entering a new phase. AI agents can now reason, plan, and act with increasing independence. From accelerating scientific breakthroughs to supporting... 🌐 Website

The 25 Best Free Resources for Learning AI in 2026 Learning artificial intelligence in 2026 is both easier and harder than it was a couple of years ago. Easier, because the best courses from Stanford and MIT, books by leading researchers, and hands-on lessons from Hugging Face are all openly available. Harder, because there is now far too much material, and the models and tools change every week. We have collected 25 proven free resources and arranged them by stage: from mathematics to LLMs, agents, and reading the latest papers. Mathematics and Intuition 1. 3Blue1Brown: Neural Networks A video series by Grant Sanderson after which neural networks stop being a black box. Gradient descent, back-propagation, and the attention mechanism in transformers are shown so clearly that the formulas read themselves afterwards. The ideal first entry point, even if you have long forgotten the math. Open the resource. 2. Mathematics for Machine Learning A book by Deisenroth, Faisal, and Ong, free in PDF. Linear algebra, calculus, probability theory, and optimization in exactly the amount needed for ML, and immediately with examples: regression, PCA, support vector machines. Convenient to read alongside courses, returning to the chapter you need. Open the resource. 3. StatQuest with Josh Starmer A YouTube channel where statistics and ML algorithms are explained step by step and with humor: from p-values and decision trees to gradient boosting and transformers. Good as a reference: if a topic in a course is unclear, look for the StatQuest video. Open the resource. Classical Machine Learning 4. Google Machine Learning Crash Course A short, practical course from Google with interactive visualizations and exercises. In a few evenings it gives you the vocabulary and basic concepts: features, overfitting, metrics, neural networks, and working with data. Good for quickly finding out whether you like this field. Open the resource. 5. Kaggle Learn Micro-courses on Python, pandas, ML, visualization, and computer vision, where all the code runs right in the browser. After them you can go straight into Kaggle competitions and look at other people's notebooks, which is one of the most underrated schools of practice. Open the resource. 6. Andrew Ng's Machine Learning Specialization An updated version of the legendary course through which millions of people came into ML. Regression, classification, neural networks, decision trees, recommender systems, and reinforcement learning in Python. The materials can be watched free of charge in audit mode. Open the resource. 7. Yandex Machine Learning Handbook An online textbook from Yandex School of Data Analysis instructors, in Russian. Rigorous but readable: from linear models and ensembles to neural networks, with mathematics and practical remarks. One of the best Russian-language sources for systematic study. Open the resource. 8. mlcourse.ai An open course from the Open Data Science community led by Yuri Kashnitsky. Its strength is practice: data analysis, feature engineering, linear models, boosting, and time series, with homework and worked solutions. It closes the gap between theory and real data excellently. Open the resource. 9. Stanford CS229: Machine Learning The classic Stanford course with serious mathematics: supervised and unsupervised learning, generalization, kernel methods, the EM algorithm. Lectures are posted on YouTube, notes on the course website. For those who want to understand why the algorithms work. Open the resource. Deep Learning 10. fast.ai: Practical Deep Learning for Coders Jeremy Howard's course is built "top-down": from the very first lecture you train working models, and you work through the theory as needed. Suitable for programmers who learn better through code, and it gives a very practical view of model training. Open the resource. 11. Dive into Deep Learning An interactive book where every chapter is runnable code in PyTorch and other frameworks. From linear regression to transformers and optimization. It is used in hundreds of universities and works well for self-study. Open the resource. 12. Understanding Deep Learning Simon Prince's book, free in PDF, with excellent illustrations and notebooks. More modern than the classics: transformers, diffusion models, graph networks, reinforcement learning. Arguably the best deep learning textbook available today. Open the resource. 13. Deep Learning (Goodfellow, Bengio, Courville) The fundamental book often called the bible of deep learning. The section on modern architectures is outdated, but the chapters on mathematics, regularization, and optimization remain a benchmark. Read selectively, when you need depth. Open the resource. 14. MIT 6.S191: Introduction to Deep Learning An intensive MIT course updated every year: fresh lectures on generative models, LLMs and their applications, plus Python labs. Good as a fast overview of the whole field in a couple of weeks. Open the resource. 15. Stanford CS231n: Deep Learning for Computer Vision The course that raised a generation of computer vision researchers. Convolutional networks, training neural networks from scratch, detection, segmentation, and generative models. The assignments that have you implement backpropagation by hand greatly sharpen your understanding. Open the resource. 16. Deep Learning School MIPT A free deep learning school from the Moscow Institute of Physics and Technology, in Russian: tracks in computer vision and NLP, lectures, seminars, and homework in PyTorch. A good option for those who find it easier to study in their native language and within the structure of a real course. Open the resource. Language Models and Transformers 17. Andrej Karpathy: Neural Networks: Zero to Hero A series in which Karpathy writes neural networks from scratch: from automatic differentiation with micrograd to his own GPT and tokenizer. Nothing is hidden behind libraries, and afterwards the design of language models becomes clear down to the line of code. Open the resource. 18. Stanford CS224n: NLP with Deep Learning The flagship university course on natural language processing: word vectors, recurrent networks, attention, transformers, pretraining, and fine-tuning. Lectures on YouTube, assignments on the course website. Open the resource. 19. Stanford CS336: Language Modeling from Scratch A course for those who want to understand how LLMs are actually built: tokenization, architecture, GPU training, scaling laws, data preparation, and alignment. The assignments require you to implement and train your own model. Hard, but afterwards papers about new models read completely differently. Open the resource. 20. Hugging Face Learn A set of free courses from Hugging Face: an LLM course, a course on agents, diffusion models, reinforcement learning, and computer vision. All hands-on with the transformers and datasets libraries and the Hub ecosystem — that is, with the tools the industry actually uses. Open the resource. 21. Jay Alammar: The Illustrated Transformer The article through which half the industry first understood the transformer architecture. The attention mechanism, encoder, and decoder are broken down in pictures step by step. Essential reading before the paper "Attention Is All You Need". Open the resource. 22. Maxime Labonne's LLM Course A popular open repository with an LLM roadmap and Colab notebooks: fine-tuning, quantization, model merging, RAG, evaluation, and deployment. Useful as a map showing what to study next, and as a collection of ready-made recipes. Open the resource. Reinforcement Learning, Research, and Fresh Papers 23. Lil'Log by Lilian Weng The blog of a researcher who worked at OpenAI. Each post is a detailed survey of a topic with dozens of references: LLM agents, hallucinations, diffusion models, adversarial attacks on models, reward hacking. The best way to quickly enter a new research area. Open the resource. 24. Sutton and Barto + OpenAI Spinning Up