Learning
AI & Machine Learning
From "what is a tensor" to fine-tuning, agents and evaluation.
Where to start
If you have never trained anything, start with the Google crash course for vocabulary or fast.ai to have something working in week one. If you can already program and want to understand what a network is really doing, Karpathy is the shortest honest route and worth doing before you reach for a framework. If you are building on top of a model rather than training one, skip straight to the provider documentation — that is a different skill and the courses above will not teach it.
Start here5
Enough to understand what everyone is talking about.
- Elements of AIStart hereFree
Non-technical and genuinely good at it. The one to send to a colleague who needs to understand AI without writing code.
- Machine Learning Crash CourseStart hereFree
The standard grounding: loss, gradient descent, overfitting, feature crosses. Short videos with interactive exercises and no framework commitment.
- AI for BeginnersStart hereFree
A twelve-week curriculum with quizzes and labs, covering symbolic AI as well as neural networks.
Trains a working image classifier in lesson one and explains the theory afterwards. The opposite order to most courses, and it suits people who already program.
- Kaggle LearnStart hereFree
Short micro-courses in a browser notebook. Good for filling one specific gap rather than learning the field.
Understand what is happening6
The part most people skip and later wish they had not.
- Neural Networks: Zero to HeroCoreFree
Builds backpropagation, then a language model, from an empty file. If you have used a neural network without knowing what it does, this is the fix.
- The Illustrated TransformerCoreFree
The diagram everybody has seen, with the explanation that makes it land. Read before any paper that says "attention".
- Dive into Deep LearningCoreFree
A full textbook where every equation has runnable code beside it, in PyTorch, TensorFlow and JAX.
- CS229: Machine LearningAdvancedFree
The mathematical treatment, with lecture notes and problem sets. Assumes linear algebra and probability.
- Spinning Up in Deep RLAdvancedFree
The clearest introduction to reinforcement learning there is, with implementations you can read.
- DistillAdvancedFree
Archived but not obsolete: interactive explanations of what is happening inside models. The feature-visualisation work is still unmatched.
Building with models12
If you are calling an API rather than training weights, this is your section.
- Claude DocsCoreFree
The reference for building on Claude: prompting, tool use, context handling, agents and the API itself.
- Anthropic coursesCoreFree
Notebooks on prompt engineering, tool use and evaluation, from the people who build the model.
- OpenAI Platform DocsCoreFree
API reference and guides for the GPT models, including structured outputs and function calling.
- OpenAI CookbookCoreFree
Working recipes rather than tutorials. Where to look when you know what to build and need the shape of the code.
- GitHub Copilot DocsStart hereFree
How Copilot actually works, including the parts people miss: custom instructions, chat participants and enterprise controls.
- Gemini API DocsCoreFree
Reference for Google’s models, with a generous free tier for experimenting.
- Hugging Face LearnCoreFree
Courses on transformers, diffusion, audio and agents, by the people who maintain the libraries you would use.
- Prompt Engineering GuideStart hereFree
A survey of prompting techniques with the papers behind them, rather than a list of magic phrases.
- LangChain DocsCoreFree
The most-used orchestration framework. Worth understanding even if you decide not to use it.
- LlamaIndex DocsCoreFree
Retrieval and indexing over your own documents, which is what most "chat with your data" projects actually need.
- OllamaCoreFree
Run open-weight models on your own machine with one command. The cheapest way to experiment without an API bill.
- DeepLearning.AI short coursesCoreFree
Hour-long courses built with model providers on narrow, current topics — RAG, evaluation, agents.
Evaluation, safety and keeping up5
The difference between a demo and something you would put in front of a customer.
- Weights & BiasesCoreFree tier
Experiment tracking. The moment you have run the same thing twice with different settings, you need this or something like it.
- arXiv: Machine LearningAdvancedFree
Where the papers land first. Unfiltered, so use a summary source alongside it rather than instead of it.
- Hugging Face PapersAdvancedFree
Trending papers with the implementations and discussion attached, which is how you tell a result from a claim. Where Papers with Code ended up.
The framework procurement teams increasingly ask about. Useful whether or not you find frameworks useful.
- EU AI Act ExplorerCoreFree
The regulation itself, browsable by article, with the risk tiers explained. Relevant to anyone shipping into the EU.
About this list
28 resources, checked automatically every week so dead links do not sit here unnoticed. Nothing on this page is sponsored and none of these links pay us — which is the only reason a recommendation on it is worth anything. Every link goes to the original source, and we take nothing with you when you leave.