Stuff weβd actually recommend if youβre getting into AI. New to all this? Start at the top.
How long it takes
π an afternoon or less Β· ππ a few days to a week Β· πππ a few weeks or more
How technical it is
π₯ beginner friendly Β· π₯π₯ some maths/coding needed Β· π₯π₯π₯ heavy going
β = my favourites/ highly recommended
Start here
3Blue1Brown β Neural Networks π π₯ β
Really nice visual explanation of how neural nets learn. You can watch it on the bus. Trust bro youβre going to like it
Andrew Ng β Machine Learning Specialization πππ π₯
Itβs a classic and pretty much everyoneβs done it. Good grounding in the basics.
Karpathy β Neural Networks: Zero to Hero πππ π₯π₯ β
Starts with backprop by hand and ends with you building a GPT from scratch. Karpathy is one of the leading AI researchers who worked at Tesla, OpenAI, Anthropic. V good course
Deep learning
Dive into Deep Learning πππ π₯π₯
Textbook with code you can run for every section. Free online, or you can buy it in print.
Stanford CS231n πππ π₯π₯π₯
Stanfordβs computer vision course. The lecture notes on CNNs and backprop are some of the best out there.
fast.ai β Practical Deep Learning for Coders πππ π₯
You train real models from lesson one and pick up the theory along the way. Good if you learn by doing.
The Illustrated Transformer π π₯π₯ β
Jay Alammarβs diagrams of how transformers work. Read this before you try to understand any LLM paper.
Reinforcement learning
Learning it
Spinning Up in Deep RL ππ π₯π₯ β
OpenAIβs written intro to RL and the main policy gradient algorithms, with reference code. Best place to start.
Deriving the simplest policy gradient π π₯π₯π₯
The section of Spinning Up where policy gradients finally click.
Karpathy β Pong from Pixels π π₯π₯
The π Karpathy again. Trains a Pong agent in about 130 lines of Python.
Hugging Face Deep RL Course πππ π₯π₯
Free and hands on, from Q-learning up to PPO, with notebooks.
David Silverβs RL lectures πππ π₯π₯π₯
The UCL course from the guy who led AlphaGo. Proper theory.
Sutton & Barto πππ π₯π₯π₯
The og bible. Free PDF. More of a reference than something you read cover to cover.
Going further
The 37 Implementation Details of PPO ππ π₯π₯π₯ β
The PPO bible. All the stuff the paper leaves out that you need to actually get it working.
Entity-Based RL π π₯π₯π₯
Clemens Winter on entity encoding, which is a cornerstone once your observations get complicated.
Libraries
Gymnasium π₯
The standard environment API. Pretty much everything below plugs into it.
Stable-Baselines3 π₯
Best beginner library. Get an agent training in a few lines.
CleanRL π₯π₯
Best library for learning. Each algorithm is one readable file.
PufferLib π₯π₯π₯
Best library for frontier work. Very fast.
Tooling
10 minutes to pandas π π₯
Fundamental for datascience/ ML work.
NumPy for absolute beginners π π₯
Arrays and broadcasting. Everything else is built on this.
PyTorch tutorials ππ π₯π₯
Start with βLearn the Basicsβ.
Weights & Biases π₯
Logs your training runs so you can compare them instead of scrolling back through your terminal.
uv π₯
Fast Python package and environment manager. Will save you a lot of dependency pain.