Andrej Karpathy builds neural networks from scratch in a video series that removes the black-box magic. The single best free starting point for understanding how models actually compute.

Jay Alammar’s visual walkthrough of the transformer architecture. Still the clearest explanation of attention mechanisms, years after publication.

Free interactive courses on transformers, NLP and agents with hands-on notebooks. The practical companion to the theory, backed by the ecosystem’s main hub.

The "top-down" alternative to Karpathy: build real models first, learn theory as needed. Free course materials with a pragmatic, code-first philosophy.

The de-facto reference for the ChatGPT API ecosystem, including function calling, structured outputs and prompting patterns that are now industry vocabulary.

Claude’s docs with genuinely useful engineering guidance — prompt caching, tool use and the "build effective agents" essays are worth reading regardless of which model you use.

The preprint server where AI research actually lands. Rough, fast and unfiltered — the closest thing to reading the field’s pulse before it reaches the blogs.

Primary source for Gemini and Google’s research, including the original Transformer paper and later system cards that describe real-world model behavior.