Learning tracks
256 concepts is a corpus, not a course. Each track below is a bounded route to one destination: every concept it depends on, in the order they build. Computed from the prerequisite graph, so they stay correct as the encyclopedia grows.
From raw text to a working chatbot. The route most people actually want.
Grounding a model in your own documents, and why that is harder than it looks.
Systems that take actions rather than only produce text.
The architecture underneath nearly everything, reached in twelve steps.
Noise to picture, and the family of models that got there.
Turning pixels into meaning. The shortest complete track here.
Why making these systems do what we intend is unsolved.
The longest chain in the corpus: everything behind models that think before answering.
How these are built. Every concept records what must be understood before it. A track is the full set of prerequisites for one destination, ordered so nothing appears before what it depends on. Nothing here is hand-written curriculum: change the graph and the tracks change.
Progress is read from the concepts you have opened, stored in this browser only. Nothing is gated, and you can jump in anywhere. Want a destination that is not listed? Plan what to learn builds a route to any of the 256 concepts.