Making the invisible visible — one reasoning step at a time.
Welcome to Annotated Thought, an organization dedicated to illuminating the how and why behind ideas in artificial intelligence and machine learning. We believe that understanding a result is only half the picture. The other half is the reasoning that got you there.
Most explanations in AI/ML tell you what happens. We annotate the thinking.
Our work breaks complex concepts, architectures, and phenomena into richly documented walkthroughs — exposing the chain of decisions, assumptions, and intuitions that underlie them. Each piece is designed to be read like a conversation with a careful thinker, not a textbook.
Whether it's tracing why a particular training dynamic emerges, unpacking the geometry of an attention mechanism, or reconstructing the reasoning behind a design choice — we make the process explicit.
- Reasoning is a first-class artifact. The thought process matters as much as the conclusion.
- Depth over breadth. We'd rather cover one concept thoroughly than many concepts shallowly.
- Rigor without opacity. Precision and clarity are not in conflict — we pursue both.
- Built for researchers. Our content assumes technical fluency and rewards careful reading.
- Annotated walkthroughs of ML concepts, training dynamics, and architectural decisions
- Step-by-step reasoning traces that make implicit assumptions explicit
- Deep dives into topics where the standard explanation leaves too much unsaid
We welcome researchers who share the conviction that understanding how to think through a problem is as valuable as any result.
- Browse our repositories to explore current work
- Open an issue to suggest a topic, flag an error, or start a discussion
- Submit a pull request if you'd like to contribute an annotation or correction
Annotated Thought is intentional. An annotation is more than a comment — it is an act of making latent reasoning legible. Our goal is to do that for the ideas that shape modern AI research.
If you can explain the reasoning, you understand the thing.