Mystery Board

(1551 Words, 9 Minutes)

Mystery Board — Unsolved Observations from Act I

During Act I (Lectures 1-3), you used AI tools to write code, explore a real codebase, and build software. Along the way, you noticed things that didn’t quite make sense — behaviors you could observe but not explain.

This page tracks those observations as unsolved mysteries. Each week in Act II, we resolve more of them as we learn how LLMs actually work. Use the dropdown below to see the board’s state after each lecture.

The Four Mechanisms

Every mystery from Act I has a mechanistic explanation rooted in how LLMs actually work:

• Tokenization determines what the model can perceive
• Attention determines what the model focuses on
• Next-token prediction determines how the model generates
• The orchestration layer determines what context the model receives

Understanding these four mechanisms transforms AI from a mysterious black box into a tool you can reason about, predict, and engineer around.