An Interactive Book
What's inside when
everything looks fine.
A guided exploration of how machine learning systems work — not what they do, but what happens inside. Each chapter is interactive: you observe, adjust, and see the system respond in real time.
What's inside when everything looks fine.
↑ About
Chapters
IThe Space Where Words Live
Words as points in geometric space. Meaning as distance. The map the machine navigates before it says a single word.
The Art of Paying Attention
How a model decides what matters. Attention as a weighted gaze — shifting, context-dependent, never fixed.
The Valley That Looks Like a Peak
Why a system can be stable and wrong at the same time. Local optima, loss landscapes, and the geometry of being stuck.
Something From Nothing
Complex behavior from simple rules. An agent that knows nothing — only moves, only collides, only accumulates. Emergence before learning.
The Table That Learns
Q-learning: an agent that begins to remember. Watch a policy crystallize from noise — one cell at a time, fully inspectable.
The Wall
Grow the maze and watch the table break. State space explosion — the moment when a perfect method hits a hard limit.
The Function
Same maze, same arrows — but the table is gone. A neural network approximates what was once explicit. Same output, different inside.
Looking Inside
Seven X-Ray moments. What we were actually doing. Look inside — not once, every time.