Mind Map — Neural Attention Visualizer

THE PROBLEM

Every AI language model — GPT, Claude, Gemini — uses a mechanism called attention to understand language. It looks at every word in a sentence simultaneously and decides which other words are relevant to understanding each one. This mechanism runs billions of times per second inside every AI response. Nobody has ever seen it. It is completely invisible.

THE APPROACH

A real language model (DistilBERT) processes any sentence you type and outputs attention weights — a number for every word pair representing how much one word attends to another. These weights are extracted and rendered as an arc diagram: click any word, and glowing lines connect it to every other word with thickness proportional to attention. This is not a simulation of how AI reads — it is the actual computed attention of a real model processing your exact sentence.

THE DEMO

WHAT I'M LEARNING