Vision: Multimodal AI

A multi-part architectural series tracing the journey from 16×16 pixel patches to unified multimodal reasoning: how machines learn to see, understand, and act.

What Happens When Computers Look at Pictures?

From a grid of numbers to a sentence that understands it: the beginning of a series on how AI learns to see.

A photograph of a green beetle among leaves, beside the same photograph cropped to a square and cut into a 14 by 14 grid.

What Does a Vision Transformer See?

A photograph cut into 196 squares, a list of a thousand words, and a squirrel the model took for a monkey.