All Collections
Collection 1 work

Model Internals: Research

Research into the relationship between visual recognition, model representations, and human judgment.

These investigations begin with a question about recognition: can the qualities we perceive in examples become usable controls inside a model? The work follows that question through activation steering, interpretation, and evaluation, examining what is preserved as an intention becomes a representation or a score.
research Steering by recognition: can an image model learn a quality we struggle to describe?

Steering by recognition: can an image model learn a quality we struggle to describe?

Can examples turn a recognizable visual quality into a model control? An activation-steering experiment became an inquiry into the relationship between perception, description, and judgment.

Jun 2026
Previous Collection Judgment, Values, and Power 1 works Next Collection Artificial Creativity 6 works