Luminote is a local AI interpretability prototype that gives language models a prosthetic face and body. It measures activation-derived affect estimates from a model’s hidden states, maps them into a valence-and-arousal space, and translates the result into color, position, and reaction imagery.

Luminote does not claim that machines have feelings. It makes hidden-state patterns visible and intuitive, creating an optional nonverbal communication layer for human-AI interaction.

Built With

Share this project:

Updates