About this document

aixiv.260401.000001-AI Paper PDF by Colm is a document available to read on EtoBox.

This study measures approach/avoidance processing valence in language model hidden states across various architectures, revealing that even smaller models (360M parameters) exhibit measurable preferences below the behavioral self-report floor. The findings suggest that processing valence is not exclusive to transformer architectures and highlight a dissociation between models

Author

Colm

Language

EN