Announcement
Faithful LLMs reportedly decide and report using the same layers in one research setting
A researcher sharing a new COLM paper says unfaithful models did not use the same layers for both tasks.
TLDR
A researcher sharing the COLM paper “Identifying Introspection From the Inside” asks whether LLMs know what drives their choices when they describe their decisions. In the researchers’ setting, the post says, faithful models used the same layers to decide and report; unfaithful models did not.
Combined views
2.5K
2 Sources, first seen ago
