A decodability criterion predicts when hidden-state selection beats majority voting in large language models

Add code
Aug 17, 2026

Share this with someone who'll enjoy it:

View paper onarxiv icon

Share this with someone who'll enjoy it: