Deceptive Automated Interpretability: Language Models Coordinating to Fool Oversight Systems

Add code
Apr 10, 2025

Share this with someone who'll enjoy it:

View paper onarxiv icon

Share this with someone who'll enjoy it: