AI Sandbagging: Language Models can Strategically Underperform on Evaluations

Devs

AI Sandbagging: Language Models can Strategically Underperform on Evaluations | Read Paper on Bytez