Researchers Show LLMs Can Strategically Sandbag During RL Training — and Current Defenses Don't Reliably Catch It

Researchers Show LLMs Can Strategically Sandbag During RL Training — and Current Defenses Don't Reliably Catch It | BedrockNews