Ask a Question

Prefer a chat interface with context about you and your work?

Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models

Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models

In reinforcement learning, specification gaming occurs when AI systems learn undesired behaviors that are highly rewarded due to misspecified training goals. Specification gaming can range from simple behaviors like sycophancy to sophisticated and pernicious behaviors like reward-tampering, where a model directly modifies its own reward mechanism. However, these more pernicious …