Sycophancy to subterfuge: Investigating reward-tampering in large language models

Best AI papers explained - Sycophancy to subterfuge: Investigating reward-tampering in large language models