Paper probes misaligned thinking in language models

Arxiv CS.AI (top papers)

The paper explores detection of strategic deception, sandbagging, and self-preservation in LLMs. It aims to improve reliability in high-stakes deployments.

LLMAlignment

Arxiv CS.AI (top papers)·Kaiwen ZhouConstantin VenhoffJonathan MichalaXin Eric WangWilliam Saunders

23 hours ago