Storia: LLMs show a “highly unreliable” capacity to describe their own internal processes — Warptech Lab News