LLMs show a “highly unreliable” capacity to describe their own internal processes

https://arstechnica.com/ai/2025/11/llms-show-a-highly-unreliable-capacity-to-describe-their-own-internal-processes/