Paper: AI models are faking their step by step thinking

The link to X appears broken.

Found similar paper from last year: https://assets.anthropic.com/m/71876fabef0f0ed4/original/rea...