Visible chains of thought are a safety advantage for AI, but that transparency is slipping away
Chain-of-thought reasoning offers safety benefits, but DeepMind warns transparency may erode. DeepMind notes that while models can articulate their reasoning steps, the practice is slipping as