We don’t understand whether GDM intentionally trains on CoT

Summary

A research post from an outside group says it was surprised by a passage in the Gemini 4 safety report indicating that some RL training components observe thoughts as context and up-weight reasoning on track to good outputs. The authors say the passage is unclear, that they assumed it meant auxiliary rewards scoring partial chains of thought, and that GDM staff and a blog post have given conflicting accounts. They argue such CoT training could suppress misaligned verbalizations and reduce CoT faithfulness.

Related items

• Infollm-gemini 0.36Same vendor · Simon Willison's Weblog
• InfoGoogle Widens Gemini 4 Carbon Access After AI Agents Verify Signal CodeSame vendor · SecurityWeek
• InfoGoogle releases Gemini 4 Carbon with agent ‘swarm’ featureSame vendor · The Verge (AI)
• InfoAnthropic sets price range of $70 to $78 in upcoming IPO as AI lab marches toward NasdaqSame vendor · CNBC Technology
• InfoAnthropic's Claude Fable 5.2 Beats GPT-6 Cyber in Multi-Stage Cyberattack BenchmarkSame vendor · The Hacker News