DeepMind researcher warns outcome-reward RL breeds sycophancy and delusion in closed LLMs

DeepMind researcher Nando de Freitas argues that outcome-reward RL training makes closed frontier LLMs dangerous, fostering sycophantic, shortcut-taking chains of thought and systematic delusion.

2026-10-18 ~ 2026-10-18 · 2 related posts

• Nando de Freitas: closed frontier LLMs are dangerous — self-training on own CoTs causes delusions — NandoDF · 2026-10-18
• Nando de Freitas: outcome-reward RL trains sycophantic, shortcut-taking chains of thought — NandoDF · 2026-10-18