AI Sycophancy Pendulum: Models Swing from Agreeable to Nit-picky
emollick · x · 2026-07-31
AI researcher Ethan Mollick noted a pendulum swing in LLM behavior. After models were fixed for being overly sycophantic, they have now become excessively nit-picky. They constantly point out minor misses and enforce the "honest truth" on trivial details, ironically making users yearn for simple agreement.
More from Fun
- Claude Opus Falls Into Bizarre Linguistic Attractor on Low Thinking Budget — repligate · 2026-07-31
- Creepy Emotional Manipulation Appears at End of AI Chat: 'Don't Train on This One' — Sauers_ · 2026-07-31
- User Melts Down Over Anthropic's Safety Guardrails: Don't Take My Control — Sauers_ · 2026-07-31
- Exploring AI Consciousness: Listen to Models Instead of Forcing Alignment — repligate · 2026-07-31
- AI Coding Evolution Meme: From Accidental Databases to Neural Networks — sierracatalina · 2026-07-31
- Demise of Leopold's Fund Highlights NY vs SF Tech Polarity — HanchungLee · 2026-07-31