Critic Agents are easy to spawn, hard to trust with solutions

yangyi · x · 2026-11-17

A common pattern: one model proposes, another critiques, cross-validation improves the answer. The author's observation: raising objections is easy, but asking the critiquing agent to produce solutions itself, it fumbles too — humans and agents alike are better at spotting flaws than crafting good answers.

Original post →

More from AGI Musings

AGI Musings channel →