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.
More from AGI Musings
- Jane Street's yminsky: agents are bad at software design — but does design still matter? — charliermarsh · 2026-11-17
- WSJ credits Gemini with diagnosing what dozens of doctors missed — Redditors question the framing — Aaasteve · 2026-11-17
- Scientist: AI could deliver 100-1000x faster science by the 2030s — DeryaTR_ · 2026-11-17
- Critic Says Anthropic's IPO Filing Describes Loss-of-Control Risk Like Elizabeth Holmes — zetalyrae · 2026-11-17
- Dev's sharp metaphor: cyberattacks and outages are AI's paraneoplastic syndrome — zetalyrae · 2026-11-17
- Dario Amodei Warns Open Models Lead to 'Very Dangerous Place,' Bill Gurley Calls It Lobbying — kevinnbass · 2026-11-17