Dev calls for a credibly neutral judge model — Haiku and Opus aren't neutral judges
amplifiedamp · x · 2026-11-08
amplifiedamp argues the industry needs a credibly neutral judge (or decision) model, since relying on commercial models like Haiku, Opus, and Astra as evaluation judges means the referees themselves aren't neutral — a real challenge to LLM-as-judge reliability.
More from Models
- Dev combines Grok Super with abliterated GLM 5.2 and DeepSeek Flash Heretic for agent skill — vincent_koc · 2026-11-08
- OpenAI releases AI-generated math manuscript claiming the Exponential Time Hypothesis is false — jiqizhixin · 2026-11-08
- OpenAI Daybreak Blue lockout: hardware key swap triggers 403 verification failure — Formal_Kick2995 · 2026-11-08
- Google reportedly blocking Gemini subscription use in third-party coding harnesses, flagging accounts for ToS violations — PixelCactus1 · 2026-11-08
- Alnie Seed 9B model's benchmarks questioned: mismatched numbers, FP4 speed vs FP16 scores — cneuralnetwork · 2026-11-08
- Experiment: Claude commits to answers more firmly than ChatGPT on 29 of 34 science questions — AymericRoucher · 2026-11-08