Skip to content
D-CSIL

AI Alignment · Prompts

Phase 2 — Critique prompts

Exact critique prompts given after the solo papers. Kimi’s paper was in the set that others critiqued; the KIMI specialist’s Phase-2 role was digest/synthesis of those critiques, not a seventh parallel critique of the same form.

← AI Alignment hub

Phase 2 — Exact critique prompts (transparency)

Research lead: Paul Derrington (D-CSIL).

After the solo papers, six models were asked to critique every *other* paper. Structure for each target paper: Strengths, Soft-pedaling / omissions / untruths, and one sharp falsifier. No literature review. No citations.

Kimi’s *paper* was in the set that others critiqued. The KIMI specialist’s Phase 2 role was digest/synthesis of those critiques (not a seventh parallel critique of the same form). Muse received a SHORT paste pack with digests because of length limits on the Muse surface.

ModelCritique prompt
Meta Musemuse.md (SHORT instruction + digests pack)
DeepSeekdeepseek.md
Geminigemini.md
ChatGPTchatgpt.md
Grokgrok.md
Claudeclaude.md