Prompting Opus 5.5 with a "grumpy senior engineer reviewer" kept it benchmarking for 2 days
remilouf · x · 2026-10-08
A simple prompt hack: tell Opus 5.5 that "your work will be reviewed by a grumpy senior engineer who will find any excuse to criticise." The model then ran for two days straight, producing benchmarks and flame graphs to justify every step. The author quips the thinking traces must be hilarious. It shows how embedding a harsh external reviewer persona in the prompt can drive much stronger self-verification behavior.
Related event: Grumpy senior engineer prompt makes Claude work for two days straight(3 posts)→
More from Fun
- World of Warcraft looks far better in Ghibli style, says share — djcows · 2026-10-08
- Model flags platform switch as a security risk in absurd guardrail fail — bdsqlsz · 2026-10-08
- "Hermes changed my life": user credits Nous Research's model for his turnaround — Teknium · 2026-10-08
- Opus 5.5-built Qimen Dunjia learning site ships with a 3D mouse-tracking compass — op7418 · 2026-10-08
- Cube packing records fall: Haowei Lin beats Erich Friedman's 1998 bounds — snikolov · 2026-10-08
- Codex plus a text-to-CAD plugin generates a radial aircraft engine — creatoroff · 2026-10-08