Prompt optimization may not need search trees: NPO paper shows teacher quality beats elaborate search
rohanpaul_ai · x · 2026-09-02
A new paper tests NPO, which keeps a single prompt lineage and has a teacher model revise it from recent rollout traces and rewards, against GEPA's multi-candidate Pareto selection. NPO matched or beat GEPA on IFBench and HotpotQA with slightly fewer rollouts (3,500 vs 3,593; 6,800 vs 6,871). The gap widened with stronger teachers—DeepSeek-V4-Flash and especially GPT-5.5—suggesting strong teacher reasoning plus rich feedback can replace optimizer-side search. Optimized prompts also transferred to other student models, especially within the same family.
More from coding & agent
- Simple trick to de-vibecode your app: feed the model a blacklist of AI-app cliches — mfckr_eth · 2026-09-02
- 'My 2T Token Life': Devs Show Off Agent Coding Token Guzzling With Tokscale — doodlestein · 2026-09-02
- Google Ships Gemini 3.8 Flash and Cyber Model for Agents and Code Security — GoogleDeepMind · 2026-09-02
- One Claude Code Skill Turns a 20-Shot Script Into 8 Finished AI Documentaries (33 Min) — illscience · 2026-09-02
- LangChain doubles down on model-agnostic stance: 'the right model today may not be right next week' — LangChain · 2026-09-02
- A useful categorization of the most common agent skill patterns — rseroter · 2026-09-02