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
- LLM one-shots a Minecraft 3x3 piston door by building its own Linux test infra — Vjeux · 2026-10-18
- Dev builds Mumbai-inspired 3D manga background generator from OpenStreetMap data — CurieuxExplorer · 2026-10-18
- Solo founder builds a money-printing data service on Claude's Pi Durable harness — MicahBerkley · 2026-10-18
- Dev builds Arch-based OS with custom desktop and browser using Claude Code — Snoo-10493 · 2026-10-18
- Five SaaS agent workflows fail the same way — and it's never model intelligence — usehive · 2026-10-18
- Claude skill generates hand-drawn interactive diagrams that teach kids how things work — CatAstro_Piyush · 2026-10-18