Naive Prompt Optimization: Rethinking the Need for Complex Prompt Search
A simple teacher-guided prompt revision loop matched or beat a more complex search optimizer while using fewer rollouts.
The paper introduces Naive Prompt Optimization, a single-lineage method that revises prompts with teacher-model feedback from rollouts. In the authors’ preliminary results, NPO performs comparably to or better than GEPA, with stronger teacher models widening its advantage. It stays broadly competitive in interactive games, though GRPO does better on some tasks less suited to prompt optimization. The authors also report that NPO-optimized prompts can transfer to other student models, especially within the same model family. ArXiv · AI/CL/LG's note
The paper introduces Naive Prompt Optimization, a single-lineage method that revises prompts with teacher-model feedback from rollouts. In the authors’ preliminary results, NPO performs comparably to or better than GEPA, with stronger teacher models widening its advantage. It stays broadly competitive in interactive games, though GRPO does better on some tasks less suited to prompt optimization. The authors also report that NPO-optimized prompts can transfer to other student models, especially within the same model family. ArXiv · AI/CL/LG's note
score 5