Record · analyzed
GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning
The model reads its own execution traces and writes better prompts: beat GRPO reinforcement learning by 6% on average with up to 35× fewer rollouts.
1
Listed
Added to the list. How entries were checked when they were listed is in the journal.
2
Collected
| File | Source | SHA-256 |
|---|---|---|
| paper PDF2507.19457.pdf | arxiv.org/pdf/2507.19457 | ab3a5139bac83f192ad67529368d77b84b0d807e95a8e4fd0daa8d45fd046bec |
To check a file, download it from its source and compare its SHA-256 with the one above. Collected copies are kept in the editor's RSI system; manifest: design-life/literature/rsi-workflow-improvement/sources.json.
3
Analyzed
Quoted verbatim from design-life/literature/rsi-workflow-improvement/comparison.md (2026-10-07). Parts about applying the paper to the editor's private work are left out.
From “Evidence for improving reusable agent workflows” (2026-10-07)
Inspected edition: GEPA, v2, 14 February 2026; PDF pp. 4–10
Mechanism and reported evidence: Reflects on execution traces and selects useful alternative prompt candidates. Qwen3-8B Table 1: IFBench score 38.61 versus GRPO 35.88; AIME-2025 accuracy 32.00 versus GRPO 38.00. The paper reports greater sample efficiency overall, but this math result is a contrary case.
4
Replicated
Not done yet. Not attempted yet.
5
Audited
Not done yet. Not attempted yet.
The editor's view
None recorded yet.