DUET 论文：让 LLM 评分器与智能体同步联合优化
一篇讲怎么同时优化智能体和评分器的论文，DUET 在 Finance Agent Benchmark 拿到 96.5%，比 GEPA 和 SkillOpt 都高，做法很实在。
DUET 方法提出在训练智能体的同时更新其 LLM 评分器，因为智能体改进后失败模式会变化。它交替固定一方、用工具调用更新器修订另一方：每一轮选取任务、运行 solver、由 grader 打分并批判轨迹。在 Finance Agent Benchmark 上 solver 达到 96.5%，超过 GEPA 的 94.4% 和 SkillOpt 的 94.0%。在 GDPval 上 grader 一致率从 68.4% 升至 78.5%，而冻结 solver 只训练 grader 时仅从 67.9% 升至 68.8%。
Nice paper on optimizing agents and their graders together.
The key finding is that you want to update your LLM grader alongside the agent instead of keeping it fixed, because the agent's failure modes change as it improves.
DUET alternates between the two.
Each round picks training tasks, runs the solver, has the grader score and critique the trajectories, then a tool-using updater revises either the grader or the solver while the other stays fixed.
On the Finance Agent Benchmark, the solver reaches 96.5%, above GEPA (94.4%) and SkillOpt (94.0%). It also beats both when they optimize against the benchmark's own reference evaluator.
Grader agreement rises from 68.4% to 78.5% on GDPval. Training the grader with the solver frozen moves agreement only from 67.9% to 68.8%.
Paper: https://t.co/qmZBl4RHu5
- Artificial Analysis10-12 17:25原文
