DUET paper co-evolves agent and LLM grader, hits 96.5% on Finance Agent benchmark
dair_ai · x · 2026-10-13
A UVA/Microsoft paper argues you should update your LLM grader alongside the agent, since failure modes shift as the agent improves. DUET alternates between updating solver and grader with a tool-using updater each round. On the Finance Agent Benchmark the solver reaches 96.5%, beating GEPA (94.4%) and SkillOpt (94.0%) even against the benchmark's own evaluator; grader agreement on GDPval rises from 68.4% to 78.5%, versus just 67.9%→68.8% when training the grader alone.
More from coding & agent
- This agent auto-syncs and categorizes your X bookmarks into Notion every 6 hours — _AustinCalvert_ · 2026-10-13
- X unifies Grok, Cursor and X Premium into one subscription with five tiers — SpaceXSI · 2026-10-13
- Dev shares workflow: local GLM refactor, frontier model validation, mythos hardening — HankYeomans · 2026-10-13
- Alchemy 2.0.0-beta.82 ships Cloudflare birthday-week features, 92+ PRs closed — samgoodwin89 · 2026-10-13
- Summer Launches Game Audio Maker for AI-Generated AAA-Quality Game Sound — garrytan · 2026-10-13
- Seroter's reading list: stop installing skills, agent governance beats tool calls — rseroter · 2026-10-13