Better with Experience: Self-Evolving LLM Agents for Evidence-Grounded Health Community Notes
Zihang Fu, Fanxiao Li, Jianyang Gu, Haonan Wang, Preslav Nakov, Bryan Hooi, Min-Yen Kan, Jiaying Wu · 2026
Read Better with Experience: Self-Evolving LLM Agents for Evidence-Grounded Health Community Notes by Zihang Fu, Fanxiao Li, Jianyang Gu, Haonan Wang, Preslav Nakov, Bryan Hooi, Min-Yen Kan, Jiaying Wu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-02215.txt.
Description
Fu and co-authors let health fact-check notes learn from past corrections. Community Notes can correct health misinformation at scale. Chat model help makes those notes timely and evidence based. But each new post starts the process from scratch. Useful experience from prior cases goes unused that way.
Their EvoNote framework builds an evolving memory of past episodes. Fine-grained credit assignment is the core of the design. Feedback at the trajectory level is grounded in note quality traits. It is distilled into action level memory for later use. Steps cover claim analysis, evidence gathering, and note writing. Tests use MM-HealthCN, a benchmark of 1.2K flagged health posts. Human notes and crowd helpfulness labels come with it. Self-evolving correction is the goal of the system.
Catalog record
| CREATOR | Zihang Fu, Fanxiao Li, Jianyang Gu, Haonan Wang, Preslav Nakov, Bryan Hooi, Min-Yen Kan, Jiaying Wu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 100 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.02215