DigestAI news desk
Research updated

CLEAR framework improves medical LLM accuracy via cross-source evidence adjudication

Researchers have introduced CLEAR, an agentic framework designed to resolve conflicts in medical information processed by large language models. While retrieval-augmented generation (RAG) allows LLMs to access current data, retrieved sources often contain irrelevant or contradictory information, which can degrade the factual accuracy of model outputs. CLEAR addresses this by independently…

1 source primary source

Key points

  • CLEAR is an agentic framework for adjudicating conflicting medical evidence in LLMs.
  • It generates answers from parametric, curated, and retrieved sources before verification.
  • Unresolved conflicts trigger targeted follow-up searches to ensure factual accuracy.

The framework employs an aggregation verifier to jointly evaluate these candidates, analyzing supporting evidence, provenance, and source quality to identify areas of agreement or conflict. An adjudication module then determines whether to preserve the initial conclusion or revise it using override-guard and challenge-audit mechanisms. If conflicts remain unresolved, the system triggers targeted follow-up searches and re-adjudication processes. This approach aims to ensure that medical LLMs provide reliable, evidence-grounded responses despite the inherent instability of external data sources.

Read the original at arXiv cs.AI · by Shuai Wang, Yize Zhao, Qingyu Chen primary source Open source ↗

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us.

Comments

via GitHub Discussions

More in Research

All →

Related stories