In the news
Verifiable by Construction: Claim-Level Evaluation of Verbatim Citation in Clinical Question Answering
arXiv cs.AI · Published · 3 min read
In 30 seconds
- What happened
- Researchers evaluated whether LLMs can answer clinical questions with verbatim citations that fully substantiate each claim, testing twelve models on 222 questions.
- Why it matters
- Clinical engineers building QA systems need this when designing verification workflows that let busy clinicians validate answers without opening source documents.
- Watch out
- Most models attach quotes to 90% of claims but only substantiate 30-40% fully. Lightweight models like Claude Haiku struggle significantly with this task.
- llm
- language model
- eval
The patterns behind this
Each one covers how the technique works, when it earns its cost, and where it breaks.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.