Chain-of-Verification Reduces Hallucination in Large Language Models
A verification workflow in which a model drafts, plans fact-checking questions, answers them independently, and writes a revised response.
Authors or creators: Shehzaad Dhuliawala, Mojtaba Komeili, Jing Xu, Roberta Raileanu, Xian Li, Asli Celikyilmaz, Jason Weston
Checked: 2026-08-01
Source note: Primary peer-reviewed conference research. Reported reductions apply to the tasks and model settings evaluated in the paper.
Claims connected to this source
- Chain-of-Verification reduced hallucinations across list questions, closed-book QA, and long-form generation experiments.
