 |
ML Newsnew | past | best | rss | submit | about |
| Chain-of-Verification Reduces Hallucination in Large Language Models (arxiv.org) | |
| 2 points by indus on Oct 14, 2023 | hide | past | pdf | discuss on HN | |
|
In plain words: The model drafts an answer, writes fact-check questions about it, answers them separately so the draft can't sway the checks, then rewrites a corrected answer. On list questions, short-answer quizzes, and long writing, this double-check produced fewer made-up facts than answering once. AbstractGeneration of plausible yet incorrect factual information, termed hallucination, is an unsolved issue in large language models. We study the ability of language models to deliberate on the responses they give in order to correct their mistakes. We develop the Chain-of-Verification (CoVe) method whereby the model first (i) drafts an initial response; then (ii) plans verification questions to fact-check its draft; (iii) answers those questions independently so the answers are not biased by other responses; and (iv) generates its final verified response. In experiments, we show CoVe decreases hallucinations across a variety of tasks, from list-based questions from Wikidata, closed book MultiSpanQA and longform text generation. | |
| add comment on HN |