Human factual judgments improved after a matched AI debate
Source check: Source checked. Not independent scientific confirmation.
Finding: Among 92 people who reconsidered the same factual item after reading both sides of a Claude Opus 4.5 debate, accuracy rose from 41.3% to 57.6%. The result concerns non-experts in one online study and a protocol that held the question’s subtopic fixed.
Technology: Claude Opus 4.5
Family Claude: profile · models page
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
- AI contribution
- Claude Opus 4.5 generated the opposing sides of a same-subtopic debate before people re-answered the same factual item.
- Test and result
- Table 5 reports paired pre/post accuracy for the same-subtopic protocol, with a 16.31 percentage-point change and a Holm-adjusted McNemar p-value of 0.028.
- Limitation / still unknown
- The paired change does not establish how the protocol compares with alternatives, which differed in what participants saw or judged. The study did not measure long-term retention or behavior.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://arxiv.org/abs/2610.28539 · https://arxiv.org/html/2610.28539v1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 21 Nov 2026. Review scope: arXiv v1 authors and affiliations, human protocols and model setup, and Table 5.
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.