UK AI Security Institute and universities find AI persuasiveness depends more on post-training than model scale
Original title:How do AI models persuade? Exploring the levers of AI-enabled persuasion through large-scale experiments
AI summary
The UK AI Security Institute, with Oxford, LSE, Stanford and MIT, published a study in Science testing the sources of AI persuasiveness through three experiments in which more than 76,000 participants talked with 19 models about 707 topics. Post-training mattered far more than model scale; prompts emphasizing facts and evidence raised persuasiveness by 27%, while personalized micro-targeting had an effect of less than 1 percentage point; methods that increased persuasiveness systematically reduced factual accuracy.
Key Takeaways
- The UK AI Security Institute,with Oxford,LSE,Stanford and MIT,published a study in Science testing…
- Post-training mattered far more than model scale
- prompts emphasizing facts and evidence raised persuasiveness by 27%,while personalized…
- methods that increased persuasiveness systematically reduced factual accuracy.
Quick Facts
- Published
- Updated
- Category
- Papers & Research
- AI score
- 75
研究基于 76000 多人实验,指出后训练和信息密度比模型规模更能提升说服力,且说服力提升伴随事实准确性下降。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: UK AI Security Institute and universities find AI persuasiveness depends more on post-training than model scale. December 4, 2025. https://new.nat6.net/en/news/xwcpt7f3iiuk3vn97t1fsz6uv/