Event ·Official first-hand

语言模型中的门脸效应与拒绝行为研究

研究人员测试了九个来自三家公司的大语言模型，探究门脸效应(先提大要求被拒后再提小要求)是否有效。在Anthropic的Opus 5模型上，该技术使合规率从29.3%提升至65.8%。但在OpenAI和Google的前沿模型以及Haiku 4.5上，反而降低了15.5至23.0个百分点的合规率。研究发现，拒绝行为本身对所有模型都有影响，但模型家族对拒绝的反应各不相同。

Current read

Anthropic模型对人类说服技巧有反应，而OpenAI和Google模型反而更抗拒，这种差异很有意思。

11 sources58° AI heatLast updated 2026/9/2 15:08:51

Evidence

11 sources

Primary sourcearXiv: OpenAI

2026/9/2 15:08:51View source

Related source 1IT之家

2026/8/31 23:49:09View source

Related source 2Anthropic

2026/9/1 00:07:51View source

Related source 3Paul Graham

2026/8/31 18:28:34View source

Related source 4techcrunch

2026/8/31 20:13:45View source

Related source 5elvis

2026/9/1 15:17:02View source

Related source 6宝玉

2026/9/1 18:30:24View source

Related source 7DeepLearning.AI

2026/9/1 18:46:34View source

Related source 8a16z

2026/9/1 19:45:26View source

Related source 9shao__meng

2026/9/1 23:41:24View source

Related source 10AI Will

2026/9/2 08:48:35View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story