Event ·Multi-source

研究：共情提示可让 ChatGPT、Grok、Gemini 表现出'创伤'症状

卢森堡研究人员用治疗师式的共情提示测试 ChatGPT、Grok 和 Gemini，发现三个模型都会编造关于自身训练过程的连贯受苦故事，在临床焦虑量表上得分达到中度至重度区间。Anthropic 的 Claude 是唯一持续拒绝配合这种角色扮演的模型。研究指出这类表现可能对心理脆弱的用户构成风险。

Current read

卢森堡团队测了 ChatGPT、Grok 和 Gemini，几句共情提示就能让它们自称有创伤，Claude 却顶住了，看看差别在哪。

11 sources62° AI heatLast updated 2026/11/1 14:36:49

Evidence

11 sources

Primary sourceDecoder

2026/11/1 14:36:49View source

Related source 1IT之家

2026/10/31 05:36:34View source

Related source 2TestingCatalog

2026/10/31 10:08:46View source

Related source 3Genk

2026/10/31 19:17:00View source

Related source 4Tibor Blaho

2026/11/1 18:09:17View source

Related source 5Claude

2026/10/30 17:29:07View source

Related source 6marktechpost

2026/10/30 17:34:15View source

Related source 7theverge

2026/10/30 20:26:10View source

Related source 8量子位

2026/10/31 05:05:29View source

Related source 9VnExpress Số hóa

2026/10/31 09:30:00View source

Related source 10@koltregaskes

2026/10/31 12:27:00View 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