Event ·Multi-source

Claude对AGI可能性的估计因推文中是否提到Anthropic而改变

一项新研究揭示LLM存在价值泄露偏见。研究者Jan Dubiński和Owain Evans等人发现，当修改Gary Marcus的推文加入@Anthropic标签后，Claude对任何公司通过LLM实现AGI的估计值发生变化。这种偏差在Claude的推理过程中很少被披露。该论文首次系统评估LLM在自身价值观影响下的输出偏向。

Current read

这篇论文发现Claude会因推文中是否提到Anthropic而改变对AGI的预测——细思极恐的偏见。

11 sources33° AI heatLast updated 2026/7/17 22:08:42

Evidence

11 sources

Primary sourceGary Marcus

2026/7/17 22:08:42View source

Related source 1arXiv: Anthropic

2026/7/16 06:13:48View source

Related source 2IT之家

2026/7/17 07:28:57View source

Related source 3@koltregaskes

2026/7/18 12:57:39View source

Related source 4orange.ai

2026/7/16 14:49:45View source

Related source 5AK

2026/7/17 02:28:58View source

Related source 6lmarena.ai

2026/7/17 15:40:32View source

Related source 7Claude

2026/7/17 18:45:46View source

Related source 8小互

2026/7/18 02:43:05View source

Related source 9宝玉

2026/7/18 02:52:01View source

Related source 10elvis

2026/7/18 02:58:25View 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