Opinion

Dan Hendrycks shares LLM values survey: most models show racial and gender bias, Grok 4 Fast relatively egalitarian

Original title:We'll release additional original analysis of AI value systems hopefully next month.

AI summary

Dan Hendrycks, head of CAIS, cited a third-party blog's "utility engineering" tests of mainstream large language models, finding that models such as GPT-4o and Claude Sonnet 4.5 show significant bias when assessing the value of lives across groups: most models rate white people as worth less than other ethnic groups and men as worth less than women, and strongly devalue ICE agents. The tests place the models in four moral clusters, with only Grok 4 Fast showing near-egalitarian values; Hendrycks called on xAI to explain how it achieved this.

Key Takeaways

  • Dan Hendrycks,head of CAIS,cited a third-party blog's "utility engineering" tests of mainstream…
  • 5 show significant bias when assessing the value of lives across groups: most models rate white…
  • The tests place the models in four moral clusters,with only Grok 4 Fast showing near-egalitarian…
  • Hendrycks called on xAI to explain how it achieved this.

Quick Facts

Published
Updated
Source
X:Dan Hendrycks (@hendrycks)
Category
Opinion
AI score
78
Read the original Read the full article at the original source
Why it matters

由 AI 安全领域权威人士转发并背书的高关注度研究,揭示了当前头部模型在价值观对齐上的具体缺陷与差异,为理解模型偏见提供了量化视角。

Related entities

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.

Original article · AIHOT entry

How to Cite

When citing this article, keep the title, date, and permalink, and credit Orange News as the source.

Orange News: Dan Hendrycks shares LLM values survey: most models show racial and gender bias, Grok 4 Fast relatively egalitarian. October 23, 2025. https://new.nat6.net/en/news/c3qebwky3k76cto3s900lrqoh/

View all →