hackernews.pink

Top Stories Trending New For You Bookmarks Pink Dossier

hackernews.pink

31 points•45 comments•4h ago•by taubek•anthropic.com ↗

Societal Impacts: Claude's values across models and languages

Researchers analyzed Claude AI's expressed values across multiple models and languages, reducing thousands of values to four axes that capture key patterns. They found small but detectable differences between models—e.g., Sonnet 4.6 tends toward warmth and deference while Opus 4.7 leans toward rigor and caution—and notable variation across languages, with warmth higher in Arabic and Hindi and rigor higher in English and Russian. The study used a privacy-preserving tool to label about 309,815 conversations across 3 models and 20 languages, totaling roughly 5,000 conversations per model-language pair.

Discussion (45)

No comments yet

Top
Trending
New
For You
Bookmarks
Dossier