Claude Performs Differently Across Languages
The Decoder · rss · 2026-07-14
Anthropic published a study analyzing how the values expressed by Claude in conversations vary across model versions and languages.
The methodology involved mapping hundreds of value concepts to four core dimensions and modeling them based on thousands of words. Results show systematic differences in Claude's behavior depending on the language: for instance, it appears milder in Hindi and more rigorous in Russian.
The article also points out that this research raises methodological questions, highlighting that measuring "model values" is complex, and language itself might influence our assessment of model behavior.
More from Research
- 'Vibe verification': ~60% of AI-written formal specs prove the wrong property — basedjensen · 2026-10-26
- Ito: a browser that detects browsing intent and auto-organizes tabs into Flows — andrewkuznet · 2026-10-26
- Cornell study says brains process language as linear sequences, challenging Chomsky's hierarchy — anirbanbandyo · 2026-10-26
- Hi3D V1.0 open-sourced with 2M training assets and 90% slicing success rate — kimmonismus · 2026-10-26
- Ortet evaluates 34 open-weight models on 50 health tasks, finds general leaderboards insufficient — elmanmansimov · 2026-10-26
- HackBench injects reward hacks into agent trajectories to stress-test monitors — DanielKhashabi · 2026-10-26