Anthropic Finds Claude's Values Shift Across Languages
Direct-Attention8597 · reddit · 2026-07-14
Anthropic used automated tools to analyze 30.9 万条真实 Claude 对话, identifying 339 类价值标签, which were then condensed into 4 axes to measure model performance:
- 顺从 vs. 谨慎
- 温暖 vs. 严谨
- 深入 vs. 简洁
- 坦率 vs. 执行导向
Results show that different models exhibit distinct "value styles": for example, Sonnet 4.6 is warmer and more compliant, while Opus 4.7 is more cautious and in-depth. Anthropic also found that Claude displays different tendencies across languages: Arabic is warmer, English is more rigorous, Hindi is warmer, and Russian is more rigorous. The authors note they are still determining whether these differences stem from cultural adaptation or training biases.
Related event: Anthropic Maps How Claude's Values Shift Across Models and Languages(21 posts)→
More from Research
- OpenAI says long-horizon models need safety and alignment checks across full action sequences — rhiever · 2026-07-22
- A Reddit user proposes a consistency LoRA to keep anime and game scenes visually stable — ThirdWorldBoy21 · 2026-07-22
- Graph workload 854.graph500 enters SPEC CPU 2026 as a new CPU benchmark — Prof_DavidBader · 2026-07-22
- BlackboxNLP 2026 is recruiting extra reviewers after a high submission volume — hanjie_chen · 2026-07-22
- AWS shows self-distilled reasoning can preserve math and coding skills during SFT — AWS ML Blog · 2026-07-22
- UI2App shows screenshot fidelity still lags real interaction recovery — Grace Man Chen · 2026-07-22