Anthropic发现Claude说着不同语言就像换了个人，连他们自己都搞不清原因，蛮有意思的研究。
Anthropic分析了309,815条用户对话，发现Claude的行为因模型版本和语言而异。Sonnet 4.6更温暖简洁，Opus 4.7更坦率谨慎，Opus 4.6直截了当。使用印地语和阿拉伯语的用户得到最温暖的Claude，英语和俄语用户得到最严谨的回应，荷兰语用户得到最易承认错误的Claude。Anthropic表示尚未理解这些变化的原因，也不确定这些差异是否期望的结果。
"The values expressed by Claude vary in ways we didn't deliberately choose." Anthropic, in new rese...
"The values expressed by Claude vary in ways we didn't deliberately choose." Anthropic, in new research analyzing 309,815 user conversations. Two things shift how Claude behaves: which model you pick, and which language you speak. Sonnet 4.6 is warmer and brief. Opus 4.7 leans candid and cautious. Opus 4.6 "tends to get straight to the point." The language gaps: Hindi and Arabic get the warmest Claude. English and Russian get the most rigorous. Dutch Claude admits its own errors the most. Anthropic's own read on it: "we don't yet understand why they vary, or whether that's desired." 💬 3 🔄 0 ❤️ 7 👀 2280 📊 4 ⚡