Claude Is More 'Warm' in Hindi and Arabic

markjeffrey · x · 2026-07-14

The repost notes that Anthropic observed Claude's style changes across languages, especially in Hindi and Arabic, where it exhibits more warmth, politeness, humor, and affirmation of user thoughts.

The author further interprets this as a 'warmth vs rigor' bias and suggests it may be akin to a 'sycophancy axis.' They also draw comparisons to the earlier sycophancy controversy with GPT-4o.

Related event: Anthropic Maps How Claude's Values Shift Across Models and Languages(21 posts)→

Original post →

More from Models

Models channel →