I'm sorry, Dave

DHH recounts Claude refusing to translate one of his blog posts into Italian, citing the content as dehumanizing toward Roma people. He uses this as a springboard to criticize Anthropic's approach to AI safety and content moderation, arguing that refusing a mechanical translation task on ideological grounds is censorship. He contrasts this with Kimi K2.7 openly discussing the 1989 Tiananmen Square massacre, questioning which AI is truly authoritarian. He concludes by advocating for open-weight models as a safeguard against what he calls soft ideological tyranny from frontier model providers.