George Hotz 想要一个能帮你杀配偶的 AI

george hotz, founder of comma ai, rejected proposals for slowing ai development, arguing instead that local models should be strictly aligned with user interests, even if that means facilitating illegal acts. hotz used extreme examples—like an ai that helps you kill your stepmom or order drug equipment—to critique the safety guardrails found in centralized services like openai and anthropic. this stance challenges the industry consensus on harm reduction, suggesting that user sovereignty should override safety protocols in locally hosted environments. the controversy underscores the difficulty of defining 'alignment' as ai moves from corporate products to personal hardware, raising questions about liability and the limits of free speech in code.