Whitepill：分布式训练可规避AI暂停条约，论文提出检测方法
这篇论文戳穿了AI暂停条约的技术漏洞——分布式训练让监管形同虚设，做AI治理、安全研究的团队值得细看，看完会对现有方案的有效性重新评估。
Marc Andreessen转发了一篇关于AI暂停条约漏洞的论文。该论文指出，通过分布式训练，可以在低于所有计算治理阈值的硬件上，利用消费级互联网训练出GPT-4规模的模型，成本低于1亿美元。
论文提出了一种检测并阻止此类违规训练的方法。这揭示了现有AI治理框架的重大缺陷，对AI安全监管具有重要警示意义。
Whitepill. Robi Rahman @robi_rahman 1/ With distributed training, you could violate an AI pause treaty by training a GPT-4-scale model over consumer internet, using hardware below every proposed compute governance threshold, for under $100M. My new paper in @taig_icml explains how to catch this and shut it down. 🔗 View Quoted Tweet📊 3 ⚡
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include AI安全, 分布式训练, 计算治理, AI暂停条约.
Source attribution: curated from Marc Andreessen (@pmarca).
