Y Combinator首席执行官加里·谭(Garry Tan)近日表示，美国应建立自己的模型蒸馏机制，允许本土开源AI实验室从前沿模型中提取知识。1谭在接受采访时表示，对于中国AI实验室使用蒸馏技术的问题，他的立场是"什么都不做"。1他认为，控制用户对闭源模型API调用的使用属于过度约束，主张平衡前沿AI实验室与开源模型的发展。1
谭的这一观点反映了他对AI行业竞争格局的看法。1他指出，AI领域最坏的情景是由单一公司垄断。1谭认为，这种蒸馏机制可以为美国提供更多健壮的开源模型选项，有助于形成更健康的市场竞争。与此同时，人工智能安全公司Anthropic本周发布的第二份报告指控中国实验室从事"非法蒸馏攻击"，这与谭的政策主张形成了对比。1
Y Combinator CEO Garry Tan has advocated for American policymakers to adopt a permissive stance toward model distillation, arguing that U.S. open-weight AI laboratories should be permitted to extract knowledge from frontier models just as Chinese labs do.1 In an interview with CNBC, Tan stated "I would do nothing" regarding regulatory restrictions on distillation practices.1 He proposed establishing an "American distillation regime" to provide the country with more robust alternatives to Chinese open-source models.1
Tan's position reflects concerns about competitive balance in AI development. He criticized existing constraints on how users and customers employ API calls to closed-weight models, stating "Controlling what users and customers do with API calls to closed weight models feels constraining."1 According to Tan, "The nightmare scenario, the doomer scenario for AI is that there's just one company," underscoring his view that fostering a diverse ecosystem of models and developers is essential to preventing monopolistic control.1 His remarks come as Anthropic released its second report this week alleging that Chinese laboratories are engaging in "illicit distillation attacks."1
评论
还没有评论，欢迎留下第一条。