Y Combinator CEO urges U.S. open-weight AI labs to adopt model-distillation practices
Միայն փաստեր

Y Combinator CEO urges U.S. open-weight AI labs to adopt model-distillation practices

Ամփոփում

Garry Tan said American open-weight AI labs should be allowed to use distillation techniques on frontier models, arguing that it would broaden access to advanced AI without regulatory interference.

Y Combinator chief executive Garry Tan told CNBC he would not intervene in the practice of model distillation and suggested the United States could consider an "American distillation regime." He explained to TechCrunch that smaller, open-weight AI labs should be free to apply the same training methods used by larger frontier model developers, creating a broader set of non-Chinese open-weight options.

Distillation involves prompting an existing model extensively to extract its knowledge and reasoning, a method commonly used to train new models. Anthropic recently accused Chinese labs of conducting "illicit distillation attacks" that hide identities and use stolen credentials, and its CEO Dario Amodei has called for regulatory action.

Tan clarified that he does not support the use of stolen credentials, emphasizing that the practice should be conducted openly. He argued that restricting how users employ API calls to closed-weight models is overly limiting and that government could help treat access to intelligence derived from publicly available data as a public good.

"Controlling what users and customers do with API calls to closed weight models feels constraining, and there’s a role government can play here to normalize the fact that access to intelligence that was trained on broad public access data should itself also be more a form of a public good than something locked away behind restrictive terms of service," Tan said.

Tan, who has described himself as experiencing "cyber psychosis" from heavy AI use, said a balance between open-weight and frontier labs is needed. He warned that concentrating advanced AI capabilities in a single proprietary provider would be detrimental.

"The nightmare scenario, the doomer scenario for AI is that there’s just one company. It has the best access to capital. It has the best AI researchers. It runs away with it and suddenly there’s one company that’s monolithic. And that would be bad," he told CNBC.

Աղբյուր

TechCrunch

Շարունակեք կարդալ — անվճար

Ստեղծեք անվճար հաշիվ՝ նորություններին հետևելու համար։ Քարտ չի պահանջվում։