Y Combinator CEO Garry Tan: U.S. Open-Weight AI Labs Should Distill Models
Garry Tan argues that U.S. open-weight AI labs should use distillation techniques from American frontier modelers.
Y Combinator CEO Garry Tan is declining to join calls for regulations against Chinese labs' AI distillation activities, instead wanting U.S. open-weight AI labs to use similar techniques.
Distillation Statement from Y Combinator CEO
Y Combinator CEO Garry Tan stated that regulators should not intervene regarding Chinese AI labs using knowledge distillation techniques. Tan expressed that U.S. AI labs can also follow the same method.
Distillation Method and Anthropic Report
Distillation means a model producer heavily prompting another model to learn how it works and reasons. Anthropic had published a report claiming that Chinese labs conducted unauthorized distillation attacks using hidden identities and stolen credentials.
Emphasis on Open Source and Copyrights
Tan stated that he is not advocating for U.S. AI labs to steal, but rather that this process should be done through legal means. He pointed out that proprietary labs have used copyrighted materials without permission during their training processes.
Historic Fear of Monopolization
Stating that the centralization of AI models in the hands of a single powerful company is the ultimate nightmare scenario, Garry Tan emphasized that open-weight models should provide freedom and access to users.