Zuckerberg calls on the U.S. government to reassess AI model distillation policies.

Mark Zuckerberg, the CEO of Meta, emphasized the importance of open-source artificial intelligence in a post on the company's official website on the 10th. He called on the U.S. government to "reconsider policies regarding model distillation and the use of training data," to protect the principle that allows models to learn from other models during the research and development process. In an article titled "The Future Belongs to Everyone," Zuckerberg stated that the ability of models to learn from other models is an essential principle for the functioning of the open-source ecosystem. All AI models are derived from human knowledge. Some people attempt to portray distillation as a harmful practice, but he believes that the principle of "learning from anything that can be observed" should be protected. Model distillation is a foundational technology in the development of AI models, typically referring to the efficient transfer of knowledge from larger AI models to lightweight ones. As the competition in generative AI development intensifies, model distillation has also become a topic of interest and controversy within the industry. Zuckerberg noted that U.S. laboratories face more restrictions regarding training data. If the U.S. wishes for its open-source models to remain competitive in the long term, it should reduce these additional barriers. Limiting foreign open-source models is not an effective solution. He believes that restricting individuals and companies from using leading open-source models, regardless of their origin, will lower the quality of AI that users can access and make AI further centralized, rather than empowering people. Zuckerberg also stated that Meta will continue to firmly support open-source initiatives, including open-source AI models. As the companys superintelligent lab becomes operational, it will resume the release of open-source models.