News
Meta Zuckerberg: If the United States wants to take the lead in open source AI, it must reduce policy obstacles
2 min read
Source: ithome.com
IT House News on August 10, according to Reuters, Meta CEO Mark Zuckerberg called on the United States to reduce policy obstacles faced by open source AI in order to better compete with Chinese competitors. Earlier today, Meta announced the new open-weight model Muse Glimmer, with more models coming soon. Muse Glimmer is smaller and positioned differently than its competitors’ top AI models. The model can run agent tasks on a Mac or PC with only one graphics card, targeting the need to run AI directly on personal devices. In his 14-page article "The Future Belongs to Everyone," Zuckerberg called for making AI widely available rather than in the hands of a few. “The idea that AI is so dangerous that it is only safe if power is highly centralized is inherently problematic.” Meta is also preparing to disclose the weight of its most advanced model, Muse Spark 1.2. The model was developed by a super-intelligence team that invested heavily last year and has high hopes of catching up again in the AI race. Meta has long supported the open model and is one of the few American technology companies to put this concept into action. However, after the poor response to Llama 4 last year, Meta was forced to adjust its strategy. Zuckerberg also announced the establishment of a US$1 billion (IT Home Note: current exchange rate is approximately 6.752 billion yuan) fund to provide support to the community, hoping to alleviate external concerns about the impact of Meta's large-scale construction of data centers on local communities. He also admitted frankly that compared with countries such as China, a major disadvantage of the United States is that it is more difficult to build infrastructure. Zuckerberg also said that if American companies hope to lead in the field of open weight models, they must re-examine a series of policies such as data use and distillation. When talking about open source models, Zuckerberg said: "Foreign laboratories currently have advantages in many aspects, because American laboratories have to comply with more restrictions on training data. If we want American open source models to last