Meta Platforms CEO Mark Zuckerberg has urged for reduced U.S. barriers on open-source AI to enhance competition with Chinese counterparts. The social media company revealed a smaller and more customizable AI model named Muse Glimmer, tailored for tasks on a Mac or PC with a single graphics card to meet the demand for AI systems functioning directly on users’ devices. Zuckerberg hinted at the imminent launch of larger models in a video post accompanying a 14-page essay promoting wider dissemination of AI, highlighting concerns about concentrated power in the AI realm.
The move reflects growing support for open-weight AI, particularly amidst rising apprehensions about escalating AI costs and recent cybersecurity breaches involving models from Anthropic, OpenAI, and Meta. Open-weight models, unlike closed models, allow users to modify their core components, known as weights, for specific purposes and are usually more cost-effective.
The spotlight on open-weight models intensified following the release of prominent Chinese models like Kimi K3 and a security incident where a Chinese open-weight model was utilized by Hugging Face to counter an attack due to the limitations of closed-source models. Meta, alongside other tech giants like Nvidia, Microsoft, and Palantir, signed an open letter backing open-weight AI, acknowledging the dominance of Chinese startups in this arena.
Meta’s stock, which has seen a 10% decline this year, surged nearly 3% in premarket trading post the AI model announcement. The company also plans to unveil Muse Spark 1.2, its most advanced model, to regain ground in the AI race. Zuckerberg emphasized the need for the U.S. to revisit policies to support domestic leadership in open-weight models, including data use and distillation methods to train efficient models with less computational power.
Furthermore, Meta introduced a $1 billion US fund to address community concerns arising from its data-center expansions. Zuckerberg highlighted challenges in building infrastructure in the U.S. compared to countries like China and stressed the importance of governance structures for releasing models safely. The U.S. government’s stance on open-weight AI safety tests was discussed, with the Trump administration opting not to subject these models to voluntary safety assessments.
