Zuckerberg Urges US to Ease Open-Source AI Policies, Announces New Models
Meta CEO Mark Zuckerberg has called on the US government to reduce policy barriers on data usage and knowledge distillation to help American open-source AI compete with China. Alongside this advocacy, Meta announced the release of its new open-weight model, Muse Glimmer, and plans to open-source Muse Spark 1.2, supported by a $1 billion community fund. This move highlights the strategic geopolitical tension in AI development, positioning open-source models as a vital tool for national competitiveness. By lowering barriers and releasing smaller, device-run models, Meta aims to democratize AI access and challenge closed-source dominance. The newly released Muse Glimmer is designed for agent tasks and can run on a single-GPU Mac or PC, targeting on-device AI needs. Zuckerberg also emphasized that US laboratories face stricter data constraints than foreign competitors, and restricting foreign open-source models is not a viable solution.
## BACKGROUND
Open-weight models are AI models whose final trained neural network weights and biases are publicly released, allowing developers to run and customize them locally. Knowledge distillation is a machine learning technique where a smaller, more efficient "student" model is trained to replicate the behavior and performance of a larger "teacher" model.