Overview
- On Monday Meta published Muse Glimmer as an open-weight model whose trained parameters are available for download and said it will make weights for a version of Muse Spark 1.2 available in the coming weeks.
- Meta says Muse Glimmer is a 30-billion-parameter, distilled model that learned from the larger Muse Spark system and is designed to run agent-like tasks on a single consumer graphics card.
- Mark Zuckerberg accompanied the release with a 6,500-word essay arguing that concentrating control of powerful AI is the largest risk and calling for wider distribution of models plus changes to U.S. rules on training data and testing.
- Meta announced a governance change giving independent directors authority to approve safety criteria for model releases as the company frames open weights as a way to speed security research after recent incidents involving other labs.
- The move is pitched as a response to competitive pressure from Chinese open-models and could force policy and infrastructure choices in the U.S., including more data-center investment and renewed debate over voluntary safety tests.