Meta open sources Muse Glimmer AI model, Zuckerberg says ‘superintelligence should empower everyone’

Meta on Monday announced that it is open-sourcing its Muse Glimmer AI model. The announcement came via a video from Meta CEO Mark Zuckerberg, who said that the company was a strong supporter of open source and would be releasing the weights for its top-of-the-line Muse Spark 1.2 model as well.

“It is increasingly important to have a clear philosophy and values for how superintelligence will benefit humanity. Meta is committed to building with the principles of individual empowerment as the source of prosperity, invention as AI's purpose, and a balance of power favouring people as the foundation for addressing safety risks,” Zuckerberg wrote in a blog.

In a video message accompanying the blog, Zuckerberg said that Muse Glimmer is a new class of on-device model with 30 billion parameters and capable of running on laptops. Zuckerberg also said that Glimmer is the ‘highest-performing model of its size’.

The Meta CEO said that there are ‘bigger’ models to be launched by the company soon as well.

In his 6,500-word blog post, Zuckerberg took a veiled dig at top AI companies like OpenAI and Anthropic and argued that superintelligence should not be controlled by a small number of companies or institutions.

Zuckerberg also noted that open-source AI will be able to prevent the centralisation of AI by giving more people access to advanced models.

The Meta CEO also revealed that the company is planning to offer a fully private mode for its personal AI agents, where even Meta would not be able to access or grant access to users' information.

Notably, this is a return of sorts for Meta, which had earlier pioneered open-weights models with the Llama series. However, ever since the restructuring under Meta Superintelligence Labs (MSL), the company had also gone back to releasing closed-source models.

Zuckerberg says US needs to rethink policies:

Zuckerberg also argued that the US must rethink its policies if it wants to lead in the open-source AI race. He also went on to defend the policy of model distillation, a technique in which a less capable model learns from the outputs of a larger model.

“Foreign labs currently hold several advantages here since American labs have to comply with many additional restrictions on training data. US policy must reduce this additional friction if we want American open-source models to lead over time,” Zuckerberg wrote.

The Meta CEO, however, maintained that the US should continue using export control measures on advanced AI chips in order to slow down China's progress. He also warned that falling behind in AI could become a larger and longer-term national security issue than individual risks associated with AI development.