Meta CEO outlines open-source AI strategy and warns of concentrated power

Meta Platforms said it is releasing two new AI models, Muse Glimmer, which can run on a personal computer, and Muse Spark 1.2, a more powerful version for developers. In a 6,500-word online essay, CEO Mark Zuckerberg argued that open-source AI would give individuals "invention superpowers" and help create new businesses, provide advanced tutoring and offer personalized lifestyle advice. He warned that concentrating "superintelligence" in the hands of a few companies, institutions or governments could produce undesirable outcomes and called for the technology to be "broadly distributed" to empower billions.

Zuckerberg contrasted Meta’s approach with other labs that focus on building AI for large organizations, suggesting that such a focus could shift power toward larger institutions. He also urged U.S. policymakers to accelerate energy infrastructure and reconsider restrictions on a technique called "distillation," which trains smaller models on the outputs of larger ones.

The release drew criticism from AI safety advocates. Anthony Aguirre, president of the Future of Life Institute, questioned Meta’s confidence in controlling rapidly advancing AI and called for a pause in development. Matt Lane of the digital-rights group Fight for the Future said open-source AI is important for security but cautioned that Meta could benefit disproportionately if its own models become the de-facto standard.