Meta may keep its most advanced AI models closed source

Meta CEO Mark Zuckerberg has hinted that the company may not open source its most advanced AI models, marking a potential shift in strategy. In a recent letter, Zuckerberg emphasized the need to balance widespread access to "personal superintelligence" with safety concerns.

Key Points:

• Open Source Commitment in Question: While Meta has historically positioned its Llama models as open-source alternatives to competitors like OpenAI and Google DeepMind, Zuckerberg now states, "We’ll need to be rigorous about mitigating these risks and careful about what we choose to open source."

• Monetization and Control: Closed models offer more control over monetization, a factor that has driven Meta’s rivals to keep their models proprietary. Zuckerberg previously argued that Meta’s revenue model—primarily based on advertising—allowed it to open-source AI without undercutting profitability.

• Recent Investments: Meta has invested heavily in AI, including a $14.3 billion deal with Scale AI and the creation of a new unit, Meta Superintelligence Labs. Reports suggest the company has paused testing on its latest open-source model, Behemoth, to focus on closed models.

• Product Integration: Zuckerberg envisions "personal superintelligence" being delivered through Meta’s own products, such as AR glasses and VR headsets, which he believes will become primary computing devices.

Meta’s Stance:

A Meta spokesperson reiterated the company’s commitment to open-source AI but acknowledged plans to train both open and closed models in the future. "We haven’t released everything we’ve developed historically," the spokesperson said.

This article was updated with additional context on Zuckerberg’s stance.