Meta's AI Strategy Shift: Is Openness Ending?

Meta Considers Shift in AI Strategy

Leading figures within Meta's newly established Superintelligence Lab have reportedly been evaluating a change in direction, contemplating a move away from their powerful, publicly available AI model, Behemoth. The discussions center on developing a closed model instead, according to a report by The New York Times.

Behemoth's Development and Internal Assessment

Sources familiar with the matter have indicated that Meta had finalized the training process for Behemoth. However, its public release was postponed following assessments that revealed underwhelming performance during internal testing. Further testing on the model was subsequently paused with the launch of the Superintelligence Lab.

Current Stance and Future Plans

It’s important to note that these are preliminary discussions, and any alterations to Meta’s approach would require the approval of CEO Mark Zuckerberg. A company representative affirmed to TechCrunch that Meta’s commitment to open source AI remains firm.

“Our intention is to continue the release of cutting-edge open source models,” the spokesperson stated. “Not all of our past developments have been publicly released, and we anticipate continuing to train both open and closed models in the future.”

Potential Philosophical Shift

The spokesperson refrained from commenting specifically on a potential departure from Behemoth. However, a prioritization of closed-source models would represent a significant departure from the company’s established principles.

Open Source as a Strategic Advantage

While Meta continues to utilize more sophisticated closed-source models internally, such as those driving its Meta AI assistant, Zuckerberg had previously championed open source as a core element of the company’s external AI strategy. This approach was intended to accelerate the pace of AI development.

He actively promoted the openness of the Llama family of models as a key differentiator compared to competitors like OpenAI, which Zuckerberg had publicly criticized for adopting a more restrictive approach following its partnership with Microsoft.

Monetization Pressures and Investment Costs

Meta is currently facing pressure to diversify its revenue streams beyond advertising, coinciding with substantial investments in AI. These investments encompass significant signing bonuses and substantial salaries for attracting leading researchers.

Furthermore, the company is funding the construction of new data centers and covering the considerable expenses associated with developing artificial general intelligence (AGI), often referred to as “superintelligence.”

Lagging Behind in Commercialization

Despite possessing a highly regarded AI research division, Meta currently trails behind competitors such as OpenAI, Anthropic, Google DeepMind, and xAI in the commercial application of its AI technologies.

Openness as Strategy, Not Ideology?

A shift towards prioritizing closed models could indicate that the previous emphasis on openness was primarily a strategic maneuver rather than a deeply held ideological commitment. Zuckerberg’s past statements suggest a degree of flexibility regarding the open sourcing of Meta’s models.

During a podcast appearance last summer, he remarked that closed models would afford Meta greater control and expanded monetization opportunities, particularly if the acquired talent can deliver superior performance.

Reshaping the AI Landscape

Such a change could also significantly alter the broader AI landscape. The momentum behind open source initiatives, largely fueled by Meta and models like Llama, could decelerate.

This could potentially shift power back towards major companies with closed ecosystems, while open source development might become more reliant on grassroots efforts. The resulting impact would be felt throughout the startup ecosystem, particularly by smaller companies specializing in fine-tuning, safety, and model alignment, which depend on access to open foundation models.

Global Implications and China's Role

On a global scale, Meta’s potential retreat from open source could inadvertently benefit China, which has actively embraced open source AI – including initiatives like DeepSeek and Moonshot AI – as a means of strengthening its domestic capabilities and expanding its global influence.