The direction of Meta's artificial intelligence development may be undergoing a major shift. According to a report from The New York Times, core members of Meta's newly established Super Intelligence Lab are discussing abandoning the current open-source AI model "Behemoth" and instead developing proprietary models. If this move comes to fruition, it would mark a fundamental change in Meta's philosophy in the AI field.
Although Meta has completed training for Behemoth, the company postponed its release due to unsatisfactory internal test results. After the establishment of the new lab, testing on the model was also halted. Although the discussion is still in its early stages, Mark Zuckerberg, Meta's CEO, needs to make the final decision. However, a company spokesperson stated that Meta's stance on open-source AI remains "unchanged."
"We plan to continue releasing leading open-source models," the spokesperson emphasized. "Although we have not released all the models we have developed in the past, we will continue to train a combination of open-source and proprietary models in the future." However, if this potential shift actually happens, it could cast doubt on Meta's open-source philosophy.
In the past, Zuckerberg viewed open-source as a key difference between Meta and competitors like OpenAI, and had criticized OpenAI's partnership with Microsoft. However, as Meta increases its investment in the AI field, the company needs to find new ways to generate revenue, especially under the pressure of billions in AI R&D costs and high salaries for top research talent.
Despite having world-class AI research laboratories, Meta lags behind competitors such as OpenAI, Anthropic, and Google DeepMind in commercializing AI achievements. If Meta decides to prioritize the development of proprietary models, it may indicate that its open-source strategy is not driven by conviction, but by commercial considerations.
This potential shift could not only change Meta's own strategy, but also have a profound impact on the entire AI industry. The trend of open-source AI may weaken due to Meta's proprietary tendencies, and power may return to large companies with closed ecosystems. Additionally, such changes could affect the development of small companies that rely on open foundational models, particularly in efforts related to model tuning, safety, and alignment.
Meta's move may also reduce the competitiveness of open-source AI on the international stage, especially in countries like China, where there is an active embrace of open-source AI to enhance domestic capabilities and global influence.
