The best members of The new Superintelligence laboratory of Meta Discussion that moves away from the powerful model to the company’s open source, Behemoth, and instead developing a closed model, Reports the New York Times.
Sources referred to the Times that Meta had completed the training on Behemoth, but delayed its release due to disappointing internal performance. When the new Supervintingigence laboratory was launched, according to what was reported, the model was interrupted.
The discussions are precisely this: discussions. The CEO of Meta Mark Zuckerberg should however sign any changes and a spokesman for the company told Techcrunch that the position of Meta on Open Source Ai is “unchanged”.
“We plan to continue the Rebelands Open Source models,” said the spokesperson. “We have released and developed historically and we expect to continue to train a mix of open and closed models in the future”.
The spokesman did not comment on the potential movement of Meta from Behemoth. If this happens so that Meta can prioritize the closed source models, it would mark a great philosophical change for the company.
While the most advanced destination meta models of source closed internally, like those who feed his assistant goal AI, had done Zuckerberg Open Source A central part of the company’s external artificial intelligence strategy – a way to maintain the development of the AI faster. Strongly positioned the Llama family Opening as a differentiation from the Likernai Commitators, which Zuckerberg publicly criticized To be more closed after collaborating with Microsoft. But Meta is under printing to monetize beyond the announcements as it pursues billions in artificial intelligence.
This included the payment of huge signature bonuses and nine -digit employees to cook the best researchers, build new data centers and face the enormous cost of development of artificial general intelligence (AG) or “Superintelligence”.
Despite having one of the best research workshops in the world, Meta is still late behind rivals such as Openi, Anthropic, Google Deepmind and Xai when it comes to marketing his artificial intelligence work.
If Meta gives priority to closed models, it could suggest that opening was a strategic, non -ideological game. Zuckerberg’s past comments suggest an ambivalence to engage in the opening of the destination models. We have podcast Last summer, he said:
Obviously we are very proen open source, but I have committed myself to release every single thing we do. In the bases I am very inclined to think that the open sand will be fine for the community and also a good for us because we will benefit from innovations. If at a certain point, however, there is some qualitative change in what the thing is capable of, and we feel as if it is not responsible for open source, then we will not do it. Everything is very difficult to predict.
The closed models would give metals more and more ways to monetize, especially if it believes that the talent it has acquired can be competent performance and the best performances.
Such a change could also reshape the landscape of the AI. The open source moment, largely guided by destination and models such as the blade, could slow down, even if Openi is preparing to release his Open model still delayed. Power could go back to the main actors with closed ecosystems, while open source development could remain a product of basic efforts. The chain effects would continue through the startup ecosystem, in particular for smaller companies focused on alignment at the end of the turning point, safety and model that are based on access to open foundation models.
On the world phase, the retreat of the goal from Open Source could potentially give in to the ground in China, which it embraced to open source – such as Deepseek e Moshot ai – as a way to develop domestic skills and global influence.
