Meta has revenue sharing agreements with Llama AI model hosts, filing reveals

A Blog post Last July, Meta CEO Mark Zuckerberg said that Meta’s publicly available Lama is “access sales” on AI models “not” [Meta’s] Business model. “Nevertheless Meta makes the least Something According to the newly unused court filing, money from Lama through a contract to share revenue.

The FilingCopyright case cadre vs cadre vs. Meta submitted by attornees in Atornas, where Meta has been accused of training on several hundred terabytes pirated ebooks of its lama models, revealed that Meta “has shared one percent of the earnings” companies that produce users from these models.

The filing does not reveal which specific hosts give the meta. However Meta enroll in several Lama host partners in different fields Blog PostAWS, Nvidia, Databrics, Grock, Dell, Azure, Google Cloud and Snowflake.

Developers do not need to use the Lama model through any host partner. Models can be downloaded, fine tune and run in a range of various hardware. However, many hosts provide additional services and equipment that make Lamama models achieve and run easy and easy.

Zuckerberg mentions the possibility of license access to Lama models During an earning call last AprilWhen he also floated Lama in other ways through the business message service and advertisements of “AI Interactions”. However, he did not publish a specific outline.

Very recently, Zuckerberg Seer Most of the values ​​obtained from the Lama come in the form of improvement of models of the Meta AI research community. The Meta Meta’s AI Assistant with its platform and the Lama models to get a number of products across the property, Meta AIIThe

“I think it’s a good business to do it in an exposed way,” Zuckerberg Meta’s Q3 2024 said while earningThe “[I]Tea is more than that makes our products better if we only create a model on an island that no one in the industry was like standardization. ”

The fact that Meta can earn a direct way from Lamama is significant because in Cadre vs Meta, the plaintiffs have claimed that Meta has not only used pirated works to develop Lamama, but “seeds,” or uploading the violation, these tasks. The plaintiffs complained that he used surreptitious torrenting methods to get E -book for meta training and in the process – due to Torrent Works work – he shared e -books with other torrentors.

Meta plans Up to significantly This year’s capital expenditure, basically thanks for investing in AI. In January, the company said it would spend $ 60 billion-$ 1 billion in Capex in 2021-in 2021, the double meta will increase the Capex-Profit Data Centers and the agency’s AI development teams.

Probably the possibility of offset a portion of the expenditure, the meta is Is being considered in the news Launching a subscription service for Meta AI that will add the assistant power to the assistant.

Meta did not immediately respond to any request for the comment. We will update this piece if we listen back.