In the July blog post, Mark Zuckerberg, a meta CEO, said, “Access sales” for Meta’s open -available LLAMA AI model said that it is “access sales” for the “META” S) business model. portion According to a newly unclean court submission, Rama’s money through a profit sharing contract.
A lawyer submitted for the plaintiff in the copyright lawsuit was accused of training an LLAMA model for hundreds of terabytes of hundreds of terabytes of the meta -pirated e -book, and Meta said that the meta -hosted company shares the ratio of profits from the user of the model.
The report does not show which certain host pays the meta. However, META lists multiple llama host partners on a variety of blog posts, including AWS, NVIDIA, Databricks, Groq, Dell, Azure, Google Cloud and Snowflake.
Developers do not need to use the LLAMA model through the host partner. Models can be downloaded, fine adjusted and executed from various hardware. However, many hosts provide additional services and tooling that can be simpler and more easily run the LLAMA model.
Zuckerberg mentioned the possibility of license access to the LLAMA model during the import call in April, and he mentioned that it could generate profits from LLAMA in other ways, such as business messaging services and advertising of “AI interactions.” But he did not explain the details.
Zuckerberg said, “(i) You are like Microsoft, Amazon, or Google, and you will basically resize these services. This is what we think we should get part of your profits.” So we’ve started to make it a little. “
More recently, Zuckerberg argued that most of the value meta derived from LLAMA came in the form of improvement of the model of the AI research community. META uses the LLAMA model to supply power to multiple products across platforms and properties, including META’s AI Assistant and Meta AI.
Zuckerberg said in META’s Q3 2024 import phone, “I think it’s a good business to do this openly.” (I) T creates a better product without creating a model that no one has been standardized in the industry than we just build an island. “
The fact that Meta can make money in a somewhat direct way in LLAMA is Kadrey V. It is important because META’s manuscript insists that META has used illegal works to develop LLAMA, but also promoted infringement through “seeding” or upload. The plaintiff insisted that Meta used a secret turret method to get e -books for training, and shared e -books with other torrent because of the way the rapids worked.
Meta plans to increase capital spending this year thanks to the increase in investment in AI this year. In January, the company announced that it will increase the data center and the AI development team of data centers and companies in 2025 to CAPEX ($ 200 billion) (approximately Double Meta’s CAPEX in 2024).
META, which is likely to offset some of the costs, is known to be considering starting the subscription service of Meta AI, which will add unknown functions.
3/21 update at 1:54 pm: A meta spokesman pointed out a copy of the import telephone for additional context TechCrunch. We added Zuckerberg’s quote. In particular, there is a quotation about META’s intention to profit share with large hosts of the LLAMA model.
