Meta announced the newest version of its Llama artificial intelligence model on Tuesday, called Llama 3.1. The latest Llama technology is available in three different versions, with one variant being Meta's largest and strongest AI model yet. Like previous versions of Llama, the newest model stays open source, meaning it’s freely accessible.
The recent Large Language Model (LLM) underscores the social network's massive investment in AI spending to maintain pace with corporations like emerging startups OpenAI and Anthropic, in addition to other tech giants like Google and Amazon.
The announcement also underlines the growing partnership between Meta and NVIDIA. Nvidia is a key meta-partner, supplying Facebook's parent company with computer chips – called GPUs – to coach its AI models, including the newest version of Llama.
While corporations like OpenAI need to earn money by selling access to their proprietary LLMs or offering services to assist their customers use the technology, Meta has no plans to launch its own competing enterprise business, a Meta spokesperson said during a press conference.
Instead, just like the way in which it did when it released Llama 2 last summer, Meta is working with a handful of technology corporations that can provide their customers with access to Llama 3.1 through their respective cloud computing platforms and sell security and management tools that work with the brand new software. Meta's 25 Llama-related enterprise partners include Amazon Web Services, Google Cloud, Microsoft Azure, Databricks and Dell.
Although Meta CEO Mark Zuckerberg has told analysts in previous earnings calls that the corporate generates some revenue from its partnerships with Llama, a Meta spokesperson said any financial profit is merely incremental. Instead, Meta believes that by investing in Llama and related AI technologies and making them available without cost via open source, it will probably, amongst other things, attract high-quality talent in a competitive market and reduce its overall computing infrastructure costs.
Meta's launch of Llama 3.1 comes ahead of a conference on advanced computer graphics where Zuckerberg and Nvidia CEO Jensen Huang will speak together.
The social networking giant is one in every of Nvidia's top customers that doesn't run its own business-focused cloud, and Meta needs the newest chips to coach its AI models, which it uses internally for targeting and other products. For example, Meta said the biggest version of the Llama 3.1 model announced Tuesday was trained on 16,000 Nvidia H100 graphics processors.
But the connection can also be necessary to each corporations due to what it represents.
For Nvidia, the indisputable fact that Meta trains open-source models that other corporations can use and adapt for his or her business operations – without paying a licensing fee or asking for permission – could expand using Nvidia's own chips and keep demand high.
However, developing open source models can cost a whole bunch of hundreds of thousands or billions of dollars. There will not be many corporations financially in a position to develop and release such models with comparable investments. Google and OpenAI are Nvidia customers, but keep their most advanced models secret.
Meta, then again, needs a reliable supply of the newest GPUs to coach increasingly powerful models. Like Nvidia, Meta is attempting to foster an ecosystem of developers constructing AI apps with the corporate's open-source software at its core, even when Meta has to essentially give away code and so-called AI weights, that are expensive to develop.
The open-source approach advantages Meta by familiarizing developers with its internal tools and alluring them to construct on them, Ash Jhaveri, the corporate's vp of AI partnerships, told CNBC. Meta also advantages because the corporate uses its AI models internally, allowing it to learn from improvements made by the open-source community, he said.
Zuckerberg wrote in a blog post on Tuesday that they were taking a “different approach” to releasing Llama this week, adding: “We are actively building partnerships so that more companies in the ecosystem can also offer unique features to their customers.”
Because Meta shouldn’t be an enterprise vendor, it will probably refer corporations inquiring about Llama to one in every of its enterprise partners, equivalent to Nvidia, Jhaveri said.
The largest version of the Llama 3.1 model family is named Llama 3.1 405B. This LLM incorporates 405 billion parameters, the variables that determine the general size of the model and the quantity of knowledge it will probably process.
In general, a big LLM with a lot of parameters can perform more complicated tasks than smaller LLMs, equivalent to understanding context in long text streams, solving complex mathematical equations, and even generating synthetic data that may presumably be used to enhance smaller AI models.
Meta can also be releasing smaller versions of Llama 3.1, the Llama 3.1 8B and Llama 3.1 70B models. They are essentially improved versions of their predecessors and will be used to run chatbots and software coding assistants, the corporate said.
Meta also said the corporate's U.S.-based WhatsApp users and visitors to its Meta.AI website will have the opportunity to experience Llama 3.1's capabilities by interacting with the corporate's digital assistant. The digital assistant, which runs on the newest version of Llama, will have the opportunity to reply complicated math problems or solve software coding issues, a Meta spokesperson explained.
WhatsApp and Meta.AI users based within the US will have the opportunity to modify between the brand new, gigantic Llama 3.1 LLM or a less powerful but faster and smaller version to reply their questions, the Meta spokesperson said.
image credit : www.cnbc.com