TLDR
- Meta has released Llama 3.1, a new open-source AI model with versions up to 405 billion parameters.
- The company claims Llama 3.1 is competitive with closed-source models from rivals like OpenAI and Anthropic.
- Meta is partnering with cloud providers to offer access to Llama 3.1, but has no plans to sell it as a service itself.
- Mark Zuckerberg argues that open-source AI is crucial for wider access and safer development of the technology.
- The model was trained using 16,000 Nvidia H100 GPUs, highlighting Meta’s partnership with Nvidia.
Meta, the parent company of Facebook, has unveiled its latest artificial intelligence model, Llama 3.1. This new open-source AI system is a big step forward in the company’s efforts to make powerful AI technology freely available to developers and researchers worldwide.
Llama 3.1 comes in three versions, with the largest boasting 405 billion parameters. This makes it Meta’s most capable AI model to date. The company claims that Llama 3.1 can compete with closed-source models from industry leaders like OpenAI and Anthropic across various tasks.
Mark Zuckerberg, Meta’s CEO, believes that open-source AI is crucial for the future of the technology.
“Open source will ensure that more people around the world have access to the benefits and opportunities of AI,”
Zuckerberg wrote in a blog post. He argues that this approach will prevent power from being concentrated in the hands of a few companies and allow for safer, more even deployment of AI across society.
Unlike some competitors, Meta isn’t planning to sell access to Llama 3.1 directly. Instead, the company is partnering with major cloud providers like Amazon Web Services, Google Cloud, and Microsoft Azure. These partners will offer access to Llama 3.1 through their platforms, along with security and management tools.
The development of Llama 3.1 highlights Meta’s growing partnership with Nvidia, a leading manufacturer of AI chips. Meta used 16,000 of Nvidia’s H100 graphics processors to train the largest version of Llama 3.1, underscoring the massive computing power required for modern AI development.
Llama 3.1 brings several improvements over its predecessors. It can converse in multiple languages, write higher-quality computer code, and solve more complex math problems. U.S.-based users of WhatsApp and Meta’s AI website will be able to interact with a digital assistant powered by Llama 3.1, showcasing its capabilities in real-world applications.
Meta’s approach to AI development differs from some of its competitors. While companies like OpenAI keep their most advanced models private and charge for access, Meta is making Llama 3.1 freely available and modifiable. This open-source strategy could lead to faster innovation and wider adoption of AI technology.
However, the open nature of Llama 3.1 also raises concerns about potential misuse. Zuckerberg acknowledges that “bad actors may be able to use the intelligence of AI models to fabricate entirely new harms.” But he argues that widespread deployment of AI will allow “larger actors to check the power of smaller bad actors.”
