Llama 3.1: Mark Zuckerberg's Meta Launches 'Biggest' Open Source AI Model Yet

Meta has launched Llama 3.1, claimed to be the largest and most advanced open-source AI model. The company said the new model now outperforms top private models like OpenAI's GPT-4o and Anthropic's Claude 3.5 Sonnet.

In a blog post, the company said Llama 3.1 is incredibly complex, with 405 billion parameters. It was developed using over 16,000 of Nvidia's top-tier H100 GPUs, costing Meta hundreds of millions of dollars. Despite this, Meta is offering Llama 3.1 mostly for free under an open-source licence, only requiring approval from companies with huge user bases.

The company said Mark Zuckerberg, Meta's CEO, believes that open-source AI models will soon be better than other models, similar to how Linux became the leading open-source operating system. He compared this strategy to Meta's Open Compute Project, which reduced data centre costs by working with other companies to improve designs.

Meta has teamed up with big names like Microsoft, Amazon, Google, Nvidia, and Databricks to help get Llama 3.1 into use. The model is cheaper to run than OpenAI's GPT-4o and can be customised by companies for their specific needs.

The company said Llama 3.1 was trained using synthetic data, improving the larger model with help from smaller versions. According to The Verge, Meta's VP of Generative AI Ahmad Al-Dahle said the model could serve as a "teacher" for smaller models, making it more affordable for developers.

Meta included extensive testing for cybersecurity and biochemical applications in Llama 3.1. The model can perform complex tasks like integrating with search engines to retrieve information and use multiple tools. This makes Meta's AI assistant, Meta AI, very versatile and available across Instagram, Facebook, and WhatsApp.

From this week, Meta AI, powered by Llama 3.1, will be available on WhatsApp and Meta's website in the US, with Instagram and Facebook following soon. The assistant will also support more languages, including French, German, Hindi, Italian, and Spanish. However, the most advanced version of the model will only be available for a limited number of prompts per week due to high costs.

Meta is also introducing a new feature called "Imagine Me," which scans users' faces through their phone cameras to create personalised images. This feature aims to meet the growing demand for AI-generated media content.

Meta AI will also replace the voice command interface in the Quest headset, helping users identify and learn about objects in their environment.

Despite the high costs and competition, Zuckerberg is confident in the open-source approach, believing it will drive faster industry progress and secure Meta’s leadership in AI.