San Francisco, CA – Meta, the technology conglomerate, recently unveiled Llama 3.1, the largest open-source AI model yet in the industry. This new model promises superior performance compared to top private models like GPT-4o and Anthropic’s Claude 3.5 Sonnet according to the company. Meta is also expanding the availability of its Meta AI assistant based on the Llama model to more countries and languages, along with adding a new feature that can generate images resembling specific individuals.
The Llama 3.1 is a substantial upgrade from its predecessors, boasting a massive 405 billion parameters and was trained with over 16,000 of Nvidia’s H100 GPUs. Despite the high development costs, Meta is offering the Llama model under an open-source license requiring approval from companies with a vast user base. This move is akin to how Linux became the dominant open-source operating system in the tech industry.
Meta’s CEO Mark Zuckerberg compares the investment in open-source AI to Meta’s success with the Open Compute Project, which saved the company billions by collaborating with outside companies. He believes that the release of Llama 3.1 will mark a significant shift in the industry towards open-source models being the primary choice for developers.
In a bid to make Llama 3.1 accessible, Meta has partnered with several tech giants like Microsoft, Amazon, Google, Nvidia, and Databricks to assist developers in deploying their versions of the model. Meta claims that Llama 3.1 is more cost-effective to run in production compared to OpenAI’s GPT-4o. The company is also sharing the model weights to enable companies to train and customize it as per their requirements.
Meta remains tight-lipped about the training data used for Llama 3.1, citing it as a trade secret. The use of synthetic data to enhance the model’s performance is highlighted, with Meta’s VP of generative AI Ahmad Al-Dahle foreseeing Llama 3.1 as a valuable resource for developers to optimize smaller models cost-effectively.
Additionally, Meta’s rigorous testing of Llama 3.1 includes cybersecurity and biochemical use cases, with a focus on identifying emerging “agentic” behaviors. This latest model integration enables complex tasks like information retrieval and tool utilization based on user queries, showcasing the model’s versatility and robust capabilities.
As Meta launches Llama 3.1 through its Meta AI assistant on platforms like WhatsApp and Meta AI website, the company is set to introduce support for new languages to enhance user experience. The “Imagine Me” feature, allowing users to insert their likeness into generated images, is introduced to meet the rising demand for AI-generated content, albeit raising concerns about deepfake technology.
Looking ahead, Meta plans to integrate Meta AI into the Quest headset to replace the voice command interface, expanding its functionality and accessibility across different platforms. Despite the industry’s anticipation, Meta refrains from disclosing usage numbers for its assistant, hinting at the ongoing evolution and potential of AI applications in the market.