On April 18, Meta released Llama 3, the latest version of its large language model (LLM), calling it a “significant leap over Llama 2.” Initially, the first two models of the current version, featuring 8 billion and 70 billion parameters, were released, with plans for upcoming models to include 400 billion parameters.
Meta highlighted that Llama 3 was trained with a “large, high-quality training dataset” of over 15 trillion tokens, 7x larger than Llama 2 and containing four times more code. Llama 3 also employs filtering techniques, including NSFW filters, to maintain data quality.
Llama 3 surpasses Llama 2 and competitors like Anthropic’s Claude Sonnet, Mistral Medium, and OpenAI’s ChatGPT-3.5 in over half of 12 use cases.
The initial releases of Llama 3 are text-based models, with future releases expected to be multilingual and multimodal. These releases will offer longer context and improved reasoning and coding performance, which Meta identifies as “core LLM capabilities.”
Meta plans to make Llama 3 available on all major cloud providers, model API providers, and other services, with the goal of launching the product “everywhere.”
Wider User Access
While Llama 3 is designed for developers, Meta is introducing new ways for end users to access AI services in the US and over 12 other countries.
One key feature is a dedicated website called Meta AI, providing users with AI-powered writing assistance, trivia games, simulated job interviews, and homework help.
Meta has also integrated Meta AI with its other products, including Facebook, Instagram, WhatsApp, and Messenger. Additionally, the service is available through Ray-Ban Meta smart glasses in the US and will be extended to Meta Quest VR headset.
In an announcement, CEO Mark Zuckerberg stated, “We believe that Meta AI is now the most intelligent AI assistant that you can freely use.”
This news of Meta’s expanded AI products follows upgrades to competing services. ChatGPT upgraded to GPT-4 Turbo on April 11, while Microsoft Copilot started using GPT-4 Turbo in March, intensifying the race among consumer-oriented AI services.