Meta Announces Upcoming Release of Llama 3: Open Source LLM Launching Next Month

Meta Confirms Upcoming Release of Llama 3, Its Next-Gen Language Model

On Tuesday, Meta announced at a London event that it plans to roll out Llama 3—its advanced large language model for powering generative AI assistants—within the next month. This follows a report from The Information indicating that Meta was nearing the launch.

“Within the next month, and hopefully even sooner, we anticipate starting the rollout of our next-generation foundation models, Llama 3,” stated Nick Clegg, Meta’s President of Global Affairs. He emphasized that multiple versions of the model, each with varying capabilities, will be launched throughout the year.

Meta's Chief Product Officer, Chris Cox, noted that Llama 3 would enhance various products across the Meta ecosystem. Following the unexpected success of OpenAI's ChatGPT, which took the tech world by storm over a year ago, Meta has been racing to close the gap.

Traditionally, Meta has adopted a cautious approach to AI development, but this strategy has drawn criticism from users who found earlier iterations, such as Llama 2—released publicly in July 2023—to be limited in scope. Llama 3 aims to overcome these shortcomings by not only improving accuracy in responses but also expanding the range of questions it can tackle, including contentious topics, in hopes of resonating more strongly with users.

“Our long-term goal is to develop a Llama-powered Meta AI that serves as the most helpful assistant globally,” remarked Joelle Pineau, Vice President of AI Research. While the company refrained from detailing the model's parameter size, it is anticipated that Llama 3 will feature around 140 billion parameters, a significant increase from the 70 billion in the largest Llama 2 model.

A key distinguishing feature of the Llama series is its open-source nature, positioning Meta as a more collaborative force in the AI landscape compared to proprietary models favored by others. However, Meta appears to be taking extra precautions, particularly regarding its image generation tool, Emu, which will not be launched yet, according to Pineau.

“Latent factors, safety considerations, and user experience are critical for generating images that truly reflect one’s creative vision,” Cox explained.

Interestingly, while preparations for Llama 3 are underway, some influential voices within Meta express skepticism about generative AI's direction. Yann LeCun, Meta’s Chief AI Scientist and a renowned figure in the AI community, critiqued the limitations of generative AI, suggesting that the future lies in joint embedding predicting architecture (JEPA)—a different methodology that Meta is leveraging for more accurate image generation.

“The future of AI is JEPA. It’s not generative AI,” he stated, suggesting a rebranding for Chris's product division might be in order.