Meta now sees Llama as an AI ecosystem that dominates the open-source scene. A future version could be multimodal.
In a podcast with Lex Fridman, Mark Zuckerberg says that Llama 2 has far exceeded his expectations for the model's success. As for Llama 3, Zuckerberg has not yet announced anything concrete.
However, Zuckerberg says that there is always a new model in training, but also made it clear that Llama 3 is not close. He hopes it can also be released as open source.
For now, the focus is on getting Llama 2 into as many applications as possible, Zuckerberg said. Meta this week unveiled Meta AI, a ChatGPT competitor powered by Llama 2 and several other generative AI features for its platforms.
Llama becomes Meta's language model ecosystem
According to Meta, Llama-based models have been downloaded more than 30 million times on Hugging Face, including 10 million in the last month alone. On Github, more than 7000 projects would mention Llama. On Google Cloud and AWS, more than 3500 enterprise projects have been launched based on Llama 2, the company says.
"Much like PyTorch, Llama has evolved into a platform for the world to build on, and we couldn’t be more excited," Meta writes. "The more enterprises and startups build on our technology, the more we can learn about use cases, safe model deployment, and potential opportunities."
Meta benefits from the community's ongoing development of its models. Internal development is accelerated as a result, Meta says, and the model serves as a common language. "LLMs have fundamentally changed AI development, and new tools and approaches are emerging daily for manipulating, managing, and evaluating models," Meta writes.
In addition to further security research and a continued strong community focus, Meta cites multimodal models as the next step: "Just as the world isn’t made up entirely of text, AI can embrace new modalities to enable even more immersive generative experiences."
Meta has already taken a first step in this direction with its ChatGPT competitor Meta AI, which offers image generation in chat in combination with the new image model Emu and the segmentation model Segment Anything.
According to a rumor at the end of August, Llama 3 will be able to compete with GPT-4, which also offers multimodal features. Llama 3 will also be free, according to the rumor.