Meta confirms its Llama 3 open-source LLM is coming next month

0


At an event in London on Tuesday, Meta confirmed it is planning the initial release of Llama 3 within the next month – the next generation of its large language models used to power generative AI assistants. Is.

This confirms a report published by The Information on Monday that Meta was getting closer to launch.

Nick Clegg, Meta’s president of global affairs, said, “Within the next month, really short, hopefully very short time, we expect to begin launching our new suite of next-generation foundation models, the Llama 3. We do.” He described what the release of several different iterations or versions of the product looks like. “There will be many different models with different capabilities, different versatility [released] “During this year, it will actually start very soon.”

The plan will be to power multiple products in Meta with Llama 3, said Chris Cox, Meta’s chief product officer.

Meta has been struggling to catch up to OpenAI, which surprised it and other big tech companies like Google when it launched ChatGPT a year ago and the app went viral, bringing generative AI questions and answers into everyday, mainstream experiences. Changed into.

Meta has largely taken a very cautious approach with AI, but this has not gone over well with the public, with previous versions of Llama being criticized for being very limited. (Llama 2 was released publicly in July 2023. The first version of Llama was not released to the public, yet it leaked online.)

Llama 3, which has a larger scope than its predecessors, is expected to address this, with the ability to not only answer questions more accurately, but also to address a broader range of questions that include more controversial topics. Can happen. He hopes that this will make the product popular among users.

“Over time, our goal is to make Llama-powered meta AI the most useful assistant in the world,” said Joël Pineau, vice president of AI Research. “There’s still a lot of work left to do to get there.” The company did not talk about the size of the parameters being used in Llama 3, nor did it offer any demos of how it would work. It is expected to have about 140 billion parameters, compared to 70 billion for the largest Llama 2 model.

Most notably, the Llama family of metas, built as open-source products, represent a distinct philosophical view of how AI should develop as a widespread technology. In doing so, Meta is hoping to play to the broader side with developers versus a more proprietary model.

But it seems like Meta is also playing it more cautiously, especially when it comes to other generic AI beyond text generation. The company is not yet releasing its image generation tool, Emu, Pineau said.

“To produce images that you are proud of and that represent your creative context, ease of use as well as security as well as latency matter a lot,” Cox said.

Ironically – or perhaps predictably (heh) – even as Meta is working to launch Llama 3, it has some significant generative AI skeptics in-house.

Yann LeCun, a renowned AI academic who is also Meta’s chief AI scientist, took a dig at the limitations of generative AI overall and said his bet is on what comes after. They speculate that Joint Embedding Predicting Architecture (JEPA), which is a different approach to both training models and generating results, will be used by Meta to create more accurate predictive AI in the field of image generation.

“The future of AI is ZEPA. This is not generative AI,” he said. “We need to change the name of the Chris’s produce division.”



Source link

About Author

Leave a Reply

Your email address will not be published. Required fields are marked *