At an event in London on Tuesday, Meta confirmed that it is planning an early release of Llama 3 – the next generation of its large language model used to power generative AI assistants – within the next month.
This confirms a report published by The Information on Monday that Meta was about to launch.
“Over the next month or less, and hopefully within a very short period of time, we hope to begin rolling out our new suite of next-generation foundation templates, Llama 3,” said Nick Clegg, President of Business Meta World Cups. He described what looks like releasing several different iterations or versions of the product. “There will be a number of different models with different capabilities and different versatility (launched) over the course of this year, and that will start very soon.”
The plan, added Chris Cox, Meta’s chief product officer, will be to power multiple products across Meta with Llama 3.
Meta has been scrambling to catch up to OpenAI, which took it and other big tech companies like Google by surprise when it launched ChatGPT more than a year ago and the app went viral , transforming generative AI questions and answers into everyday consumer experiences.
Meta has largely taken a very cautious approach to AI, but this has not been well received by audiences, with previous versions of Llama being criticized as being too limited. (Llama 2 was released to the public in July 2023. The first version of Llama was not released to the public, but it was still leaked online.)
Llama 3, which is larger in scope than its predecessors, is expected to address this issue, with capabilities to not only answer questions more accurately, but also answer a wider range of questions that may include more controversial topics. He hopes this will allow the product to appeal to users.
“Our goal over time is to make Llama-powered meta-AI the most useful assistant in the world,” said Joelle Pineau, vice president of AI research. “There’s still a lot of work to do to get there.” The company hasn’t talked about the settings size it’s using in Llama 3, nor has it offered a demo of how it works. It is expected to contain around 140 billion parameters, compared to 70 billion for the larger Llama 2 model.
Most notably, Meta’s Llama families, designed as open source products, represent a different philosophical approach to how AI should develop as a broader technology. In doing so, Meta hopes to gain favor with developers over more proprietary models.
But Meta also seems to be playing more cautiously, especially when it comes to other generative AI beyond text generation. The company has not yet launched Emu, its image generation tool, Pineau said.
“Latency is very important, as is security and ease of use, to generate images that you are proud of and that represent whatever your creative context,” Cox said.
Ironically – or perhaps predictably (heh) – even as Meta works to launch Llama 3, there are some significant generative AI skeptics in the house.
Yann LeCun, a famous AI scholar and also Meta’s chief scientist, looked at the limitations of generative AI as a whole and said his bet was on what comes next. He predicts it will be Joint Integration Prediction Architecture (JEPA), a different approach to both training models and producing results, which Meta has used to create more predictive AI precise in the field of image generation.
“The future of AI is JEPA. It’s not generative AI,” he said. “We’re going to have to change the name of Chris’s product division.”
techcrunch
