At a London event on Tuesday, Meta announced that it will launch Llama 3, its latest large language model for AI assistants, in the coming month.
“Within the next month, actually less, hopefully in a very short period of time, we hope to start rolling out our new suite of next-generation foundation models, Llama 3,” said Nick Clegg, Meta’s president of global affairs. He talked about what seems like the launch of multiple versions or updates of the product.
“There will be a number of different models with different capabilities, different versatilities [released] during the course of this year, starting really very soon.”
Chris Cox, the Chief Product Officer at Meta, stated that their plan involves integrating Llama 3 into several Meta offerings.
Meta is quickly trying to match OpenAI’s pace after being taken by surprise when OpenAI launched ChatGPT over a year ago. The success of ChatGPT made interacting with generative AI a regular occurrence for many people, catching Meta and other big tech companies like Google unprepared.
Meta has been careful with its AI development, but this strategy hasn’t been well-received, as earlier Llama versions were seen as too restricted. Llama 2 became available to everyone in July 2023, while the initial Llama version wasn’t shared with the public but still ended up online.
“Our goal over time is to make a Llama-powered Meta AI be the most useful assistant in the world,” said Joelle Pineau, vice president AI Research. “There’s quite a bit of work remaining to get there.”
The company didn’t share details on the number of parameters in Llama 3 or show any demonstrations of its functionality. It is believed that Llama 3 will have around 140 billion parameters, which is double the 70 billion parameters of the largest Llama 2 model.
Meta’s Llama series, created as open-source tools, show a unique way of thinking about AI’s growth as a broader technology. By choosing this path, Meta aims to gain more support from developers compared to more closed-off models.
However, Meta is also being more careful, particularly with AI that goes beyond creating text. For now, the company isn’t launching Emu, its tool for generating images, according to Pineau.
“Latency matters a lot along with safety along with ease of use, to generate images that you’re proud of and that represent whatever your creative context is,” Cox said.
Interestingly, or maybe as expected, Meta faces internal skepticism about generative AI even as it pushes forward with the release of Llama 3.
Yann LeCun, a renowned AI expert and Meta’s top AI scientist, criticized the current state of generative AI, suggesting his focus is on future technologies.
He believes the next big thing will be the joint embedding predicting architecture (JEPA), a new method for training models and generating outcomes. Meta is already applying JEPA to develop more precise AI for creating images.
“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.”
What we think?
I think Meta’s Llama 3 launch will be big news. They’re making lots of versions to help with different tasks. They want to catch up with ChatGPT.
Before, their AI wasn’t shared much, but now they’re opening it up, hoping more people will help make it better.
They’re still careful, especially with AI that makes pictures. There’s some doubt inside Meta, but they’re still moving forward with new ideas.