Meta is already working on Llama 4, which will require ten times more computing power for training

Aug. 1 (Portaltic/EP) –

Goal has already begun to work in Llama 4, the next generation of its family of large language models, which will require ten times more computing power than that used in Llama 3 for training.

The technology firm plans to launch next year the most advanced model in the industry, Llama 4, which they are already working on. However, achieving this goal and maintaining it in future launches requires a high computational cost.

Meta CEO Mark Zuckerberg has explained in the call to investors held this Wednesday that Llama 4 will require almost ten times more computing power in its training than that used in Llama 3the model currently available.

He has also warned that future models may require even more. “It’s hard to predict how this will play out several generations into the future, but at this point I’d rather take my chances.” building capacity before it is needed rather than too late, given the long lead times for getting new infrastructure projects up and running,” he said.

The company has recently launched Llama 3.1, a version of its foundational Llama 3 model that It is on par with the most advanced closed rivals with its 405 billion parameters (405B) and skills in general knowledge, orientation, mathematics, use of tools and multilingual translation.

Llama 3.1 powers the AI ​​Studio tool that allows you to create and discover AI agents as well as develop personalized chatbots with a name, personality, tone, avatar and slogan.

All these models They are open sourcea model that Zuckerberg has defended as benefiting both AI development and Meta itself. “We are dedicated to creating the best experiences for consumers and advertisers. To do this, we need access to the leading technological infrastructure and not be limited by what competitors allow us to do,” he said.

Consider that the models They are not “isolated pieces of software”“, but are part of an ecosystem, That is why, if they want “the most robust ecosystem of tools, efficiency improvements, silicon optimizations and other integrations” to be part of their models, they need “these to be widely used by developers across the industry.”

METAVERSE

Artificial intelligence is also shaping the metaverse, Meta’s other big project. Reality Labs’ investments have accelerated the arrival of smart devices, such as the glasses it develops with Ray-Baneven anticipating technologies such as holograms.

The Ray-Ban Meta glasses, in particular, have been a success “sooner than expected, partly thanks to AI,” said the executive. The visor has also exceeded his expectations. Quest 3something that, in his opinion, is due to the fact that “It’s not only the best mixed reality headset for the price, but the best headset on the market“.

He also recalled that the latest developments in artificial intelligence and the metaverse will be shared on September 25, at the annual event Meta Connect.