Faster than GPT-5? Mark Zuckerberg confirms: Meta Platforms is training Llama 3

After Altman first revealed GPT-5, Zuckerberg also announced that he is training Llama 3.

At around 2:00 am Beijing time on January 19th, Zuckerberg updated his status on Instagram, revealing that "Meta Platforms is currently training Llama 3." By the end of this year, Meta Platforms will have approximately 350,000 Nvidia H100 chips, and if other GPUs are included, there will be about 600,000 H100 equivalent computing power.

Zuckerberg said in the video:

"It is becoming increasingly clear that the next generation of services needs to build comprehensive general intelligence, build the best AI assistance, AI for creators, AI for enterprises, and so on. This requires progress in various fields of AI, from reasoning to planning, from coding to memory and other cognitive abilities."

"This technology is very important, and the opportunity is so great that we should open source it and use it as widely as possible so that everyone can benefit. We are building a lot of infrastructure to support this."

"The short-term goal is to have 350,000 H100 chips by the end of this year, and if other GPUs are included, the total computing power will be equivalent to nearly 600,000 H100 chips."

Zuckerberg also mentioned that the existing two AI research efforts (FAIR and GenAI) will be combined to "support our long-term goal of building general intelligence, responsibly open sourcing it, and making it available and useful to everyone in our daily lives."

Zuckerberg has not provided more details about Llama 3. However, he also mentioned that smart glasses will be important devices to help humans communicate with AI and combine AI with the metaverse. The Ray-Ban Meta Platforms glasses have made "strong progress":

"Over time, I think many of us will be talking to AI frequently throughout the day."

"I think many of us will use glasses to do this because glasses are the ideal form for AI to see what you see and hear what you hear. The Ray-Ban Meta Platforms smart glasses - Meta AI have had a very strong start."

Just before this, OpenAI CEO Sam Altman said in an interview with the media at the Davos Forum on Wednesday that his top priority now is to launch the next generation of large-scale models, which may be called GPT-5, and compared to existing models, GPT-5 "can do more, much more." According to Altman, GPT-5 will be a multimodal model that supports speech, images, code, and videos, and will have significant updates in terms of personalization and customization features, with stronger reasoning abilities and higher accuracy. The common problem of hallucination in current large models will also be addressed in GPT-5.