Facebook founder Mark Zuckerberg has set his sights on OpenAI. Facebook parent company Meta is developing a new artificial intelligence (AI) system that it hopes will be as powerful as the most advanced models offered by OpenAI, people familiar with the matter said. Currently, the most advanced model used by OpenAI's ChatGPT is GPT-4, which is in a leading position in the AI industry.
Informed sources said that Meta hopes that the new AI model will be ready next year, and its performance will be several times more powerful than the Llama2 model just released two months ago. This AI system is still being planned, and details may change. It will help other companies build services that provide complex text, analytics and other content output.
Meta CEO Zuckerberg established a group at the beginning of this year to accelerate the development of generative AI tools. The new AI model is the result of this group’s work. Some people familiar with the matter revealed that Meta expects to start training the new large model in early 2024. As it lags behind rivals like OpenAI in the AI race, Zuckerberg is working to ensure Meta remains a significant force in the field.
The new model is several times more powerful than Llama2TAGP H16
Meanwhile, Meta is currently building the data centers needed to train new AI large models and purchasing more H100 chips. The H100 is the most advanced Nvidia chip available for this type of AI training. While Meta has partnered with Microsoft to offer Llama2 on Azure, Microsoft's cloud computing platform, it plans to train the new model on its own infrastructure, some people familiar with the matter said.
Like Meta’s early AI products, Zuckerberg is pushing the new model to be open source so that other companies can use it to build AI tools for free. Zuckerberg will participate in a summit hosted by Senate Majority Leader Schumer on Wednesday to discuss how to deal with AI. Technology executives such as OpenAI CEO Sam Altman and Google CEO Sundar Pichai will also attend the summit.
However, the big model Meta is developing may not close the gap between it and its competitors. Meta hopes its performance will match OpenAI’s GPT-4, launched in March this year, which underpins OpenAI’s revenue-generating plans. For example, the ChatGPT commercial tool recently launched by OpenAI is based on the GPT-4 large model. OpenAI has also been courting other companies to build AI tools based on the technology in an attempt to offset the huge cost of developing advanced AI models. Meta's open source strategy has certain advantages. Zuckerberg supports open source AI models, which are popular for being cheaper and easier to adapt.