Meta is building a new AI system to rival OpenAI's best

Meta is not satisfied with just playing catch-up.

According to a report from The Wall Street Journal citing people familiar with the matter, Meta is quietly building a new, much more powerful AI system designed to go toe-to-toe with OpenAI's most advanced model. The goal is to have it ready sometime next year.

Aiming for OpenAI's crown

Meta reportedly plans to kick off training for this massive new language model in the early months of 2024.

While the project is still in its planning phases, the internal expectations are massive. The new system is anticipated to surpass Llama 2, Meta's flagship model released back in July, by a considerable margin.

Unlike Meta's early research-only releases, Llama 2 was made widely accessible via direct download, AWS, Microsoft Azure, and Hugging Face to compete directly with ChatGPT and Google Bard.

With this upcoming powerhouse, Meta wants to help external businesses build services capable of handling intricate text generation, complex data analysis, and other advanced outputs.

Meta has so far declined to comment on the project. Meanwhile, the pressure is mounting across Big Tech. Reports also indicate Apple is quietly testing its own Ajax framework and an internal chatbot dubbed Apple GPT.

Why it matters

If Meta pulls this off, it completely changes the pricing and power dynamics of the entire AI race.

Right now, developers turn to open-weight models like Llama 2 for privacy and cost, but they still pay OpenAI for the hardest reasoning tasks. If Meta delivers a model that truly rivals OpenAI's flagship and keeps its open distribution playbook, that frontier monopoly disappears overnight.

For businesses, this could mean getting state-of-the-art analytical horsepower without handing their proprietary data over to a closed API. But there is a catch: training a model of this magnitude requires enormous compute resources and infrastructure, and it remains to be seen whether Meta can maintain its open-access philosophy as its models get dangerously capable.