Facebook and Instagram parent company Meta is working on a new AI system that could be as powerful as OpenAI, the startup behind ChatGPT. The system is intended to help other companies build services that produce sophisticated text, analysis and other output, according to The Wall Street Journal. Plans could change, but Meta wants to have the system ready by next year.
Meta declined to comment.
Meta is trying to regain its footing in the AI technology race, which broke into a full sprint early this year when ChatGPT went mainstream and showed the public what a generative AI system could create. This rabid interest, however, has been tempered a bit by industry experts who fear that these systems could have dangerous unintended consequences. The Big Tech companies have largely dismissed these concerns, and have focused on catching with OpenAI, with the belief that having the most effective AI system means potentially creating the tightest relationship with consumers.
Meta Chief Executive Mark Zuckerberg is putting more resources towards this goal by creating a group to accelerate the development of generative AI tools that can mimic human expressions, the WSJ reported. The company could start testing the new large language model in early 2024. It’s uncertain whether the new system will give Meta the edge it’s looking for, especially when it’s likely the new system would release after Google’s advanced large language model, Gemini.
If all goes to plan, the new system could be several times more powerful than Llama 2, the AI model the company released in July. The report said Meta is building data centers to handle the job and acquiring H100s, Nvidia’s most advanced chips. The new model would train on Meta’s own infrastructure. Zuckerberg intends for the system to be open sourced so other companies can use it to build their own AI tools for free. While an open-source model is more adaptable at a lower cost, it also comes with risk of potential copyright infringement and being used to spread misinformation.
Llama 2, or Large Language Model Meta Artificial Intelligence, was trained on 40% more data than its predecessor, Llama 1, which launched in February according to the model’s website. The open-source large language model is available for free for research and commercial use. At the time, Meta teamed up with Microsoft to make Llama 2 available on Azure. It’s also offered through Amazon Web Services and other providers.
“We support an open innovation approach to AI,” Meta’s statement reads on its website. “Responsible and open innovation gives us all a stake in the AI development process, bringing visibility, scrutiny and trust to these technologies.”
Around the time of Llama 2’s release, Meta and Qualcomm announced a partnership to enable on-device AI applications using the model. Qualcomm wants to make Llama 2 available on flagship smartphones and PCs starting next year. Having on-device AI helps increase user privacy, address security preferences, make apps more reliable and enable personalization at less cost to the developer in comparison to the sole use of cloud-based AI implementation and services, according to Qualcomm.
Upcoming summit to discuss how AI is handled.