Meta goes all out on artificial intelligence. Sometime next year, the social media giant plans to deploy an AI model capable of holding its own against ChatGPT 4. The project does not involve Microsoft, which nevertheless collaborated with Meta on Llama 2.
In the wake of Microsoft, OpenAI and Google, Meta launched into the race for artificial intelligence. The Menlo Park giant has indeed announced a host of innovations, made available to researchers and developers, in recent months. These include Llama 2, an alternative linguistic model of OpenAI’s essential GPT, Audiocraft, an AI capable of generating music, or even CM3Leon, an image generator similar to Midjourney or Adobe Firefly.
On the same subject: This open source AI model challenges ChatGPT, Google Bard and Meta’s Llama 2
Launch in 2024
Mark Zuckerberg’s firm does not intend to stop there. According to our colleagues at the Wall Street Journal, Meta is currently working on a new AI model capable of matching or even surpassing GPT-4, the latest version of the model behind ChatGPT. Citing sources close to the matter, the media affirms that the artificial intelligence is also intended to be twice as powerful as Llama 2. A working group dedicated to the project was brought together a few months ago.
The company plans to begin training this new language model from the beginning of 2024. If training goes as planned, Meta should offer AI sometime next year. The release would probably come after the announcement of Gemini, the next model from Google and intended to improve the Bard chatbot, which is expected in May 2024.
Bye bye Microsoft
To make the AI work, Meta invested in a mountain of Nvidia H100 chips. This graphics processor, sold for more than $40,000 per unit, is the first card to have been designed expressly for generative AI. It allowed Nvidia to double its turnover.
Thanks to a stock of components, Meta hopes to be able to free itself from Microsoft and develop its own infrastructure. To form Llama 2, the Menlo Park giant was indeed forced to rely on Azure, Microsoft’s cloud computing platform. Meta apparently wants to stand on his own two feet with his next language model.
An always open source approach
Internally, Mark Zuckerberg, CEO and founder of Meta, is campaigning for the future model to be offered freely to the community of researchers and developers, like the group’s other AI. It is also thanks to this approach that the open source community has evolved massively in the field of artificial intelligence. In recent months, a host of open source alternatives have emerged on the web. According to a Google engineer, open source represents the ultimate competitor to the giants engaged in the AI race.
Concretely, Zuckerberg wants any company to be able to rely on Meta’s innovations to design new chatbots, reveals the Wall Street Journal.
At the same time, Meta hopes to use generative AI to revolutionize its social networks. The company plans in particular to integrate a chatbot, available in 30 personalities, into Instagram. Finally, Meta readily suggests that AI will manage to save the metaverse, Zuckerberg’s costly fad, from collapse. The stakes are therefore very high.
Wall Street Journal