Meta would like to develop a new AI model that should be able to do at least as much, and ideally even more, than GPT-4 from OpenAI.

According to the Wall Street Journal, Meta is currently preparing its next step in the AI ​​race. The company has already secured the necessary supply of Nvidia H100 GPUs and is also increasing its computing capacity in order to be more independent from Microsoft. Training of the model will begin in 2024.

Better than GPT-4

Meta’s ambitions are clear: to be better than OpenAI. The company claims to be able to develop an AI model that can at least compete with GPT-4, the benchmark to beat in the AI ​​world. Meta wants to develop chatbot applications with this model for its own platforms, but, like OpenAI, also offers them to other companies for payment or without payment. A chatbot for Instagram that can switch between thirty different personalities is already being planned.

Meta is certainly no stranger to the AI ​​industry. The company’s most powerful model is LLama 2, which it now offers fully open source and has also been converted into a code generator. Meta also experiments extensively with technologies for converting text into (moving) images. This fits with the company’s new strategy. Zuckerberg has come to the realization that big money cannot yet be made in the Metaverse, but with AI.

Shortage of chips

The race between technology companies to develop the best AI model has consequences. Meta isn’t the only one wanting to surpass OpenAI. Microsoft, Google, Amazon and Apple all want to have the best AI model. This requires special hardware. TSMC is gradually no longer able to keep up with the production of chips and therefore Nvidia GPUs are in danger of becoming scarce. This will be particularly detrimental to smaller companies that no longer have enough GPUs to develop AI applications.