Media: Meta is developing a more powerful AI model
Date: 2023-09-11 Author: Karina Ziganova Categories: BLOCKCHAIN, BUSINESS
Meta is developing a new Large Language Model (LLM) that will compete with OpenAI's ChatGPT. WSJ reports this with reference to its sources.
It is expected that it will be “several times” more powerful than Llama 2. Let us recall that the developer presented this LLM in July 2023.
According to the publication, training of the model will begin in the first months of 2024. After this, some more time will be required for preparation, but the product will probably hit the market before 2025, the WSJ noted.
Company sources said Meta is already actively promoting the project. In particular, it plans to order even more Nvidia H100 chips to create additional data centers.
The Llama 2 model is trained with 70 billion parameters. There is no such data for the flagship ChatGPT 4. According to various expert estimates, in this case we are talking about 1.5 trillion parameters.
Meta plans to release a commercial version of its LLM. However, according to the publication's sources, the new solution will be open source.
Notably, in August 2023, user traffic on the ChatGPT app decreased by 3%. And this is not the first fall in recent months.
