Journalists reveal details of Meta’s neural network

In 2024, Meta will release a new version of its artificial intelligence language model Llama 2, which will be several times more powerful than its predecessor, according to the WSJ.

According to the publication, the AI project will help other companies create services that generate complex textual, analytical and graphical data.

The company is currently building the data centers required to operate the neural network and is also purchasing Nvidia’s H100 GPUs.

As part of the initiative, Meta has begun collaborating with Microsoft to release Llama 2 on the Microsoft Azure cloud platform. At the same time, the AI will run on Meta’s own infrastructure.

The code for the upcoming neural network will be released as open source, and developers will receive the tools needed to build applications based on the new AI.

The company expects that Llama 2 will not lag behind OpenAI’s GPT-4.

Meta introduced its first language model in February 2023. The July release of the AI was trained on 70 billion parameters.

In August, the Financial Times reported on the company’s plans to create a chatbot with support for different personalities.

As previously reported, after April–June 2023, the operating loss of Reality Labs, the metaverse unit at Meta, amounted to $3.74 billion.

Earlier, Bloomberg learned about Apple’s plans to create a competitor to ChatGPT. The company has developed a framework for creating large-language models Ajax.