Meta, the parent company overseeing social media giants Facebook and Instagram, has disclosed its ambitions to create a new artificial intelligence (AI) model intended to compete with OpenAI's cutting-edge models, according to exclusive reports by The Wall Street Journal.
Meta's objective is to develop an AI model that surpasses the capabilities of its Llama 2 model, which was launched earlier this year. Insiders familiar with the matter have revealed that Meta intends to make this new AI model "several times more powerful" than its predecessor.
What's distinctive about Meta's strategy is its decision to make the plans for this system open source. This means that Meta will allow other organizations to harness the technology to build their own AI tools for generating advanced text, analytics, and various other forms of output.
In preparation for the development of these advanced AI systems, Meta has been expanding its data center infrastructure and acquiring more of Nvidia's H100 semiconductor chips, which are highly sought after due to their exceptional power and performance.
While the Llama model was trained on 70 billion parameters, OpenAI's GPT-4, whose parameters have yet to be officially disclosed, is estimated to consist of around 1.5 trillion parameters. Meta anticipates commencing training of this large language model (LLM) in early 2024, with plans for a release in the following year. It's expected to launch after Google's forthcoming LLM Gemini.
Although Microsoft is a significant supporter of OpenAI and has partnered with Meta to make Llama 2 available on its Azure cloud computing platform, insiders suggest that Meta intends to train its upcoming models using its own infrastructure.
This development occurs in the midst of a global race among major technology firms and governments to create, deploy, and manage advanced artificial intelligence systems. For instance, the UK government recently unveiled plans to invest $130 million in high-performance chips to support the development of artificial intelligence systems. Meanwhile, China has introduced new legislation on artificial intelligence, with over 70 AI models reportedly released by the country.
