Meta, the parent company of Facebook and Instagram, is immersed in an ambitious artificial intelligence (AI) development project that seeks to rival the most advanced models of the competition, including GPT-4 from OpenAI. This new AI model, expected to be ready next year, aims to be considerably more powerful than its recent launch, Llama 2. The initiative, led by Mark Zuckerberg, CEO of Meta, aims to accelerate the development of tools Generative AI capable of producing images and text with human expressions.
The company plans to begin training the new artificial intelligence system, known as a "large language model," in early 2024. These plans are part of an effort by Zuckerberg to establish Meta as a major force in the world of AI. , in debate on technological regulation. To achieve this, the company is building the necessary data centers and acquiring Nvidia H100 AI training chips.
Unlike its collaboration with the company founded by Bill Gates to make Llama 2 available in the Microsoft Azure cloud, Meta plans to train the new model on its own infrastructure. Furthermore, Zuckerberg advocates for this new model to be open source and therefore freely available for other companies to create AI-powered tools.
The open source approach has advantages such as lower cost and adaptability, but it also poses legal and ethical challenges, such as the risk of misuse of the technology. Meta lawyers have raised concerns about the potential generation and spread of disinformation or other malicious actions with this technology.
Zuckerberg will join a group of top technology executives including Elon Musk, Google CEO Sundar Pichai, Microsoft CEO Satya Nadella at a summit hosted by US Senate Majority Leader Chuck Schumer, the Next September 13th. In it, they will discuss the challenges and regulations related to AI.
Despite Meta's aspirations, the model in development may not fully match its competitors, as it is expected to be roughly as capable as OpenAI's GPT-4, released in March.
Building large language models, like Meta's Llama 2, tends to increase in power as they are trained on more data. The most powerful version of this model, announced in July, was trained with 70 billion parameters.
While OpenAI has not revealed the size of GPT-4, it is estimated to be about 20 times larger, with 1.5 trillion parameters. Size aside, experts in this technology point out that there are other approaches to match the power of GPT-4 without having to reach its size.
In addition to Meta, other companies, such as Amazon and Apple, are also investing in the development of generative AI for use in their own services, such as virtual assistants like Alexa and Siri.
