Meta, the parent company of Facebook, unveiled its latest major artificial intelligence on Thursday: Llama 3. Presented as the most powerful open-source, Meta promotes it as an AI accessible to everyone. The most curious thing is that, that same afternoon, Meta’s Chief AI Scientist, Yann LeCun, announced at the Imagination in Action conference that the company is developing an even more powerful successor that could surpass current market leaders like GPT-4 from OpenAI and Gemini from Google.
Llama 3 was launched this week in two versions: one with 8 billion parameters and another with 70 billion. However, Meta is already working on an even larger model, with over 400 billion parameters. LeCun anticipates the release of numerous variants in the coming months, although no date has been given for the launch of the most powerful model.
On Tuesday, Meta also introduced Meta AI, a new chatbot based on Llama 3, claiming to outperform previous open-source models with a similar number of parameters. While giants like Google and OpenAI keep their AI advancements “behind closed doors,” LeCun argues that open-source code progresses faster and predicts that it will accelerate the evolution of AI towards human-level intelligence.
Meta’s strategy of offering open-source AI has revitalized interest in generative AI. Llama 2, launched in July 2023, provided startups and researchers with more powerful models. Unlike GPT-4 and Gemini, which are only available through paid APIs, Meta offers its models for free, funding their development with massive investments in research and hardware.
Mark Zuckerberg, CEO of Meta, revealed in January that he is investing billions of dollars in acquiring the necessary GPUs to train AI algorithms. Meta’s ultimate goal is to develop general artificial intelligence, which can match human intellectual capacity, a goal shared by OpenAI and Google.
