Meta just dropped a bombshell: they're seeing "early signs" that their AI can teach itself. At the same time, the company's saying they'll be way more careful about publicly releasing their most cutting-edge models.
In a letter about the future of superintelligence, CEO Mark Zuckerberg revealed that Meta's new systems are gradually improving themselves without much human intervention. "The development of superintelligence is now in sight," he wrote, emphasizing that this technology comes with new risks that require strict safety measures.
Zuckerberg hinted that Meta's most powerful models might not be released publicly, unlike previous Llama versions. He said openness remains a key part of their strategy, but the company "will carefully consider what they make public" to avoid helping competitors and putting users at risk.
Meta is going all-in on superintelligence: The company is setting up its own Meta Superintelligence Labs, recruiting top researchers with massive compensation packages reaching hundreds of millions of USD, and plans to spend around $69 billion on AI development and infrastructure in 2025.
