Anything a human can do, I can do better

Meta boss Mark Zuckerberg has announced that he wants to create the artificial general intelligence, or AGI. This is the ultimate goal of AI, to make machines that can think and do anything humans can.

He’s not the only one: Google’s AI boss Demis Hassabis is also after the same prize. But Zuckerberg is shaking things up by moving Meta’s AI research team, FAIR, to work with the team making AI products for Meta’s apps.

Zuckerberg wants Meta’s AI breakthroughs to reach his billions of users faster. That is if he does not boot them out and refuse to let them back in with borked two factor id software. 

“We think that, to make the products we want to make, we need to make machines that are generally smart,” Zuckerberg tells me in an exclusive chat. “I want to say that because a lot of the top researchers want to work on the big challenges.”

Zuckerberg admits he can’t define “general intelligence”  or if it will be like human intelligence, better, or much better.

“The important thing is how wide it is, which is that intelligence can do all these different things where you have to use logic and gut feeling.”

He thinks the general intelligence technology will happen slowly, not all at once. He said Meta’s new, wider focus on AGI was inspired by the launch of Llama 2, its latest AI language tool, last year. The company didn’t think that making it write code was useful for how people would use it in Meta’s apps. But it’s still a good skill to have for making smarter AI, so Meta did it anyway.

External research says that Meta’s H100 chips for 2023 will be 150,000, a number that is the same as Microsoft’s and at least three times bigger than everyone else’s. When you add its Nvidia A100s and other AI chips, Meta will have a stash of almost 600,000 GPUs by the end of 2024, according to Zuckerberg.