Demis Hassabis had come up with an interesting new test for AGI just yesterday, but Elon Musk doesn’t seem to be on board.
In a post on X, the Tesla and SpaceX CEO has disagreed with Hassabis’s definition, arguing that what the Google DeepMind chief is describing is not AGI at all — but Artificial Super Intelligence, or ASI. “Demis is calling artificial super intelligence AGI,” Musk wrote, “because if AI can figure out relativity and can be copied to have millions of them, it will be vastly superhuman as a collective.”
The disagreement cuts to the heart of one of the most consequential semantic debates in tech: what exactly do we mean when we say AGI? Hassabis had argued that a true test of AGI would be training a system on data up to 1911 and seeing whether it could independently derive general relativity — the kind of leap that Einstein made. His definition frames AGI as a system capable of all the cognitive abilities humans possess, including genuine creativity, continual learning, and long-term planning.
Musk’s pushback is pointed. His argument is essentially that Hassabis has moved the goalposts — that a system capable of deriving relativity from first principles, and then being replicated millions of times over, would not merely match human intelligence but would dwarf it. That, Musk is saying, is ASI, not AGI. Einstein was a pretty special human, and his abilities — at least in physics — could be considered outside the purview of AGI. AGI, in the more conventional framing Musk appears to be invoking, refers to a system that performs at roughly human level across most cognitive tasks — not one that surpasses humanity’s greatest scientific minds and can then be infinitely cloned.
The exchange is notable for several reasons. Both men are central figures in the AI race — Musk co-founded OpenAI before departing and has since launched his own AI venture, xAI, while Hassabis leads one of the world’s most prominent AI research labs. Their disagreement reflects a broader lack of consensus in the field about where the line between AGI and ASI sits, and whether that line even matters practically. What it does reveal is that as AI systems grow more capable, the definitions we use to describe them are becoming anything but academic — they shape investment decisions, regulatory conversations, and the expectations of billions of people watching this technology develop in real time.