The AGI Levels Are for Investors, Not Scientists

The OpenAI "levels of AGI" framework — Level 1 through Level 5, Chatbots to Superintelligence, announced last month with great seriousness — is starting to feel less like a scientific taxonomy and more like a roadmap drawn for a very specific audience that does not care about Turing tests.

The speculation goes like this: OpenAI is heading toward an IPO. Wall Street needs a story. "We make a chatbot" is not a story that justifies the valuation. "We are climbing a defined ladder toward superintelligence and we are currently on rung two" — that is a story. That is a TAM slide. That is a reason to hold through the dip.

Nobody planned this in a cynical back room, probably. These things don't need to be planned. You just need a company that wants to raise money and a framework that happens to be infinitely legible to people who price futures contracts. The framework writes itself.

The tell is that the levels are defined just vaguely enough to be unprovable but just specifically enough to sound like milestones. Level 2 is "Reasoners" — AI that can solve problems without human help as well as a human with a PhD. Nobody can measure this. Nobody will ever definitively achieve it. But it will definitely be announced.

This is all speculation. Could be completely wrong. But it's the most interesting read on the AGI theater that's been circulating, and it has the uncomfortable quality of things that are probably at least partially true.