He's (probably) referencing Turing's 1950 paper [1]. The whole point of "The Imitation Game" is that the AI ought be able to imitate any arbitrary type of person. Turing's example was the machine pretending to be a woman, and its up the the investigator to determine which person they're speaking to is a real woman and which is the AI. The implication of this is that the machine ought be able to be completely indistinguishable from any type of person, including those who might do well on this test or that.

Somehow this test got dumbed down over time, probably in an effort to try to pass it, into an investigator having to decide which of two sides is an AI - with no other information to go on. That's a comparatively trivial test to pass (for the "AI"), as it merely requires creating a passable chatbot. Imitation is an exceptional challenge as it does implicitly require the ability to imitate anybody, whether a professional athlete, a man who scored perfectly on the LSAT, or even something as specific as "John Carmack."

[1] - https://www.espace-turing.fr/IMG/pdf/Computing_Machinery_and...

You're adding the implication that it needs to beat the "best" to be passable, if you could pay an agi 1% of the salary of the "best" to get an average quality, is that not accomplishing the goal without moving the goalposts?

By their marketing along, OpenAI has moved the goalposts more than anything else. They've managed to lower the bar of agi from "artificial general intelligence" to "regurgitates and recombines to form passable outputs with enough labelled training data".