> Moreover, I doubt any human programmer will have troubles with the "Beginner" problems, regardless of their specialization.

FWIW, I think you overestimate humans. Particularly those who haven't actively been practicing leetcode-style coding. E.g., many of the problems are specified in "competition language", not "human-friendly language" (where "human-friendly", e.g., is something you'd be happy to see in a design doc). (Should that matter to GPT-4? I dunno.)

I do think it is fair though to say that, with some baseline level of practice (which is potentially the relevant comparison point), a lot of people would probably nail the "beginner" tests.