I think these new definitions are mostly PR manoeuvring; I doubt very much they are what most people think of when they envisage a human-level intelligence of this nature.
Maybe, but regardless of the motivation for their definition, it only makes sense to interpret OpenAI’s claims and ambitions about AGI in the context of their own definition of AGI. If you prefer your own definition, that’s fine! Substitute “median-human-level automaton” or some other less PR-friendly term in OpenAI’s claims about AGI. But while OpenAI say that their goal is (paraphrasing again) to create an aligned AGI, as far as I can tell, they don’t expect or intend the system they create to be sentient.