I'd think if it had the ability to generate text as it does now, then iteratively apply the sorts of value judgments humans do on said output, and refine/discar

I'd think if it had the ability to generate text as it does now, then iteratively apply the sorts of value judgments humans do on said output, and refine/discard/preserve accordingly, it wouldn't necessarily be so far from "genuine" creativity. Indeed maybe it does already do something like that. I've played around a bit with getting it to write stories etc., and they do often seem quite "creative", the problem is eventually they often end up making little sense or contain fairly obvious contradictions or non-sequiturs in a way I wouldn't expect to see in the output of a typical human author (and certainly not a skilled writer). Indeed I'm not sure ChatGPT has the ability to formulate any sort of longer "story arc" within which to frame the text it generates. But I also suspect it will gradually be able to develop that capacity with future refinements.