I've thought about this a bit more. Disclaimer: I'm no computer scientist and no expert in machine learning etc, but still I think I know enough about it to have an educated opinion on this.
My thoughts are based on what ChatGPT can do in the area of human interaction, which is the original intention ("chat"). Looks like it's a model which has been trained with a LOT of data! And with LOT I mean it in the most possible capital writing …
LOT!
(And there's other "magic" going on I'm sure, which I cannot comprehend) What was it trained with? It was trained with real human interaction, because that's what it's supposed to imitate. It has been on the media quite a lot, that apparently this lead to ChatGPT being e.g. quite racist. So what is ChatGPT next to being impressive? It's a mirror of our (western) society. Looks like it's nearly impossible to exclude such "character traits" in an AI.
Another example of an AI that is only as good as the data it was trained with:
Spleeter. Spleeter is an AI that can extract e.g. vocals from normal audio files. I've installed it recently and it's very apparent, that it works best on well known songs and struggles quite a lot on not so popular music, music styles, soundscapes etc. E.g. with ABBA it works fantastic. So the AI can only be as good as the data it was trained with.
Now to the development of computer code, more specifically to computer code which the AxeFX and other such products are based on. Was ChatGPT also trained specifically with such data? I highly doubt it! Good simulations of such highly complex circuits with all the interactions between the components are still quite new and not really publicly available. I might be wrong here, but that's what I've learned by owning AxeFX'es and following this forum for many years.
tldr: afaik ChatGPT wasn't trained with data about modelling amps etc. in code, so I highly doubt that it can create code of a modelling that surpasses the AxeFX'es and all the other comparable modellers out there.
However, I'm not so sure about profilers. There are quite a lot of profiling plugins out there based on machine learning, which are done as a hobby project and they are very convincing. So I can imagine that profiling can be improved a lot based on good machine learning models. Maybe we can get profilers with native controls soon? This could become a real challenge for all modellers! But that's just a gut feeling, because like I said earlier: I'm no expert in this.