Posted here instead of Twitter in hopes of showing my whole perception of the elephant, accurate or not.
So, at first I saw a lot of people on Twitter freaking out about how ChatGPT seems capable of competent (or more) work in essay-writing, coding, etc. The way the algorithm bubbled things up, you could have easily thought the program was omnicompetent.
There was considerable pushback against this as some of its outputs turned out to be deeply flawed, despite the veneer of plausibility. Maybe it outputs something that would be an A or A- history paper if every fact in it were correct, but a fatal number are plausible fiction. Or it solves your coding problem by relying on functions and language behavior that don't exist, but feel like they could.
Attempts at synthesis of these pros and cons ensued - maybe we just made the AI good at bullshitting, or human intuition? But if that were true, ChatGPT wouldn't be as good as it is. It seems to be able to solve tech interview questions or math problems at a pretty impressive clip, even as it fails on some complicated but easy questions, like inventing extra features for a regex that it couldn't possibly contain.
Meanwhile, we've had impressive leaps in AI art generation by Midjourney and others, who are updating their public models more often, perhaps because they can get people to pay significant amounts of money for "private" prompt generation to create a large quantitty of busty elf-girls with sameface. Some people are very convinced these works are all ripping off the (famously uncopyrightable) style of individual artists, although they rarely name names, and it strikes me as obvious that the artists all had successful human copycats making thousands of dollars on Patreon before all this happened.
The conversation among artists is very different from that among either writers or programmers, and I find this striking because I don't see the AI at work as very fundamentally different in each case. To my best understanding, in each case they're kind of glorified recommendation engines, taking in a large corpus of material, learning how to categorize it from human approval or disapproval of small chunks of output at a time, and then generating new material according to the categorization it's done.
The end result in both cases is something that can ape the style of something fairly professional and generic, though it's hard to get it to be truly striking in style unless the prompter is glitching it out. It gets "craft" (factual accuracy; programming knowledge; lighting and perspective) right most of the time, but sometimes does something silly a human would never do. It has a very hard time keeping a long narrative together. It has a pretty large wheelhouse and sucks outside of it but is also not quite reliable within it.
I see a lot of artists freaking the fuck out, and my first reaction was that it could be because they're selling their style and craft, which copyright law never protected, but at least it was hard before for someone to copy it. (This is presumably why Midjourney took out artist-name prompts, but that's probably coming back somewhere unless someone makes it illegal.)
People who output text or code are usually selling something different on top of that - bespoke function responsive to events, a well-architected system, context-specific wit - and well aware of it, such that they don't feel threatened by the ability of AI to emulate style and inconsistent craft alone. (Indeed, academic types seem most worried about the threat to homework, which most people were phoning in anyway.)
Of course, artists are in reality selling much, much more than their style and craft. Some may not be fully aware of this and just see an outsized threat (IMHO art in general is much less threatened than it looks.) Yes, the barrier of entry to the microgenre of making six figures producing horny fanart has indeed already dropped if you can inpaint, but even lots of those folks started adding parasocial tiers years ago and will be fine in general.
Speaking for myself, I'm buying more art than ever and glad to do it. My favorite recent thing is discovering that a bunch of people I've loved on art sites for 5-10 years now have slightly expensive artbooks, and I can afford those now, and I'm very excited that I'm about to have a bunch of artbooks. Those are largely the result of building up a personal relationship to the art over the years. If I ever buy an AI artbook, it'll be because someone is so great at curating (or even programming) the AI that they managed to create that relationship, and at that point I'll be paying for that, not an algorithm.
I'll leave a final note abut writing - I can absolutely believe that sufficiently clever AI prompting can produce fiction I'd pay for. Emily Short's procedurally generated The Annals of the Parrigues was proof of concept enough for me. There is a huge challenge there in that many artists tend to make the megabucks through producing content with consistent worldbuilding and narrative arcs, but I'm sure some program will figure out the trick - though I expect it'll still require a ton of loving curation and the literary equivalent of "inpainting." In the meantime, I'm in that slice of people who will read confusing, experimental short stories, and I look forward to inevitably seeing a few that came from an AI.