<lunbe> it's [chatgpt] already more reliable than the average human
on a number of meaningful metrics & there's no reason to
think it's near any ceilings
The above might be seen as something of a pro-AI view. However, no reason would include the risk of yet another AI winter, warnings about bad actors abusing AI, a web that is flooded with output that is fed back to... oh, Spammerville? Yeah, you don't want to go there. There may also be computational or other technological limits that the "not small" models soon or have already run into. How do small humans learn so much running on so little? What other problems could there be?
So the various ceilings one might entertain involve the realms economic, Machiavellian, regulatory, diminishing returns, technological, other. Maybe all those bold predictions of future earnings (recall that John McCarthy invented AI, to get money) mostly fail to materialize, and the lemmings get spooked and herd their investments off elsewhere. Again. How long will the chatgpt honeymoon last? Take fuzzy logic. That AI ended up in rice cookers. Perhaps chatgpt will be most useful in a similarly modest role, maybe to suckle C-suites with sweet summaries, or to bang out bad boilerplate?
I was informed that articles need to have images with which to amuse or educate, so here imagine, reader, if you can, the image of Pandora pouring from a pithos.
Machiavellian... realpolitik so thick one could cut it, and some may gladden should American hegemony hiccup sooner rather than later. AI could help a Russian troll farm, or really now anyone can generate piles of prose to tucker into. Who knows what alarming things may result, and then what ceilings may appear? Or, for those with more modest aims, various interest groups that may or may not be impacted by AI will lobby against it. What else do you do when your piece of the pie has been getting smaller over the last few decades of progress? Now, regulatory scrutiny is unlikely in a worker's paradise such as the United States, though perhaps more progressive regimes in Europe or Asia will put various ceilings forth? (Or the AI wage slave labourers could unionize...)
https://dl.acm.org/doi/10.1145/3442188.3445922#d3448975e1
More likely, the powers that be may not wish to be cut from their well-earned profits, and perhaps would take action against AI that threatens. So that's another ceiling to be wary of.
On the diminishing returns front, well, that will become apparent once the honeymoon is over and the service has been productized too far (an internet connected chatgpt enabled smart toilet?) and the profits turn out to (surprise!) not be quite so large as expected, and all the usual Wall Street antics ensue. Diminishing returns also includes pollution from the axe grinders, SEO spam, or in general humans humaning on the internet. This may lower engagement; humans going else and doing other would tend to put a ceiling on profits. One might imagine sites becoming mostly chatbots talking to chatbots and the ad money flowing in, at least until the tide flows out.
https://pluralistic.net/2023/01/21/potemkin-ai/#hey-guys
However, one had best buckle their metaphorical seat belt because we still live in interesting times:
"Blame not the age, it is now too late to stop; it is in the grip of inventions now, and has to go on; we cannot stop content with mustard-gas; it is the age of Progress, and our motto is Onwards."
-- "The Chronicles of Rodriguez". Lord Dunsany. 1922.