Zompist Bboard Again

Travis B. wrote: ↑Thu Feb 05, 2026 11:11 am When I see AI-generated pics my brain almost immediately thinks "this is probably AI-generated" -- they have a specific kind of look to them that most human-generated images lack.

I don't know what to tell you. I like that look. It's possible the trainers of those models do too, and that's why they have it.

I would pay human artists to manually draw "AI art". Then again, I'm the kind of guy who buys generic artwork being sold on the pavement. I am not an art critic with discerning tastes.

I've just had the thought that an AI creating art that is indistinguishable from human-created art is about as likely as someone creating a conlang which looks exactly like their first language.

so... very?

Torco wrote: ↑Thu Feb 05, 2026 2:50 pmso... very?

People's first conlangs tend to be in the details very similar to their first language... except then they usually read the Wiki article on Finnish cases and put 'every' case in their language just because they think cases are the thing to have (and more cases are better, amirite?)

Another classic thing to do is do something like have no tenses and only perfective versus imperfective aspect (I did that in one of my early conlangs).

And when they aren't cloning Finnish's case system or making their language a direct cipher of their native language, they are making their language relexified Latin or Sanskrit...

as an amateur conlanger, I feel slightly called out here

/ˌnɐ.ˈɾɛn.dɚ.ˌduːd/ wrote: ↑Thu Feb 05, 2026 4:44 pm as an amateur conlanger, I feel slightly called out here

we were all there, and then moved on to the kitchenlang. it's all part of the great circle of life
(cue mr eltohn john)

Travis B. wrote: ↑Thu Feb 05, 2026 8:41 amThe thing is that they're not truly passable as human-made.

Perhaps if you know exactly what defects characterize AI generated images and text and how they differ from the kind of mistakes humans would make. Most people don't have that kind of expertise, though, and much human-made content is formulaic enough that AI could easily replicate it.

malloc wrote: ↑Fri Feb 06, 2026 9:37 am

Travis B. wrote: ↑Thu Feb 05, 2026 8:41 amThe thing is that they're not truly passable as human-made.

Perhaps if you know exactly what defects characterize AI generated images and text and how they differ from the kind of mistakes humans would make. Most people don't have that kind of expertise, though, and much human-made content is formulaic enough that AI could easily replicate it.

AI-created content just has a certain look and feel, a certain je ne sais quoi, to it that human-created content tends to lack; this is not really a matter of expertise per se.

it absolutely is tho. my mum isn't young but she doesn't have substantial cognitive decline -knock on wood- and she's remarkably bad at telling what's generated and what's not. it's fair to say anyone on this board is probably in the world's top 10 or 20% by computer savyness.

Torco wrote: ↑Fri Feb 06, 2026 10:00 am it's fair to say anyone on this board is probably in the world's top 10 or 20% by computer savyness.

Apparently one of the few exceptions is malloc himself. Could that explain why he sees this stuff so differently from most of us?

I don't know how computer savy malloc is but from what i've read of his takes on llms and generative models, yeah, possibly.

Torco wrote: ↑Fri Feb 06, 2026 11:10 am I don't know how computer savy malloc is but from what i've read of his takes on llms and generative models, yeah, possibly.

He certainly takes the hype about AI at face value when many people who actually deal with generative AI deprecate it not because it is this existential threat that will replace humans or decide in a millisecond to exterminate humanity once it reaches self-awareness but precisely because of the opposite -- that it doesn't live up to the hype, with the content being generated by it being low-quality 'slop', and the threat it poses is because of its being pushed by people such as executives despite its obvious incompetence.

Travis B. wrote: ↑Fri Feb 06, 2026 11:57 amHe certainly takes the hype about AI at face value when many people who actually deal with generative AI deprecate it not because it is this existential threat that will replace humans or decide in a millisecond to exterminate humanity once it reaches self-awareness but precisely because of the opposite -- that it doesn't live up to the hype, with the content being generated by it being low-quality 'slop', and the threat it poses is because of its being pushed by people such as executives despite its obvious incompetence.

Above all, my concern is caution more than anything. Even if the current wave of AI cannot replace humans, the trajectory of AI development clearly points to humans facing an existential threat in the future. If we take decisive action now to restrain AI, we can avoid catastrophe in the future. On the other hand, if we wait until AI has developed to the point that it can overtake humanity, we will find that resisting it has become impossible. Too many times in history, from fascism to global warming, humanity has procrastinated on dealing with problems because they seemed minor at the time, only to find them overwhelming later on.

Travis B. wrote: ↑Fri Feb 06, 2026 9:56 amAI-created content just has a certain look and feel, a certain je ne sais quoi, to it that human-created content tends to lack; this is not really a matter of expertise per se.

Quite often it does, but it really depends on the subject matter and sophistication of the model. Plenty of human drawings are incredibly derivative and generic and plenty of AI images have impeccable anatomy and composition. I guarantee that even you would struggle to identify at least some drawings as AI generated.

malloc wrote: ↑Fri Feb 06, 2026 12:44 pm

Travis B. wrote: ↑Fri Feb 06, 2026 11:57 amHe certainly takes the hype about AI at face value when many people who actually deal with generative AI deprecate it not because it is this existential threat that will replace humans or decide in a millisecond to exterminate humanity once it reaches self-awareness but precisely because of the opposite -- that it doesn't live up to the hype, with the content being generated by it being low-quality 'slop', and the threat it poses is because of its being pushed by people such as executives despite its obvious incompetence.

Above all, my concern is caution more than anything. Even if the current wave of AI cannot replace humans, the trajectory of AI development clearly points to humans facing an existential threat in the future. If we take decisive action now to restrain AI, we can avoid catastrophe in the future. On the other hand, if we wait until AI has developed to the point that it can overtake humanity, we will find that resisting it has become impossible. Too many times in history, from fascism to global warming, humanity has procrastinated on dealing with problems because they seemed minor at the time, only to find them overwhelming later on.

The thing, though, is that the trajectory of AI development points at a bubble that will pop sooner or later, and will leave massively failed investments and idle and half-built data centers in its wake, not the achievement of superintelligence that will replace or exterminate humanity. This is obvious to most people who are paying attention to what is going on in this area. You, on the other hand, seem to have bought into the marketing hype of the AI techbros who want people to think that they will achieve superintelligence despite the glaringly obvious fact that the emperor has no clothes.

malloc wrote: ↑Fri Feb 06, 2026 12:44 pm

Travis B. wrote: ↑Fri Feb 06, 2026 9:56 amAI-created content just has a certain look and feel, a certain je ne sais quoi, to it that human-created content tends to lack; this is not really a matter of expertise per se.

Quite often it does, but it really depends on the subject matter and sophistication of the model. Plenty of human drawings are incredibly derivative and generic and plenty of AI images have impeccable anatomy and composition. I guarantee that even you would struggle to identify at least some drawings as AI generated.

AI-generated images have a special sort of mediocrity to them that typical derivative human-generated images tend to lack. This is not merely a matter of counting fingers or looking for distorted non-text in images, even though those are often tell-tale signs.

Travis B. wrote: ↑Fri Feb 06, 2026 12:54 pm AI-generated images have a special sort of mediocrity to them that typical derivative human-generated images tend to lack. This is not merely a matter of counting fingers or looking for distorted non-text in images, even though those are often tell-tale signs.

And AIs have nothing to tell people by means of their texts and images, and that is actually quite easy to tell.

One thing that AI's can't do is tell the difference between AI- and human-generated content -- people've tried to do this, e.g. in "AI detectors" for grading papers in school, and it's had disastrous results with genuine human-generated papers being falsely flagged as AI-generated while AI-generated papers are not detected. Of course, the people grading often don't understand this, which has resulted in incidents like when one grader at Texas A&M University failed every single final paper for a course because they simply asked ChatGPT whether it had generated said paper and it said yes (falsely) for every single one.

Travis B. wrote: ↑Fri Feb 06, 2026 12:54 pm The thing, though, is that the trajectory of AI development points at a bubble that will pop sooner or later, and will leave massively failed investments and idle and half-built data centers in its wake, not the achievement of superintelligence that will replace or exterminate humanity. This is obvious to most people who are paying attention to what is going on in this area. You, on the other hand, seem to have bought into the marketing hype of the AI techbros who want people to think that they will achieve superintelligence despite the glaringly obvious fact that the emperor has no clothes.

You're right for now. And that goes for WeepingElf's point, too. But malloc might have a little bit more of a point if you think really long-term.

IMO, when discussing "AI", it's important to distinguish between "AI" as in "the general idea and concept of artificially created intelligence" and "AI" as in "the stuff that gets marketed under the label 'AI' right now". Assuming that modern technological society keeps going, I think it's quite likely that at some point in the future, though perhaps not during the lifetime of any of us here in this thread today, we'll get "real" general purpose artificially-built intelligence. (And no, I don't really buy zompist's standard line where he seems to assert that, because he can't think of any real use for that kind of thing, no one else will want it, either. I don't really see the point of various pimp-your-vehicle style car accessories, myself, but as long as there are some people around who are not me and who do see the point of these things, some companies will keep making and selling them.)

Travis B. wrote: ↑Fri Feb 06, 2026 1:18 pm which has resulted in incidents like when one grader at Texas A&M University failed every single final paper for a course because they simply asked ChatGPT whether it had generated said paper and it said yes (falsely) for every single one.

As they say, *Head. Desk.*

Travis B. wrote: ↑Fri Feb 06, 2026 1:18 pm One thing that AI's can't do is tell the difference between AI- and human-generated content -- people've tried to do this, e.g. in "AI detectors" for grading papers in school, and it's had disastrous results with genuine human-generated papers being falsely flagged as AI-generated while AI-generated papers are not detected. Of course, the people grading often don't understand this, which has resulted in incidents like when one grader at Texas A&M University failed every single final paper for a course because they simply asked ChatGPT whether it had generated said paper and it said yes (falsely) for every single one.

"But surely with the vast amounts of money being poured into AI research one day AI's will inevitably become clever enough to be able to detect the difference infallibly!"