Someone on another site mentioned that it's been a year since the LLM "AI" hype started, and it hasn't transformed anything.
This is coming just after Sports Illustrated was caught using (likely) AI-generated content with fake AI-generated headshots to pass it off as the product of real people.
I think that's interesting to think about where AI has been used, where it hasn't, and maybe why that is.
AI seems to be pretty good at helping students cheat on fairly short essay assignments.
Teachers- at least anecdotally-seem to be genuinely noticing and grappling with LLM content passed off by students as their own work. They can catch it sometimes, but not always, and detectors aren't terribly successful.
LLM-generated news articles, on the other hand, seem not to have really penetrated the news industry.
I am not 100 percent confident about this- it's always possible it's penetrated more deeply than I've been able to detect- but as best I can tell it's very much on the fringes, with high profile instances of outlets getting caught publishing either stories with made up nonsense in them or very-low-quality work (like the SI reviews).
LLMs also don't seem to be that widely used for things like routine work emails, but are being used by spammers.
Putting all this together, I have a theory of when and where LLMs are useful:
LLMs are useful to people with very low ability or knowledge to produce written content they don't care about and would find unpleasant to produce themselves.
They can't currently replace humans with medium or higher knowledge/ability, and aren't used on tasks humans find quick, easy, or pleasurable.
Why do students use them? By definition someone learning a skill or topic has low knowledge/ability, and students hate writing papers.
Why don't people use them for routine emails? Easy task for most adults, not made that much easier with automation.
Why do spammers use them? Low ability, joyless task.
Why don't most news sites? High knowledge and skill task, beyond LLM capabilities. Paying a human to fact check the LLM's work not much more efficient than paying them to write the story.
Someone in the replies mentioned codng applications being much more legit- I don't know anything about that, sounds plausible, and I should have realized on Masto that's an application there's a lot of interest in.
I was thinking about LLMs as a tool for some sort of written communication, not code. I don't know anything about the coding applications.
@0x00string@paarth OK, you're getting me at my most curmudgeonly, but we wouldn't need LLM to replace search engines if search engines still worked as well as they did 10 years ago.
It went from a world of information at your fingertips to oceans of crap with the few mostly-reliable results coming from a few websites, but almost nothing you can do to find the info you want if those few websites aren't in your results.
@0x00string@e_urq@paarth I maintain, advertisers are the Illuminati. All bad things of modern society can be traced to their seemingly limitless flow of money. Google is gunning to be chief.
@0x00string@e_urq@paarth I'm less interested in the potential future harm than I am in the amount of unpaid/underpaid labor being used to build them, and the computing resources used to operate them. Whatever interesting uses there may be now, they're being subsidized by those apocalypse cult weirdos and a ton of exploitation. If the tech needs that to survive, it's not worth it
Add comment