Zompist Bboard Again

alice wrote: ↑Fri Sep 25, 2026 2:49 pmA useful question regarding any kind of "AI" might be: "What can AI help us do now that we couldn't do just as well before?" Thus:

- "Good" AI: difficult but important analytic things which up to now have needed a lot of time, effort, and re$ource$.
- "Bad" AI: Save money by not having to pay humans.

The problem is drawing the line between those things. The people pushing AI insist that their models fall into the first category and talk about all the wonderful things that AI can produce if we just get out of its way. The fact that it saves money by avoiding the need to pay humans is a happy side effect. Furthermore, saving time and effort is not necessarily desirable. Automating the production of art or scientific knowledge might save us considerable work, but we also lose something important when humans are no longer the ones making art or conducting research.

jal wrote: ↑Fri Sep 25, 2026 5:28 am Almost nobody says "AI slop" meaning LLM text output.

Excellent post, except this line. A lot of people have to deal with poorly written LLM output and they hate it. Samples:

https://www.forbes.com/sites/johndrake/ ... asured-it/

https://www.reuters.com/legal/litigatio ... 026-09-17/

https://www.vulture.com/article/literar ... sions.html

https://futurism.com/the-byte/editors-s ... ed-ai-slop

https://socket.dev/blog/ai-slop-polluti ... -platforms

https://nymag.com/intelligencer/article ... -spam.html

Or, my favorite: AI conferences are overwhelmed by AI slop and are trying to restrict it:

https://www.ft.com/content/54e274c5-de8 ... 25a6b1a6=1

There's a whole subgenre of concern about AI in universities. As you say, LLMs make people dumb. Students cheating is nothing new, but it's now incredibly easy. Professors are driven to find other ways to grade, such as going back to handwritten tests; meanwhile their institutions are often forcing AI on them.

zompist wrote: ↑Fri Sep 25, 2026 4:01 pm
Or, my favorite: AI conferences are overwhelmed by AI slop and are trying to restrict it:

https://www.ft.com/content/54e274c5-de8 ... 25a6b1a6=1

Brilliant.

jal wrote: ↑Fri Sep 25, 2026 5:28 am

rotting bones wrote: ↑Thu Sep 24, 2026 3:10 pmI love AI FWIW.

rotting bones wrote: ↑Thu Sep 24, 2026 3:12 pmI do hate the AI industry.

Those are one and the same, you cannot separate them.

Disproof by example:

I love this book: https://github.com/udlbook/udlbook/rele ... 9_26_C.pdf

This book is not the AI industry. It teaches AI technology while arguing against the practices of the AI industry.

zompist wrote: ↑Fri Sep 25, 2026 4:01 pmThere's a whole subgenre of concern about AI in universities. As you say, LLMs make people dumb. Students cheating is nothing new, but it's now incredibly easy. Professors are driven to find other ways to grade, such as going back to handwritten tests; meanwhile their institutions are often forcing AI on them.

Indeed. Artificial intelligence poses quite a significant threat to humans in academia and knowledge work more generally. Even if we avoid extinction or mass unemployment, we face the prospect of catastrophic stupification after outsourcing all our thinking and reasoning to these machines. Hopefully the universities will find effective ways to curb AI-powered cheating, although I worry that people will increasingly avoid higher education entirely. Why bother going to college when AI has taken all the jobs in mathematics and science and so forth?

I came across this in the first link (Forbes):

Gartenberg invoked George Orwell’s essay “Politics and the English Language,” with its examples of politicians burying meaning in abstraction. AI prose, she said, reads like those politicians: dense, vaguely impressive, hard to follow.

and found it hard not to think of Saruman, who somebody once remarked "talks like too many politicians". JRRT was ahead of his time.

zompist wrote: ↑Fri Sep 25, 2026 4:01 pmExcellent post, except this line. A lot of people have to deal with poorly written LLM output and they hate it.

Thanks, I stand corrected.


JAL

Thread on Mastodon from a radiologist on whether AI can do radiology. Short answer: no.

radiologist dude wrote: Jensen Huan[g]... seemed to suggest that the task of automating our field had essentially been completed. ...Dario Amodei claimed that it's well-established that AI can read a scan better than a radiologist...

These people have zero idea what we do. And if they have zero idea what WE do, then that means they've got zero idea what anyone else does.

I do wonder about rogue agents hacking into stuff, though.

Raphael wrote: ↑Tue Sep 29, 2026 4:00 pmI do wonder about rogue agents hacking into stuff, though.

That's a good thing, as it shows how vulnerable our critical infrastructure is. If AI agents can do it, hackers can do it (even without AI to help them). We also mustn't forget that all these "rogue agents" stories are about "escaped" experimental bot farms trained specifically for the task by the AI companies, using millions of dollars worth of tokens before they were set loose, so real-world "application" is sketchy at best, for now.


JAL

jal wrote: ↑Wed Sep 30, 2026 3:53 am

Raphael wrote: ↑Tue Sep 29, 2026 4:00 pmI do wonder about rogue agents hacking into stuff, though.

That's a good thing, as it shows how vulnerable our critical infrastructure is. If AI agents can do it, hackers can do it (even without AI to help them). We also mustn't forget that all these "rogue agents" stories are about "escaped" experimental bot farms trained specifically for the task by the AI companies, using millions of dollars worth of tokens before they were set loose, so real-world "application" is sketchy at best, for now.


JAL

Ah, thank you!