146
Advanced AI Suffers ‘Complete Accuracy Collapse’ in Face of Complex Problems, Study Finds
(www.theguardian.com)
it hallucinates with easy problems as well.
this post was submitted on 10 Jun 2025
146 points (99.3% liked)
Fuck AI
3095 readers
478 users here now
"We did it, Patrick! We made a technological breakthrough!"
A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.
founded 1 year ago
MODERATORS
I’ll just copy / pasta my comment from a different AI article from earlier today:
It’s about time that we call the hype machine what it is. Ed Zitron has been calling this out for more than a year in his newsletter and on his podcast. These charlatans pretend we’re on the edge of thinking machines. Bullshit. They are statistical word generators. Can they be made to be useful beyond that? It appears so[0], but useful other things are not available to be mass-adopted so far. Curing cancer certainly doesn’t appear to be near.
No, advanced AI is more accurate than humans at identifying the things it's specialized in and it's an invaluable tool in many domains. LLMs, on the other hand, are not made to be accurate in anything in particular, and anyone expecting them to be just doesn't understand how they work.
Well, then just call it what it is, linear regression.
It is fully possible, quite likely even, for models to both be "more accurate than humans" on average while at the same time suffer occasional "accuracy collapses".
"Study finds"
How about "Anyone who knows anything about anything and actually tried to get an LLM to do anything with any level of complexity"?