At least 80 million (3.3%) of Wikipedia's facts are inconsistent, LLMs may help finding them
A paper titled "Detecting Corpus-Level Knowledge Inconsistencies in Wikipedia with Large Language Models",^[1]^ presented earlier this month at the EMNLP conference, examines
11 Comments
Derpenheim@lemmy.zip · 27 pts · 297d
My knee jerk is no, because fuck ai, but LLMs are literally made to parse vast amounts of data quickly. The analysis and corrections needs to be done manually, but finding these errors are literally what they were originally made to do
CptBread@lemmy.world · 5 pts · 296d
Well it could have the issue of overloading volunteers with issues. Especially bad if the false positive rate is high enough.
fodor@lemmy.zip · 2 pts · 296d
That isn’t what most LLMs were designed for, though. It’s just one possible use case.
Aussiemandeus@aussie.zone · 12 pts · 297d
It can probably help make 160million
FriendBesto@lemmy.ml · 2 pts · 296d
I watch a YT channel that talks and researches History on Wales, and on that somewhat narrow topic alone, he has found some ridiculous mistakes on Wikipedia. There are tons but few people are aware as they may lack the suffiency in knowledge or background to know how wrong they are. AI will surely make that problem worse. I have caught ChatGTP to be wrong numerous times on some topics within my wheelhouse. When I tell it is wrong it "apologizes," corrects itself and just adds what I told it. Well, if it had found the data before, then why does it have to wait until it is corrected? If kids use this for school, they are so fucked.
Who wants to put glue on their pizza?
kalkulat@lemmy.world · 1 pts · 296d
msage@programming.dev · -4 pts · 297d
No.
TheLeadenSea@sh.itjust.works · 18 pts · 297d
I know everyone on Lemmy hates LLMs, but analysing large amounts of text to fond inconsistencies is actually something they're good at. Not correcting them, of course, that can be left to humans. Just finding them.
msage@programming.dev · 0 pts · 296d
It's hard to believe then their output is at best inconsistent.
artyom@piefed.social · 6 pts · 296d
That's why you have to manually review them. The biggest problem with LLMs is abuse. People just print their outputs without ever checking their validity.
msage@programming.dev · 0 pts · 296d
Is it faster than doing it all by yourself?
artyom@piefed.social · 5 pts · 296d
Doing what? Manually reviewing the entirety of Wikipedia? Absolutely.