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]](https://en.wikipedia.org/wiki/Wikipedia:Wikipedia_Signpost/2025-12-01/Recent_research#cite_note-1) presented earlier this month at the EMNLP conference, examines
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.
It’s hard to believe then their output is at best inconsistent.
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.
Is it faster than doing it all by yourself?
Doing what? Manually reviewing the entirety of Wikipedia? Absolutely.