KingDec 1

[Opinion] AI finds errors in 90% of Wikipedia's best articles

https://blackneon.net/post/72051

How could you do this to me? - BlackNeon.net

> For one month beginning on October 5, I ran an experiment: Every day, I asked ChatGPT 5 [https://en.wikipedia.org/wiki/ChatGPT_5] (more precisely, its “Extended Thinking” version) to find an error in “Today’s featured article [https://en.wikipedia.org/wiki/Wikipedia:About_Today%27s_featured_article]”. In 28 of these 31 featured articles (90%), ChatGPT identified what I considered a valid error, often several. I have so far corrected 35 such errors.

Show threaddukemirageDec 1
legitimate use of a LLM
Show threadTreczoks
Yep. Let it flag potential problems, and have humans react to it, e.g. by reviewing and correcting things manually. AI can do a lot of things quick and efficiently, but it must be supervised like a toddler.
Dec 1 at 2:16pmWeb
Show threadbuffing_lecturerDec 2

This is an interesting idea:

The “at least one” in the prompt is deliberately aggressive, and seems likely to force hallucinations in case an article is definitely error-free. So, while the sample here (running the prompt only once against a small set of articles) would still be too small for it, it might be interesting to investigate using this prompt to produce a kind of article quality metric: If it repeatedly results only in invalid error findings (i.e. what a human reviewer  Disagrees with), that should indicate that the article is less likely to contain factual errors

File:Symbol declined.svg - Wikipedia

Show threadarchitectDec 2

So… the same as most employees but cheaper.

People here are above average and overestimate the vast majority of humanity.