Elektrine lite

← Feed

@Bredroll@mas.to

2026-09-24 13:49 UTC

@tansy@wandering.shop the whole 'ai detection' thing is so pointless, anything that even slightly works at 50%+ one day will be entirely 0% accurate within a week or two. These things are the latest toy for people looking for even the thinnest justification to punch downwards. its just like "most people arent disabled, so you can't be" or "your symptoms are probably just a cold, it cant be anything else". "the app is never wrong"..

Replies (1)

  • @aud@spectre.asta.lgbt 2026-09-24 19:26

    @Bredroll@mas.to @tansy@wandering.shop I don't know specifically what algorithm the various "AI detectors" use but at best they're worthless in a week or two, like you said, and at worst they're literally trying to do something impossible.  I have a suspicion that a lot of them boil down to utilizing an LLM on the text, in which case it's fundamentally nonsense regardless of what they do with the output. If that were the case, it would be... say you take a survey of a classroom and get everyone's birthdays.  You now have a distribution of birthdates for the class.  You then ask someone for their birthday, then check if that birthday is within the distribution (it is, because you built it on the classroom).  Then you accuse them of it not being their real birthday, but one they took from the distribution. Do these "AI detectors" work like that?  I don't know.  But I don't trust them *not* to do something that statistically useless.

    Open ##5064518