Even, for the sake of argument, if these AI detectors did work, suspending a student for a 98% probably of AI detection on something they submitted one of that year would mean, if you have a big university with thousands of students that mostly are behaving honestly, that you suspend potentially dozens of completely innocent ones every year. The fact that these systems dont work (and honestly, fundamentally cant work given that there's no 30 page sequence of words that its possible for a bot to write but impossible for a human to write, and a bot trained on human text wouldn't likely generate such a text if there was) is a problem, but considering a 1 in 50 chance of someone being honest as effectively zero when considering a life altering punishment is a problem regardless of the circumstances surrounding it, that's just not as low a chance as it feels.
Oh I’m really pissed about the OOP. Anyone know if it got sorted? This kind of shit will only get worse over time. My own English has always been a little mechanical and I have been accused of pasting machine output as my own writing by people who (I say this with full “humility”, I don’t think my writing is extra beautiful but those people think that about slop) can’t distinguish the two
I agree the tools aren’t reliable but the gettysburg thing isn’t really a gottem imo; if that text showed up in a paper without attribution it would indeed be plagiarism. Using the tool ON original sources like famous speeches should always return a plagiarism flag.
7 Comments
CarbonIceDragon@pawb.social · 24 pts · 81d
Even, for the sake of argument, if these AI detectors did work, suspending a student for a 98% probably of AI detection on something they submitted one of that year would mean, if you have a big university with thousands of students that mostly are behaving honestly, that you suspend potentially dozens of completely innocent ones every year. The fact that these systems dont work (and honestly, fundamentally cant work given that there's no 30 page sequence of words that its possible for a bot to write but impossible for a human to write, and a bot trained on human text wouldn't likely generate such a text if there was) is a problem, but considering a 1 in 50 chance of someone being honest as effectively zero when considering a life altering punishment is a problem regardless of the circumstances surrounding it, that's just not as low a chance as it feels.
ggtdbz@lemmy.dbzer0.com · 5 pts · 81d
Oh I’m really pissed about the OOP. Anyone know if it got sorted? This kind of shit will only get worse over time. My own English has always been a little mechanical and I have been accused of pasting machine output as my own writing by people who (I say this with full “humility”, I don’t think my writing is extra beautiful but those people think that about slop) can’t distinguish the two
WorldsDumbestMan@lemmy.today · 4 pts · 81d
Yeah, sounds like they just pick people to fail, is all.
reallykindasorta@slrpnk.net · 2 pts · 81d
I agree the tools aren’t reliable but the gettysburg thing isn’t really a gottem imo; if that text showed up in a paper without attribution it would indeed be plagiarism. Using the tool ON original sources like famous speeches should always return a plagiarism flag.
StopTech@lemmy.today · 19 pts · 81d
It was flagged as AI-generated, not plagiarism. But it may be that the creators of the tool conflate the two.
figjam@midwest.social · 2 pts · 81d
The address was 100% included in training data because it is an important document that is in the public record.
reallykindasorta@slrpnk.net · 1 pts · 81d
True, definitely mislabeled at the very least. I was picturing the dashboards on Canvas and similar when I commented