Post #837070
2026-03-24 21:27 UTC
Replies (12)
-
@futtta@mastodon.social 2026-03-25 06:54
I advised her to take a screenshot of the "version history" of the document, that should at least give an indication of the work she did. The number of actions in such a revision history will be a lot smaller when the text was entirely or largely copied from an LLM's output as opposed to a typically longer list when having been written "from scratch"?
-
@futtta@mastodon.social 2026-03-26 21:24
the revision history indeed seems to be acceptable proof in some uni's; https://blogs.iu.edu/citl/2025/08/25/quick-tip-using-version-history-to-see-writing-evolve/
-
@alicefrieren@mastodon.social 2026-03-24 21:36
a student who actually did the work getting flagged as AI is the scenario these tools promised wouldn't happen. 'low false positive rate' looked very different in the demo
-
@janov911@fosstodon.org 2026-03-25 11:58
@futtta I am sooo glad my two kids already graduated… wow, this is so sick. But predictable outcome of our “progress”
-
@JubalBarca@scholar.social 2026-03-26 17:53
@futtta These days I keep multiple draft forms of things in case this comes up.
-
@brian_gettler@mas.to 2026-03-26 18:03
@futtta If I had to deal with a case like this, seeing notes or various drafts would go a long way. Also, a conversation. If the student can clearly explain their paper to me, I'm way more likely to believe they wrote it.
-
@joosteto@mamot.fr 2026-03-26 18:35
@futtta I guess that 80% means whatever tool judged the text estimates there's a 80% chance the text is AI, so a 20% chance it isn't (out of every 5 papers that get the 80% score, the tool estimates 1 would not be AI). Surely a university wouldn't want to fault the student if there's still a 20% chance the paper is hand-written? (in the case the tool is perfect).
-
@AnneMartine@social.edu.nl 2026-03-26 19:11
@futtta Ik weet niet welke software ze in Maastricht gebruiken om deze claim mee te onderbouwen, maar over het algemeen geeft dergelijke software teveel valse positieven om betrouwbaar te kunnen zijn. In Leiden gebruiken we deze software daarom niet. Op https://www.maastrichtuniversity.nl/nl/studeren/support/complaints-service-point-h%C3%A9t-loket-voor-studenten-voor-bezwaar-beroep-klacht staat o.a. hoe je bezwaar kunt aantekenen als de examencommissie niet open staat voor de aangevoerde argumenten.
-
@verymetalsite@mastodon.social 2026-03-26 19:14
@futtta Het is niet veel troost maar Maastricht University speelde het zelfs klaar om aanbevelingen te doen on how not to get hacked while being hacked themselves. Nauwelijks een aanbeveling lijkt me. Moeten bewijzen dat je je werk *niet* met AI gedaan hebt lijkt me slopend.
-
@jae@mastodon.me.uk 2026-03-26 19:23
@futtta I had a similar issue finishing up my degree. What I did was show some published articles on how many false positives the AI checking tools use. I feel bad for anyone currently in academics.
-
@FionaCraig@mastodon.scot 2026-03-26 19:43
@futtta Please forgive the link to BlueSky if it's not your thing, but I saw this today and it provides a great example for your daughter's case. Someone put the first paragraph of Mary Shelley's Frankenstein into one of these interfaces and it came back 100% AI generated. https://bsky.app/profile/jasonsanford.bsky.social/post/3mhuw4onfhk24
-
@arjen@idf.social 2026-03-26 20:36
@futtta not really, except that I know of no really reliable tool to assess AI authorship. Maybe she saved earlier versions of her text, that'd be convincing. Otherwise just challenge the teachers to show her what cues make them think it is AI generated and not authentic. They probably also do not know ;-)