Philipp Steinkrüger

Reading this semester's essays, I note several common features that I haven't seen like this in past years:

- very few typos and grammatical oddities (which are otherwise common in first year students)
- very repetitive

I think I'm seeing Chat-GPT at work...

Mar 23, 2023 at 1:25pmWeb
Show threadPhilipp SteinkrügerMar 23, 2023
I just asked Chat-GPT to write an essay for the question I asked the students to write about and it's definitely using the same patterns and expressions that I see all over the student essays. So I fear I have to conclude that something like 75% of submissions are Chat-GPT plagiarized.
Show threadPhilipp SteinkrügerMar 23, 2023
Another hint: students usually are very fond of providing tables of content, even for the shortest of essays. But this year, very few essays have TOCs. But Chat-GPT also doesn't give TOCs...
Show threadPhilipp SteinkrügerMar 23, 2023

After five Chat-GPT essays in a row finally a real original essay:

- 10 item TOC for a 9 page essay
- "Oral" sentence structure in the first paragraphs
- typos and stylistic errors down to non-existent words

Show threadRick 'n RollMar 23, 2023
@philippsteinkrueger
The real deal.
Show threadTatjana SchefflerMar 23, 2023
@philippsteinkrueger 😭Tears of gratitude, B+ “thank you for your effort!”
Show threadPhilipp SteinkrügerMar 23, 2023
@tschfflr Totally, my first thought!
Show threadRev. GothAliceMar 23, 2023

@philippsteinkrueger Grammar itself will be a giveaway for a while. There are subtle patterns in human dialog that are dead giveaways, changes in which become highly noticeable. We've evolved to feel subtle wrongness strongly; the "uncanny valley" exists for a reason.

It's easy for those at work to notice if I've written something, or if someone else wrote it, even if I dictated. Comma use is a huge one. 😜

AIs don't tend to "flourish". Just the meat 'n potatoes of facts.

Show threadwzhkevinMar 24, 2023
@philippsteinkrueger This is depressing to read. That apparently this many students are prepared to cheat given the chance. And also that they didn’t even bother to try harder to hide the fact.
Show threadPhilipp SteinkrügerMar 24, 2023
@wzhkevin I think it’s because they have heard about Chat-GPT and they give it a try and they are impressed by the output and so they don’t even think that someone who reads lots of essays might pick up on the differences with human created content… they’ll probably improve their efforts if given the chance…
Show threadRobert Weißgraeber🇪🇺Mar 25, 2023
@philippsteinkrueger there is also deepl.com/write - grammar check on steroids.
Show threadRobert ViefMar 23, 2023
@philippsteinkrueger just test for it. There is e.g. the GPT Detector on this website https://www.writefull.com/
Writefull

Writefull is automated writing and proofreading for academics, using the most advanced AI to date.

Writefull
Show threadPhilipp SteinkrügerMar 23, 2023
@robertvief Thanks, will try!
Show threadRobert ViefMar 23, 2023
@philippsteinkrueger Unfortunately, the free version has some limitations (maximum of words per query and limit of queries per day), but still useful to test the supicious essays...
Show threadPhilipp SteinkrügerMar 23, 2023
@robertvief I'm afraid I can't even try, as I neither have Word nor Overleaf...
Show threadThe Febrile MuseMar 23, 2023

@philippsteinkrueger Call in each student and have them recite some of essay, from memory?

If they worked hard on it, they'll know their own words. If they can't do this, allow them to fess-up?

Show threadPhilipp SteinkrügerMar 23, 2023
@CMDoran I thought about that. There’s no way these essays are not AI generated, so they won’t be able to engage in any conversation about the content. The question is why I would do that. I can’t fail them because of it as they could always claim that they had forgotten about the content (unlikely, I know) and so the only thing I stand to win is more work…
Show threadPet PharmerMar 24, 2023
@philippsteinkrueger Humans are just one hot mess of a species. A brain is a terrible thing to waste.