still kind of think the way to attack LLM use in graded writing is to make students submit two drafts and explain their changes. has the advantage that if they’re using LLMs it teaches them how to use LLMs
May 7, 2025 13:18there is no way to LLM-proof homework and also they’re probably going to be using LLMs for the rest of their lives and careers also. make them learn how to attack a draft.
this isn’t particularly scalable but in my undergrad we had weekly tutorial sessions where you, another student, and the prof would orally critique the papers you and the other student wrote that week. that would work too.
what you are trying to teach and what they need to learn, almost always, are critical reasoning skills.
those matter more not less in the LLM age.
so square up and teach em.
[This post was deleted]
Some of this is also just cultural norms. It really should be a shameful thing to cheat via LLM. It's pretty normal to have "worse" ways of cheating as is
really at this point why not just eliminate homework, make the school day one hour longer and the kids have a "study hall" every day to do tasks that would otherwise have been homework like essays, with a pen.
with calculators, students still needed to show their work
same applies for LLMs
You can see when a document was created and last modified in Word. If you're just dropping a ctrl+v on a doc, seems like Microsoft could make and auditable change log that would make it pretty apparent if you actually wrote it or not.
Or write in class! I had a fantastic course where we spent the first 20 minutes of each class responding to a writing prompt.
I also hypothesize that rigorous use of Turnitin software will help. The more text is submitted to Turnitin, the easier it will be to use the one pattern matching software to defeat the other pattern matching software
So many professors are going to hate that this means they have to read two drafts, heh.
i do a rough draft/final draft process along with a written conversation about changes, and I’m not sure it’s the answer. the only thing i’ve found that works currently is 1-on-1 meetings to figure out how the students are using it, and that’s an arduous process requiring time and small classes
what do the failure cases look like there?
As always, it is the process as much as the product that needs to be assessed. As a teacher, my goal is to model and then provide scaffolding so that students acquire the critical thinking skills necessary to conduct research, develop a thesis, and synthesize ideas in a summative assessment. The LLM
will be used regardless of what Byzantine (and ultimately impractical to implement) measures to punish students for using it we adopt. Ultimately I need to adapt my instruction to demonstrate what LLM’s can - and cannot do wrt facilitating the research/drafting process.
Need to confront how poor writing instruction is even at the best schools (hs and college) and that includes teachers who consider themselves excellent writing instructors. A lot of them want to engage with young writers who are already good to make them great, not improve overall aptitude.
So the thing is you can just ask the LLM to do this too.
It won’t be particularly good, but an increasingly contingent academic work force has been inflating grades to save their jobs anyway.
make them write their own version of the raymond queneau literary masterpiece exercises in style
99 versions of the same incident retold in different literary styles
Yes--almost any kind of revision work will help make AI less useful/attractive.
