Very considered and thoughtful piece by @Carwil on banning AI-generated text from students.
'In the end, the two main things I’m looking for in class essays are self-reflection and research. And while I can get the appearance of both from large-language model the first is a lie and the second an uncertain and fragile illusion.'
It is interesting to note the changes in university pedagogy since I finished my degree in 2001. I have distinct memories of having a social science professor declare that we undergrads did not have anywhere near the depth of understanding in the field to be promulgating our own feelings and opinions. Our essays were to demonstrate that we had read and synthesized the writing of people who had enough background to know what they were talking about.
Synthesizing a large number of experts into a short essay is something that LLMs are actually reasonably good at -- at least as good as college sophomores. It is interesting that as technology has acquired these skills, that students opinions and experience have become part of the syllabus.
@rspfau
So my plan is (1) to lean more heavily into the requirements that I think generative AI can't meet well: vivid description, effective examples, grounded persuasion. My assignments will seek this explicitly. People whose prose is limited in these ways will do less well in my class.
@rspfau
(2) As I say in the final paragraph, I'm making an appeal to not use AI in this way, especially not in the essentially deceptive maneuver where AI gets to say "I think" and "I feel" on your behalf.
@rspfau @antares @prachisrivas @Carwil Is a big issue for these llms that typically higher ed assigns so much writing that their really isn't assessment? My wife finished her doctoral program and I remember calculating how many pages of writing were graded. This seemed low compared to many profs who talk of having to grade full classes of papers over a weekend. Doing the math it seems that often it's just a few who get graded while many have their formatting checked.
@rspfau @antares @prachisrivas @Carwil Using an llm seems very disrespectful, but also assigning work that is only minimally reviewed seems to be as well. I est 8 hrs per page for tech writing, so a 10 page report is a solid 2 weeks. If I slip in 4 pages of Loren ipsum, that's significant.
@hwyaden @rspfau @antares @prachisrivas
I agree that writing into the void is a frustrating experience.
There's a place for very short response that get an up/down check, but a much larger space for writing that can be either shared to all or deeply appreciated and commented on by an expert instructor.
I've been on a journey of pushing assignments toward being a joy to read and pleasant to share for a while now.
@rspfau @hwyaden @antares @prachisrivas
LLMs can use I statements because they appear in human-written texts. This is something I mentioned in the piece.
But the "better" these simulation are the greater the lie.
This breaks down into two problems...
@rspfau @hwyaden @antares @prachisrivas
There are those who believe that LLMs offer a shortcut to actual learning. This kind of fake self-transformation doesn't help that case.
And there are those concerned w/ the ability of instructors to catch academic dishonesty. The latter is going to require more in-class work: "You two had radically different takes on the reading, can you talk about it in class?"
@rspfau @hwyaden @antares @prachisrivas
Formative assignments, those done in the process of learning, have long been game-able by copying, Greek houses'test banks, paper-writing services, and smart but unethical friends.
It's been many years since every intro calculus problem was solvable by Mathematica.
That doesn't change their value for learning. But we will have to explain that process better to students. It might involve the phrase "cheating yourself."
@antares @prachisrivas @Carwil How will @carwin enforce it?