Finally created an official policy for AI/LLMs in class, both for writing and for coding (https://datavizf24.classes.andrewheiss.com/resource/ai-bs.html)
AI, LLMs, and BS – Data Visualization with R

Use R, ggplot2, and the principles of graphic design to create beautiful and truthful visualizations of data

Data Visualization with R
AI, LLMs, and BS – Data Visualization with R

Use R, ggplot2, and the principles of graphic design to create beautiful and truthful visualizations of data

Data Visualization with R
@andrew Isn't mean(x, na.rm = TRUE) accurate R code?
@smach Yep, it is, but it's unusual to see it on students' very first exercise before we've talked about it or covered it
@smach Like, that's all completely valid code and it works, but they're learning tidyverse-flavored R, so it's unexpected to see read.csv, and the print() is superfluous, etc. I typically get 50% of assignments with that identical code, thanks to ChatGPT
@andrew @smach I point out that the code students produce needs to be the code that I talked about in lecture, and I don't talk about read.csv at all (and not about na.rm in mean() until later), so they are *wrong* in the context of my course. (I started doing this during covid, when I wanted some evidence that the students were actually learning from me and not from some guy on chegg.)
@nxskok @andrew Understand and that makes sense. Although in the real world, people often look up answers to coding questions -- paper books and coding "cookbooks" in the old days, then online, now LLMs.
@smach @nxskok Yeah, and that's totally fine - it's just that when brand new beginner students copy/paste directly from ChatGPT like that, I've found that it's detrimental to learning the stuff. LLMs (esp. GH Copilot) are a great resource when you know what you're doing

@andrew @smach this I think is the key observation: students learning the stuff don't have enough baseline knowledge to understand whether the stuff coming back from chatgpt (or google or chegg) is any good.

I make a point that everything in my courses can be done using the course materials, with the exception of a few places where I explicitly ask students to look up something and report back on what they find (with a citation to where they found what they were looking for).

@andrew it’s wild how much of a tell the comments are
@grrrck “Certainly! Here’s how you do it:” leaking over published academic prose is a bleak, bleak thing to witness.
@andrew these are great! I may be pointing my students at these.
@andrew haha, I've been using R for 20 years, and the third image is exactly the sort of code I write, with the exception that I wouldn't use read.csv. Comments on everything, explicit print statements, I haven't started using the new pipe because muscle memory, the whole deal.
@andrew Thanks, I may steal some of this!
@SamCrawley @andrew Thanks for sharing, I'll use some bits of it also on my syllabus!
@andrew thanks for this excellent example!

@andrew

This might interest you.

I wrote it to basically plead with my university to get off the bandwagon.

They didn't really listen: you cannot convince people to understand AI when their salary depends on the AI hype

https://github.com/paezha/University-AI-Panel

GitHub - paezha/University-AI-Panel

Contribute to paezha/University-AI-Panel development by creating an account on GitHub.

GitHub
@andrew @matherion This is awesome! Might need to dot add something similar to my courses.