Since we're chatting about ChatGPT, someone posted this on Facebook. (I don't know who it is .. maybe someone in a sphere of influence of an acquaintance ... maybe someone even more random)
-----------------------------------------------------------
Then someone posted this in reply
It's on NPR, so have a ball.
A college student made an app to detect AI-written text : NPR
-----------------------------------------------------------
Cheating and Chat GPT
Yesterday I discovered that 8 out of 70 of my students had generated either parts of their terms papers or parts of their take-home Finals using Chat GPT; I've caught another 2 (update: 4!) this morning.
This is a far higher percentage of students cheating that I've encountered before at The College of (name withheld by me, DLD)
But after playing around with Chat GPT for a while I'm no longer worried at this--in large part because (obviously!) it's very easy to detect work that has been generated by it.
The work that Chat GPT generates is pretty distinctive. If asked about a controversial topic it will produce an anodyne paper along the lines of "Some say this, some say that, it's complicated". If asked to recount a theoretical point it will provide a short summary in neat declarative sentences. After playing with it for a while answers that have been generated by it become fairly easy to spot.
Of course, merely suspecting that a student's paper has been generated by an AI isn't enough to establish that they've cheated.
Luckily (well, for me...) it's very easy to use Chat GPT to do this.
I started off by typing in both my actual essay prompts and variants on them. Chat GPT doesn't (yet) vary its answers, and so this was sufficient to identify several cheating students. (This was made easier by students not realizing that others are using Chat GPT to cheat, and so handing in identical work!)
Other students were sneakier, and used Chat GPT's answer as the "framework" for their response. They then simply tweaked it themselves, or ran it through something like QuillBot, or moved around sentences or paragraphs with minor changes. And others had clearly typed in a slightly different prompt than the ones I'd used, and so received a different answer.
But these approaches to cheating were also simple to catch: I simply pasted their responses verbatim into Chat GPT. It then responded with something like "Yes, that's correct.... " and then offered its own response--which was that which the student had stolen and tweaked.
Reports of Chat GPT enabling students to cheat with impunity are greatly exaggerated.
As, alas, many of my students will be discovering next week.
With two of my five classes graded (70 students out of 175) I'm curious how many students will end up using this.
Then someone posted this in reply
It's on NPR, so have a ball.
A college student made an app to detect AI-written text : NPR