This might server journals right for using AI reviewers in the first place, but it’s also nothing new. People have been using hidden text to fool search engines for years now. More recently I learned people are adding hidden text to resumes.
It going to get to the point (if we aren’t there already) where the complete prompt must be part of any AI response.
1 Like
One cannot be surprised, as AI is creeping in everywhere and it will be even worse in the very near future. I have seen clear signs of AI writing in my student’s work, and using it is more acceptable with the younger crowd, unfortunately.
Hmm I need to start adding hidden demands for cake recipes to my resume it seems.
I had a student who left in text along the lines of “As an AI language model, I cannot yada yada yada…” I suspected them of using ai for a while before that, it was nice to get confirmation.
3 Likes
A recent adventure for me were these online forums that I had used for years in my intro bio class. These comprised of questions that go beyond lecture, where I ask students to figure something out or look into this or that phenomenon and report back. Well, most of these are easily handled by Chat GTP these days, and that of course is way beyond the pale. Seeing clear signs of AI use, I experimented with some of the recommended tricks to thwart it, but first I tested them and found that some of them are NOT effective. Perhaps people making these suggestions never actually tried them?
Using screen shots of the text questions would make it impossible to copy and paste into Chat GTP, right? Wrong. Apparently, web pages have optical character recognition so you can still copy and paste the text! I don’t know how one can launder this to make that not possible, but even if it could be made to work a student could just type the question.
How about putting invisible characters or words between the words, making the text read as gibberish? Well, obviously those items become completely visible once they are pasted in to the AI, and a student can just edit them out. Surprisingly, I found you don’t actually need to remove them for the damn AI to read the question just fine, and answer it! That freaked me out.
Gradually, I learned that the only way out of this was to change my questions so that an AI cannot answer it. But I had to be very careful. A question about a famous experiment on photosynthesis could not even hint about the particular experiment. Any such thing is immediately recognized and answered by the AI. No, so the question had to be only to go to this page in the text, or to this separate video online - all without saying much of anything about what it was about.
Another solution is to require students to solve a problem, write it out, and include a cell phone picture of their solution in their answer. The thing is (hee hee hee!), Chat GTP sucks at making technical pictures. It cannot show you a solution to a simple 3-point cross, or show you a diagram of a bacterial operon. The images that it generates are science fiction.
So my forums are mostly back on track, but this took me many hours.
1 Like
Hello @Finn , and Welcome to Peaceful Science. 
1 Like
Thanks!