Q: What do you call a sandwich without meat and without cheese?
A:

After a technical discussion someone I respect has sent me the output of ChatGPT (with a proper disclaimer) pointing out that the references may be worth checking out.
My verdict after one hour of fact-checking: the first reference is wrong (the method is correct but the application is not), the second one does not exist, and the third one is a generic statement with no sources.
Is there a word for "throw bad arguments at someone and let them sort them out"? Because I feel there must be.

This graph has been living rent-free in my head for over a month now. It shows the distribution of exam points as a function of AI usage and/or cheating. If you wanted to show the effects of AI in learning, this graph seems as clear as it gets - it's not everyday that one says "have you considered old-fashioned cheating instead?", but here we are.
Source: "CheatGPT? How Using AI for Programming Homework Influences Exam Results" (Markus Brenneis, to appear in the proceedings of HCII 2025)

I just learned that Roxette is touring again and I don't think I'm ready yet to move on. Perhaps in another decade or two?

AI answers for technical problems are wild: "Your <really specific error> with <program> can be caused by problems with your internet, antivirus, or access rights. Check you internet connection, the software, antivirus, firewall, access rights, server, free disk space, contact the non-existent technical support, try another computer or try the non-existent paid version of your app. Source: a page with a review of an unrelated software".
The only one missing is "turn it off and on again".

Doing ethics reviews for conferences can be exhausting.
"In this dataset we explore how to build a State-of-the-art synthetic dataset suitable for building a Torment Nexus, as seen in the Sci-Fi novel 'Don’t Create The Torment Nexus'.
(... 8 pages later ...)
Ethics statement: this dataset is for research purposes only and should not be used to build the Torment Nexus because doing so could result in harm to its users. Also our method is really limited and would never work anyway."