Posted on Nov 16
Building a public test set of bad AI citations. What should be in it?
ChatGPT once gave me references that looked perfect but were wrong. I only noticed after they were in my draft. That’s why I built CiteWise.
I’m putting together a public test set of known-bad citations: fake DOIs, wrong years, retracted papers, preprints with no DOI. I’ll post what it catches and what it misses.
Ever been burned by a bad reference? What did it look like?