2 Comments
User's avatar
Siebe's avatar

Hmm, the "AI promotes retracted articles and doesn't flag methodological issues" evaluated ChatGPT 4o-mini (very outdated) and only gave it title + abstract. Their prompt seems alright I guess.

https://onlinelibrary.wiley.com/doi/10.1002/leap.2018

Ironically, this is pretty bad research! (Or at least, just very outdated due to long publication timelines - a really common issue in science of AI).

Paul Litvak's avatar

ha, that is ironic. yes, this is indeed a problem. i know some folks who are working on different probes to test for this with more current models and they are finding similar issues.