As of mid-2025, almost 2% of all citations on papers uploaded to the Social Science Research Network are hallucinated.On arXiv, PubMed Central, and bioRxiv, the rise in hallucinations has also been substantial.https://x.com/cremieuxrecueil/status/2054575327481942049
The rate of completely made-up citations in biomedical research has gone up by almost 15 times since the advent of modern LLMs.
>the Social Science Research Networkthankfully it doesn't matter
arXiv has been the number one service for posting made up bullshit for years. Why would you expect it to be any different now that it's easier than ever to make up bullshit?
>>16975533As it becomes easier to make fake papers they will start citing each other, so the fake citation rate might start going down at some point, but the number of fake papers can still go up.
The appendix gives these 3 papers as examples using made-up citations:https://pmc.ncbi.nlm.nih.gov/articles/PMC11972227/https://pmc.ncbi.nlm.nih.gov/articles/PMC12154452/https://pmc.ncbi.nlm.nih.gov/articles/PMC12886048/
>>16975719You evidently fail to understand why AIs hallucinate citations.
