Reference hallucination rates have improved considerably
One way to spot ChatGPT-generated manuscripts (for now): non-existent references
Some worse than others, ranging from fully correct to just a wrong date to a random (but real-sounding) mash up of authors, titles, and journals
Caveat lector