A lot of folks in my networks have been chatting about OpenAI / ChatGPT -- the latest GPT-3 release. Based on the examples I've been seeing, the coherence level has increased dramatically from previous releases, such that I probably wouldn't be able to distinguish a paper that was substantially written by an AI if a student were to submit it.
Here, for instance is the entirety of the prompt I submitted to ChatGPT:
There are clearly still some glitches here -- a student who submitted this as-is would probably not get an 'A' -- and I would have several question marks and "redundant phrasing" comments in the margins. But it's definitely both impressive and a little concerning how good this is.