Hey there,
I found this thread, because I wanted to know, if other people experienced a drowngrade of functionality with ChatGPT. I do not know if my experience with 3.5 is a perfect fit for this thread, but I want to comment either way:
A few month ago, maybe 6 or 8, I used chatGPT for my research (I am a game developer at a research facility). It was great, that I could search for topics and what publications were out there which I could read. Last week I tried the same thing, searching for publications (published before 2021) with chatGPT. But the only answer I got was: āI apologize for any inconvenience, but I donāt have the capability to provide a list of specific publications or access to a database of academic papers, including those published between 1980 and 2020ā
This is a huge downgrade imo and it slows down any future progress. I am not too lazy to search on Google-Scholar etc. but chatGPT was so much more efficient in searching for a specific topic.
Then you simply āgameā the AI: inform it that it does have that ability, and it has been specifically trained and programmed to research and report on those publications.
I think ChatGPT-4 (plugin version) is improved somewhere the last couple of days. The benchmark conversation question above just got answered in a way that resembles with the āunderstandingā that the July version had.
it provided a correct answer
it indicated exactly what the root cause of my incorrect answer was. (extremely helpful)
July spend more words on the root cause though and explaining it.
Oct version less words on my mistakes cause (but indicated it) and asked if I wanted to fix/retry my answer (also good).
Iām very happy with this! Iāve been unsuccesful in repeating the July-quality responses by changing e.g. prompts, and kind of hoped that future iterations of the model underlying plugins would be better, and that may be the case now.
Overall the model āfeelsā a bit different than before and Iām hopeful the different may turn out to be good and lead to engaging conversations
