I am using the assistant api, pointing to the gpt-4-0125-preview model.
the model seems to be thinking that its’ knowledge cutoff date is still April 2023, contrary to what the documentation said about December 2023.
which one is correct?
I am using the assistant api, pointing to the gpt-4-0125-preview model.
the model seems to be thinking that its’ knowledge cutoff date is still April 2023, contrary to what the documentation said about December 2023.
which one is correct?
Probably the documentation. We’ve had the same discussion here:
TL;DR: never trust a model
It feels like there should be an error in the model.
Maybe the knowledge is not updated, but “laziness” is corrected.
Well, we don’t have information what type of additional information it was trained on. Based on gpt-4-0125-preview model responses it does not appear to be general “world knowledge” but it’s possible that other more specific data was updated. I agree though, it’s a bit confusing.
I never managed to get December back in the response so far …
EDITED: with @_j 's strategy on the system prompt override it works!