e.g. compared to gpt-3.5-turbo-1106
i find 1106 would enter infinite repetitions & not do so well w RAG personally
How is gpt-3.5-turbo-0125 better?
1 Like
I haven’t had a chance to see if it’s really better or not better
I need some time to test
1 Like
gpt-3.5-turbo-0125 has performed significantly worse when tested against fine tuning gpt-3.5-turbo-1106. In one case it inserted “Sexual navigator, mama’s boy.” and “Director/frontrunner Mapper Anna” in its suggestion for a personal essay where nothing even close to this was mentioned. The generated text did not improve on provided content. It also referred to earlier questions despite being told to forget prior inputs. Its demonstrably bad like an early model with severe hallucination or possibly confusing/integrating user inputs. @luke
1 Like