I’ve tried the newer 3.5 models (most recently gpt-3.5-turbo-0125) and they feel considerably dumber. To be fair, I haven’t done a blind test, but it’s obvious to me that they perform worse, at least for my personal use case. I guess it’s the additional safety training causing the subpar performance?
Given that the shutdown date is drawing near, what can we do to make it clear to openAI that we would like to keep using these older versions of the models?