Similar experience here. It seems that the releases one week after the plugin-introduction around mid May, drastically decreased reasoning and coding performance. I think that during my use in the last months i have explored the limits of GPT-4 in my use cases: code generation and reasoning about linear algebra. And in the last weeks i noticed, that the limits are moving. To get useful responses, i need to decrease the complexity of my questions/prompts, in such a way that they are almost not of interest any more. With current performance, I feel like i am a baby sitter or teacher, and i am paying for it.
2 Likes