GPT-4o has this problem that seems to over compensate for 4 turbos issues.
Sure its not “lazy” but its also way too ambitious with its responses. Especially in GPTs actions.
GPT-4 and turbo was pretty interactive and careful with what it did, and were i to say “write an updated version of ___.py” or whatever it would give me placeholders that would irritate me.
GPT-4o does not - but it also fails to engage on a project, it fails to be as interactive as gpt-4o.
Honestly what i think is a perfect solution is to take the 4 turbo model and have it flip between the 4o model- depending on the context.
like have the two models share a context in which the 4 turbo decides.
make 4o its… assistant so to speak.