I disagree, the 3-4 versions of GPT-4 I've used have face planted in terms of usefulness. No doubt in my mind this is to do with RLHF. It maybe be cost effecti

I disagree, the 3-4 versions of GPT-4 I've used have face planted in terms of usefulness. No doubt in my mind this is to do with RLHF. It maybe be cost effectiveness of running the models sure. I'm trying to be open minded.