How to deal with "lazy" GPT 4

This is a huge issue, but having used this a lot I remember using old chats uses the original model, but old chats have issues when they get too long.