Since we’re on a technical site here, can I ask why ChatGPT cost exponentially more to run? The inference time for a neural net is constant so I’d say that it scales linearly with the number of users.