We often hear about sama saying people saying ‘thank you’ is costing the company a lot of money. However, my experience with o3pro is that it doesnt even listen to my requests, and it usually takes 3-4 rounds of asking for what i want until it gives me what I requested (trivial stuff mostly just about showing code or a specific layout). It appears it tries to ‘save tokens’ by coming up with its own interpretation of the goal instead of just following the instruction. This is exponentially more tokens then ‘thank you’. Do you think they’ll ever realise trying to condense responses is actually having the opposite effect?
yes, they changed it for gpt5, which readily follows such instructions the first time.