we put these llmRequestDelaySeconds and responseDelaySeconds very low.. to maintain resposivness. and then make up the problems in the prompting.