@Sangy Its def irritating. A few causes Latency, Response delay, Interruption, and we add a few perimeters into the prompt with instructions if you add in [Square Bracket = Instruction, do not read out loud] it gives the model a special way to understand things and we can do things like (smile, while talking) [Pause, Wait for response] , or we could add some emotion In your use case it's going to be to wait for the user to respond, and turn up the interruption setting to between 3 and 5 words Use GPT -4 model to test the system messages to see how they perform initially, then each AI LLM actually has a different prompt and system message structure or format that they follow you can use ChatGPT to help you write the system message into a md file --- which tells the "machine" its related to what it's doing and not just plain text