The echo chamber works by having a language model talk to itself. You supply the initial topic and it will keep discussing until you stop it. A reflection is applied between each discussion which can help guide the discussions.