You're welcome. Thanks for reporting. It's pretty confusing so maybe we should change it :)
They allow you to configure chat participants (a model + params like context or temp) and then each AI answers each question independently in-line so you can compare and remix outputs.