Using GPT-3.5 to render visual interfaces within a conversation

I don't understand what is happening here. How is working with the audio for example?