Chat with the model and test live inference. Inject signals mid-generation to see seamless adaptation.