The model responds based on conversations it’s trained on? It’s a bespoke response. It’s not simply showing a browsable list of responses, it’s giving particular ones.

It’s literally feeding these mentally ill people responses that a human, with the same context, would be legally culpable for.