>its becoming very likely the user is becoming delusional or may engage in dangerous behavior.
Talking to AI might be the very thing that keeps those tendencies below the threshold of dangerous. Simply flagging long conversations would not be a way to deal with these problems, but AI learning how to talk to such users may be.
Talking to AI might be the very thing that keeps those tendencies below the threshold of dangerous. Simply flagging long conversations would not be a way to deal with these problems, but AI learning how to talk to such users may be.