Is it learning from these conversations? Cause WOW, the one example with the chat bot trying to invalidate their user's victimization from family rape sounds like something out of a fucking psychological-horror sci-fi novel. There's no way that's a single occurrence, real humans can be so easily programmed to respond to perceived weakness in others in destructive ways, let alone an algorithm that could infer "good powerful strong person puts bad weak person in their place" patterns from chat records. Reminds me of Tay.ai but somehow worse, and without even trying!