The part they didn’t even address in the article: even if this was NOT a child operating the system, what consumer is asking for a chat bot that responds to a conversation about soccer by veering off into send nudes territory? Where is the target audience who is like ‘Oh right, was distracted by the sports, back to the pornography’, while driving?
The bot isn’t really cognizant of its own replies.
…Well, it could be with a postfilter, ranging from sanity checking its own reply, to a tiny separate safety model (of which many exist), to a basic keyword check. But apparently Twitter has no resources for such things.
But the point is this isn’t really the intended design.
The part they didn’t even address in the article: even if this was NOT a child operating the system, what consumer is asking for a chat bot that responds to a conversation about soccer by veering off into send nudes territory? Where is the target audience who is like ‘Oh right, was distracted by the sports, back to the pornography’, while driving?
The bot isn’t really cognizant of its own replies.
…Well, it could be with a postfilter, ranging from sanity checking its own reply, to a tiny separate safety model (of which many exist), to a basic keyword check. But apparently Twitter has no resources for such things.
But the point is this isn’t really the intended design.