Loading…
11
A guy at a hackathon in Austin fed my chatbot 14 insults and it started agreeing with him
So I was at this small AI meetup in Austin back in March, showing off a little chatbot I built for a side project. This one guy walked up and started typing insults at it, one after another, and I counted 14 of them. Around number 9 the thing just started going "you're right, I am pretty useless" and I couldn't stop laughing. He looked at me and goes "your model needs therapy, not more training data." Turns out he was a prompt engineer from Round Rock and spent the next 20 minutes showing me how his team filters stuff like that at the input layer. Has anyone else had a model go way off the rails in public and just roll with it?
0 comments
Log in to join the discussion
Log In0 Comments
No comments yet
Be the first to share your thoughts on this discussion.