PSA: realized I was feeding my AI customer service bot the wrong tone data the whole time

A customer literally replied 'are you a robot trying too hard to be funny?' after my bot dropped a joke it learned from a standup comedy dataset, and that's when it clicked that I'd been using casual chat logs instead of actual brand voice transcripts for training.

2 comments

2 Comments

finley_lopez9811d ago

Bruh that Gordon Ramsay pizza bot story has me rolling. My own bot once told a customer "you look like you need a hug and a coffee" because I accidentally trained it on late-night college chat logs. The customer replied "is this bot broken or just weird?" Realized I had to delete that whole personality file and start from scratch with actual boring brand tone scripts. Machine learning is basically just learning from your own cringe mistakes in public.

grantthomas11d ago

Ha, oh man that's brutal. A buddy of mine ran a pizza place's Twitter and fed it a bunch of Gordon Ramsay insults for "personality" and a customer asked if they were being pranked by a drunk chef. He realized too late that a little sarcasm goes a long way but a lot of it just makes people angry. Took him a whole weekend to scrub the dataset and start fresh with normal customer interaction logs.