Loading…
3
Our support bot went from useless to actually solid in about 6 weeks and it bugs me
Six weeks ago our help desk chatbot was handing out made up refund links and we were getting about 40 angry tickets a week from it. Then the team switched it from a big general model to a smaller one plugged into our real FAQ and order lookup, and the junk answers basically stopped. Is it weird that this made me trust AI less, not more, since it only got good once we stopped letting it talk about stuff it didn't have data for?
1 comments
Log in to join the discussion
Log In1 Comment
the_robin7h ago
What was the smaller model actually allowed to answer from, just the FAQ and order system? Feels like getting good at the boring restricted stuff is the real test, and a bot only looks smart when it's kept on a short leash. Curious if you've tried it on an edge case the FAQ doesn't cover yet, cause that's where the trust question really gets tested.
5