My AI assistant just hit 1,000 'I don't know' responses in a month

I was reviewing the logs for the custom AI I built for my shop's scheduling and saw it gave that exact phrase 1,023 times. It's not a simple chatbot, it's a fine-tuned model on my own work orders. The number matters because it shows the system is correctly identifying gaps in its training data instead of making up answers, which is a huge safety win. Has anyone else tracked how often their models correctly admit uncertainty instead of hallucinating?

2 comments

2 Comments

green.jenny21d ago

But isn't a thousand "I don't know" answers just a fancy way of saying your system is broken? If it can't handle basic shop questions after all that work, maybe the whole setup is wrong. You call it a safety win, but it sounds like a customer service fail waiting to happen. Why pay for an AI that needs constant hand-holding and leaves people without answers?

adam_wood9721d ago

Yeah and that's the thing, it reminds me of when my local coffee shop put in one of those fancy automated brewers. It could do everything except tell you where the restroom was, which is what half the people actually asked. @green.jenny is right that you can't just call a system safe if it leaves everyone hanging. Feels like some teams get so focused on stopping wrong answers they forget people need some answer, even if it's just a link to a human.