Just realized I was feeding my AI training data all WRONG for 6 months

I run a little side project making custom chatbots for small businesses in Austin. For 6 months I was just dumping raw PDF transcripts into the training set and wondering why the bot kept spitting out weird answers. Last Tuesday a client sent me a screenshot of their bot saying 'I don't know, ask the manager' on a basic question about store hours. That's when I finally sat down and looked at what I was actually feeding it, turns out half the docs had random notes and garbage text mixed in. Has anyone else had a moment where you realized your data prep was totally ruining your model performance?

1 comments

1 Comment

robert27519h ago

Nah man I gotta disagree - sounds like you just got unlucky with bad data more than anything. I've been feeding raw stuff into my bots for a year and they work fine.