Heard a data scientist say 'garbage in, garbage out' at a conference last week and it finally clicked for me

I was sitting in the back of a workshop in Portland when this guy just casually dropped that phrase while talking about training datasets. I've read it online before but hearing it in person with real examples made it stick. He showed how a model trained on messy data with wrong labels got 60% accuracy, then after cleaning it up it jumped to 92%. Made me realize I need to spend way more time on my own data prep before jumping into the flashy AI tools. Has anyone else had a moment where a simple saying finally made sense in a new way?

2 comments

2 Comments

ivanb4117d ago

Actually that's a pretty common saying but people often get it backwards. It's not just about garbage data going in, it's also about the model not being able to fix bad data on its own. A lot of folks think AI can somehow figure out what you meant even if your training data is a mess, but that's not how it works. I saw a guy at a hackathon once who spent weeks tweaking his neural network architecture but his labels were all wrong and he couldn't figure out why his results were terrible. Clean data first, then worry about the fancy algorithms, that's the real lesson people miss from that saying.

harper_owens17d ago

yo @ivanb41 hit the nail on the head. i wasted a solid month once trying to get a model to work with sloppy data and it just kept giving me nonsense outputs. cleaned up my dataset and suddenly everything clicked after like two days of tweaking.