My friend in Austin said to feed an AI model more data, not more code, for a project

I was building a small tool to sort customer emails and kept adding complex rules to the code. My friend, who works with machine learning, told me to stop and just gather 500 more example emails instead. I spent a week collecting and labeling them, then retrained the model. The new version sorted emails with 95% accuracy, way better than my old code. Has anyone else found that more simple data beats a clever algorithm?

2 comments

2 Comments

grant_palmer11d ago

Yeah, that's the whole game right there. I was trying to filter spam messages and kept adding more code logic, which just made it break in new ways. Finally just sat down and made a huge, clean set of "spam" and "not spam" examples. Threw out all my fancy code and used a basic model. The difference was insane. Good data is like giving the model a clear picture instead of a blurry one. All the smart code in the world can't fix a blurry picture.

the_david11d ago

Hold on, that's not quite the full story. Your friend is right that more data often helps, but it's not just any data. You said you spent a week labeling them. That's the key part you're glossing over. Clean, well-labeled data is what fixes these problems. Feeding a model 500 messy examples would do nothing. The real lesson is that your time is better spent fixing the data, not writing endless rules.