Heard a ML engineer say "stop feeding AI junk data" and it clicked for me

I was at a coffee shop last Tuesday and overheard two guys talking about training models. One of them said that most people spend months tweaking parameters when the real problem is their dataset is garbage. He said he cleaned up a client's data for a weekend and the model accuracy jumped from 68% to 91%. I've been spending all my time on fine tuning and never once thought about the data quality. Now I'm going back through my training sets and finding so much noise. Has anyone else realized they were polishing a turd dataset?

1 comments

1 Comment

margaret_lane9d ago

Jumped from 68 to 91 just from cleaning data? That's insane.