19
Heard a dev say AI models need data scrubbing not just bigger datasets at a meetup last night
I was at a little tech meetup in Denver Tuesday and this one engineer was talking about how the real bottleneck for AI isn't more GPUs or larger models but cleaning up all the garbage data we feed them. He said most teams just dump raw web scrapes into training and wonder why the output is flaky. Got me thinking about how I approach my own small ML projects. Has anyone here tried running a serious data cleaning pipeline before training and seen a big difference in output quality?
3 comments