Got roasted by a data scientist for my training data choices

Ngl, this guy from a startup in Austin tore apart my AI model because I was using scraped social media posts without checking for bias. Now I only pull from verified datasets and my accuracy jumped 15 percent - has anyone else had to rebuild their pipeline after getting called out?

2 comments

2 Comments

patricialee14d agoMost Upvoted

Got called out myself last year by some guy on Reddit who pointed out my model was basically learning to be sexist from Amazon reviews. Felt like an idiot but it forced me to rebuild everything from scratch. Ended up spending two weeks just cleaning and balancing my datasets. The accuracy jump was real though, went up about 12 percent after I fixed the bias. Now I check everything three times before I train anything. Sometimes getting roasted is the best thing that can happen to your work.

fisher.adam14d ago

Hold on, 12 percent is a pretty big jump from just fixing bias, are you sure that wasn't mostly from the data cleaning?