in fact, i'd be very surprised if many large platforms weren't already doing this; can't really get meaningful results otherwise
---
RT @pee_zombie
@eigenrobot if I were a Twitter ML engineer trying to crowdsource dataset labeling I would simply insert items with known correct labels & classify contributors based on the accuracy of their labels for them, & then weigh contributions using this contributor confidence score, simple as
https://twitter.com/pee_zombie/status/1385436526071500802