[mgraczyk]

Hi, I used to work on ML and ran experiments like this one at FB. Just want to point out that the effect of experiments like this on an individual is smaller than the effect of making the app start 0.1 seconds faster, or retraining the network used to filter hate speech. The only reason the effect is significant is because N>650k, so tiny effect sizes can be detected.

[thatcat]

whats the variance on that effect?? doesn't a large data set like that just lead to type ii errors?