Comment by olmo23

> While unlikely, we cannot rule out that de-identified data derived from [Buckmaster and Alpöge’s] usage of our products helped improve our models.

Well yeah, if you use the free product they train on your data, ... I thought this was widely understood?