Yep, it is dangerous. If you're not quantifying uncertainty, you can't make safe predictions. I think this is reason for the obsession with "data cleaning" in t

Yep, it is dangerous. If you're not quantifying uncertainty, you can't make safe predictions. I think this is reason for the obsession with "data cleaning" in the ML community, "outliers" aka rare observations sink general models.