Certain activities, for certain highly restricted purposes, yes. Blanket data collection for a purpose unrelated to safeguarding, which can be trivially leaked, as above, no.

It should be an unpleasant necessity, highly regulated. Not just something that you do on a blanket basis because you suffer under the delusion that you might create the next magic robot (I assume the driving force here is that Facebook is very, very behind in LLM-land.)