nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqy04dfr40v8nr6hpvdq5gg0ac22dplhlt9ncgpz5lp5s8jx26n40q8lfc7h (nprofile…fc7h) For the examples I have in mind:
1. Yes. The IP for the input data is unequivocably owned by the company producing the model and all necessary consents have been obtained.
2. My final sentence was: "There needs to be a conversation about whether training it represents an acceptable use of resources..." - so, it depends.
Given the much smaller number of variables generally needed for such models then the training costs *might* be acceptable, but we need to measure it.