I think that's aspirational, whereas the censoring of training data can reasonably be operationalized. Whether it'll work out in practice and produce something of value, I have no clue. [Edit: sorry, I see what you mean. Yes, I concede that rebuttal.]