Even if you have open training, the corpus were compiled from millions of sources, labeled by experts manually, by AI, by outsourcing, etc. Is it "open by first principle" ?