we need open training data models, and not just open weights. if you cant prove that the source code wasnt trained on, how can you show that its not a copy of

we need open training data models, and not just open weights. if you cant prove that the source code wasnt trained on, how can you show that its not a copy of the copywrited original?