I agree full transparency on data adds several other challenges. Still, even releasing the software and infrastructure aspects would be a huge step from where we are now. Also, some recent work has shown pretraining filtering to be possible and beneficial which could help mitigate some concerns of sensitive data in the datasets.