I'm very curious how this TOS holds up considering the fact that OpenAI likely broke many TOS in order to train their models. I can't imagine it would be possible to train their models without breaking a number of TOS. The vast amount of data used just seems impossible they could make that many deals already to cover it all. Maybe I'm wrong and it's totally possible.