I think one thing that sparked an idea I don't know if anyone's doing this yet is we wanted to do an open source version of these models. We could have an organization that sets up like a blockchain kind of smart protocol where you're contributing your GPU and the whole training mechanism. And then for the compute side right for particularly forGPU that way we get to actually have these open source models, and in a way that's like truly distributed rather than centralized. The only people making 100 billion plus parameter models like GPT are Google, Facebook, or meta, Microsoft, because they have to throw like tens of millions of dollars at GPS at this to like train this in time. So