I'm rapping now. >> All right, what is going on? I mean, I guess this like >> Oh, I should be taking notes for this. >> I guess there's like uh quite a few things there, right? Cuz I think a lot of these um speech speech models themselves um are trained to really be sort of voice assistants is one thing. And um and that's why the model itself just really wants to sort of wait for you to say something. And if you interrupt, it really tries to be quiet so that you can listen to what the user is saying. I I it's really interesting that it can actually beatbox is the other side of it. The fun part though is that like if you listen to sort of the beatboxing, the rhythm isn't always consistent because you know, they're still just like LLMs at the at the bottom. And so, it's just like auto-regressively generating all these tokens and it doesn't quite have a very precise sense of timing.