A good model requires both effective training methods and good data.

“Um, but then the other axis is to make a better model is just to have better data. More data, uh, what type of data, what kind of reward signals in that data, and obviously, you need both of those, uh, doing well in order to actually have a good model at the end of the day.”