zeshen comments on Reward is not the optimization target

[zeshen]

Are you just noting that the model won’t necessarily find the global maxima, and only reach some local maxima?

That was my takeaway as well, but I’m also somewhat confused.