> I don't know how to solve this. Don't use an algorithm which produces its response according to a probabilistic arrangement of tokens when solutions require

> I don't know how to solve this. Don't use an algorithm which produces its response according to a probabilistic arrangement of tokens when solutions require accuracy / correctness? Most probable and most accurate are not the same thing. Hoping that we can get the errors down to something acceptable using an algorithm that is fundamentally inappropriate to solving the problem seems like a fool's errand to me.