Again, I am asking you for a theory. WHY would strings of text reflecting word distribution in some unknown training corpus be useful for the work that you are doing? >>

→ View original post on X — @emilymbender