cartgr comments on Are language models good at making predictions?

[cartgr]

Why didn’t you look at the token probabilities for the “yes” and “no” tokens?