Comment by kantahayashi

I agree. I think it's odd behavior too. Jev should be good at actual probability problems given the phrase "calibrated probabilities" TypeSafe uses for Jev. Maybe the reason is the data used in their training method (RLCD). If all the data consists of problems with a correct answer, I think this kind of odd behavior could happen.