[name] asked: Why does the model keep choosing the next token even if the probabilities of options are low?