Yeah, it should be sentences that have low next token distribution entropy. Where an LLM is sure what the next word is. I bet people do real well on those too. By the way, I also had 5/15.