Did I just fix the data overfitting problem in LLMs through thoughtful prompting? LLMs can easily be tripped up by simple twists on common puzzles, because they like to rely on common answers instead of reason. My paper, Mind over Data: Elevating LLMs from Memorization to Cognition I propose a fix.