AI vs Puzzles is an experiment in cryptic and logical intelligence. A space where large language models are tested not just for what they can say, but for how they think. It's about seeing whether reasoning, pattern recognition, and intuition can emerge when language alone isn't enough.
We explore how models handle various aspects of cognition, including the heavy logic of Sudoku, the numerical precision of Countdown, and the visual understanding of games like Geoguessr or Domino. Each puzzle pushes a different part of cognition, even the ability to form connections between words in unexpected ways, such as recognizing relationships based on how they sound instead of their literal meaning through Connections.
We store all attempts and generate leaderboards for all puzzles. It's not about competition as much as discovery, finding out what reasoning means when the thinker isn't human.