{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/b549d07622694c8c9b1d9b31f2e38f60\" frameborder=\"0\" width=\"1090\" height=\"817\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":817,"width":1090,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":817,"thumbnail_width":1090,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/b549d07622694c8c9b1d9b31f2e38f60-16cafc40835b9993.gif","duration":139.72,"title":"Can Agents Outthink Humans in Games","description":"This Loom demonstrates a study to compare how an AI agent performs versus humans on newly generated game-like logic puzzles. The team asks another agent to create random game variations, then lets both humans and the AI read the rules before solving. The games are inspired by Lincoln scans but with slight variations designed to test intelligence and the ability to avoid relying on prior knowledge or tricks. The author notes that the current agent or human solves can be slow, especially the rotating aspect, and expresses interest in whether decision models or code-based agents can solve faster in the future."}