[hackpert]

If anyone else is curious about which ARC-AGI public eval puzzles o3 got right vs wrong (and its attempts at the ones it did get right), here's a quick visualization: https://arcagi-o3-viz.netlify.app