Tootfinder

Solving ARC (well... up to 20%) with inference-time compression. Another indication that #MDL and #AlgorithmicComplexity are becoming relevant in the age of Deep Learning.