What if AI intelligence could be measured in bytes instead of test scores? In this episode of This Is AGI, Alex Chadyuk introduces a principled, information-theoretic approach to evaluating artificial intelligence, moving beyond benchmarks like ARC or FrontierMath toward entropy reduction as a true measure of generalization. Learn how Shannon information, uncertainty reduction, and model size combine into a scientific framework for “intelligence density,” and why this matters for the future of AGI and frontier AI model evaluation.