Lance Fortnow

ML models that accurately mimic cryptographically secure pseudorandom generators are doomed to output near-uniform distributions, revealing a fundamental limit in their ability to capture certain types of complexity.