Mathematician Says AI Lacks Proof Verification

a16z shared quotes from University of Toronto mathematician Daniel Litt stating that AI models cannot yet produce long complicated proofs because they lack the ability to check correctness. Litt noted that models handle short proofs but face bottlenecks on verification. The posts also describe an interview with a16z's Lisha Li where Litt said current models excel at computations and drawing from many papers but cover a narrower slice of math than headlines suggest.