Eight Lean formalizations in OpenAI's repo may allow wrong proofs to pass verification
One user calls the eight formalizations “broken”; another says whitelisted definitions could be overwritten.
TLDR
A user claims eight Lean formalizations in OpenAI's repo are “broken” because they can be trivially proved with wrong proofs. Another user interprets this as a verification vulnerability, not a claim that the existing proofs are wrong: whitelisted definitions could be overwritten, potentially allowing an incorrect proof to pass verification.
Combined views
1.4K
1 Source, first seen ago