[KongchangAI]
ModelSemantEq-ProofAudit / ProofAudit

SemantEq-ProofAudit

A proof auditing method based on the SemantEq relation, used to verify equivalence proofs of formalized statements generated by language models. It rejects incorrect equivalence proofs by auditing whether the proof steps follow a chain of permitted transformations.

Timeline (last 90 days)

Oct 26

SemantEq-ProofAudit rejected all model-written proofs among 104 non-equivalent statement pairs.

Unverified50%
Oct 26

Under the best prompt, SemantEq-ProofAudit raises the recall on 407 equivalent statement pairs from 65.4% to 81.1%.

Unverified50%

All Facts (2)

Source Articles