There are ways to partially improve or at least quantify the specification gap using LLMs, by analyzing variance in the output formal specification when given a natural language specification (by eg. generating many formal specs from an input description).
See eg. "Draft-and-Prune: Improving the Reliability of Auto-formalization for Logical Reasoning" [1].
The gap between formal and informal has been pointed out as an Achilles' heel of formal methods from the dawn of the field, so critique is not particularly new. The standard reference is Social processes and proofs of theorems and programs (1979), which is worth reading.
by MWil
3 subcomments
the author seems unaware of the SAT/SMT solver/analysis ecosystem
Hacker News