Grigore Rosu: Program execution is proof generation, unlocking infinite training data from code

LingmingZhang · x · 2026-09-19

Formal methods researcher Grigore Rosu argues that "program execution IS proof generation": within semantic frameworks, execution is rooted in formal language semantics, so every step has a justification. Composing those steps turns a finite execution into a proof of how the program moves from initial to final state.

The payoff: across unbounded programs and inputs, code supplies infinitely many finite execution proofs, and the resulting data — program states, applied rules, and derivations — can be used directly as training data generated from code.

Original post →

More from Research

Research channel →