SCOPE compiles natural-language agent policies into predicate programs
Programs trained with a weak executor transfer to other models and match frontier accuracy at lower cost with more stable verdicts.
Stories per day · 30 days
3 stories, newest first · one card per story
Stories tagged #neurosymbolic
Structured Four-Stage Legal Translation: From Natural-Language Traffic Rules to PROLOG
A structured four-stage prompt framework translates raw traffic rules into executable Prolog, reaching 75% accuracy versus 60% and 55% for baselines.
Beyond Solver Verdicts: Generative Reward Models for Autoformalization
Researchers introduce Generative Verification (GenV), a generative reward model achieving 0.961 AUROC in detecting unfaithful autoformalization that preserves solver verdicts.