Jev, Prolog, Pi, and the dream of probabilistic logic programming
4 hours ago
- . Jev is now supported by the latest versions of DeepClause and its Pi extension, offering a natural mapping of its concepts to logical predicates in DML/Prolog.
- . This integration provides greater speed and determinism for use cases where pure LLM/Agent approaches are too expensive or indeterministic, benefiting SOP2AGENT-style applications.
- . The system uses a `judge` predicate that calls Jev (or an LLM fallback) to answer questions like multiple choice (`choose`), scaling (`rate`), yes/no (`verify`), and calibrated probabilities (`probability`).
- . The approach revives the notion of probabilistic logic programming, which previously existed in academic works like the 1990 paper on meta-interpreters for rule-based inference under uncertainty and DeepProbLog, but often lacked scalable probability sources.
- . Jev, trained on large datasets, now provides reasonable probabilities at scale, potentially making probabilistic logic programming practical for real-world applications.