Develop fine-tuned LLM proposal oracles for GRASP

Determine whether fine-tuning the large language model used as GRASP’s rule-proposal oracle improves the quality and validity of the proposed rules.

Background

GRASP currently treats its LLM as a black-box proposer that generates candidate symbolic rules, which are then evaluated symbolically and incorporated through gradient boosting. The conclusion explicitly leaves open whether fine-tuning this LLM can improve the quality and validity of the generated rules.

References

Finally, grasp currently treats the LLM as a black-box proposer; fine-tuning the LLM could improve the quality and validity of proposed rules.

— Learning Probabilistic Logic Programs with Functional Gradient Guided Language Models  (2610.12303 - Mathur et al., 8 Oct 2026) in Section 5, Conclusion