Learning When to Trust LLM Priors: A Validated Framework for Semantic Prior Integration

TL;DR

Statsformer is a validated framework that adaptively learns when to trust LLM-derived semantic priors in supervised learning, improving predictions by calibrating the influence of various prior-informed models.

Contribution

It introduces a method to selectively trust LLM semantic priors using out-of-fold validation, ensuring robustness and performance guarantees in supervised learning.

Findings

Informative LLM priors enhance prediction accuracy.

Unreliable priors are automatically downweighted.

The framework guarantees performance close to the best convex combination of models.

Abstract

Large language models (LLMs) encode rich semantic knowledge that can be useful for supervised learning, but their outputs are unreliable as statistical priors: they may be noisy, misspecified, or hallucinated. Existing LLM-informed learning methods either trust such signals directly, leaving predictions vulnerable to unreliable LLM guidance, or restrict semantic integration to a single model class. We introduce Statsformer, a validated framework for learning when to trust LLM-derived semantic priors in supervised statistical learning. Statsformer maps LLM-derived feature scores into a family of learner-specific prior-injection mechanisms across a heterogeneous library of linear and nonlinear predictors. It then uses out-of-fold validation to adaptively calibrate the influence of each prior-informed learner, allowing useful semantic information to improve prediction while attenuating weak, misspecified, or adversarial priors. This yields a guardrailed statistical learning system with an oracle-style guarantee: up to statistical error, the final predictor performs no worse than the best convex combination of its in-library candidates, including prior-free learners. Across diverse prediction tasks, informative LLM priors improve performance, while unreliable priors are automatically downweighted. These results position Statsformer as a reliability-oriented approach to LLM-informed statistical learning: rather than trusting LLM knowledge directly, it validates semantic priors against data before allowing them to influence the final predictor.