Chain-based Adaptive Reconfiguration Over Lattices for Hallucination Reduction
For LLM practitioners, CAROL offers a principled method to reduce hallucinations at test time with provable guarantees, though it is an incremental improvement over existing uncertainty-based and retrieval-augmented approaches.
CAROL introduces a probabilistic framework for test-time hallucination reduction in LLMs, using semantic uncertainty based on consistency with trusted context. It reduces hallucinations and improves reliability on QA and multi-agent benchmarks compared to baselines.
We introduce CAROL (Chain-based Adaptive Reconfiguration Over Lattices), a probabilistic framework for test-time hallucination reduction in large language models. Rather than relying on token-level uncertainty, CAROL defines a semantic uncertainty measure based on the consistency between generated responses and a trusted context, inducing a string-submodular objective over a lattice of textual sequences. This formulation enables hallucination mitigation to be cast as a Markov chain accept-reject process with provable convergence and near-optimality guarantees, allowing the model to iteratively refine outputs toward semantic consistency. By operating at the level of meaning, CAROL unifies hallucination detection and mitigation within a single framework. Empirical results on question answering and multi-agent reasoning benchmarks show that CAROL significantly reduces hallucinations and improves reliability and interpretability compared to likelihood-based and retrieval-augmented baselines, while maintaining competitive computational efficiency.