Interventional Grounding Audits: Black-Box Premise-Dependency Tests for LLM Chain-of-Thought via Predicate Substitution

본문 미리보기

arXiv:2607.13069v1 Announce Type: new Abstract: Large language models produce chain-of-thought (CoT) reasoning that appears logically sound yet may not genuinely depend on its stated premises. We introduce interventional grounding audits, a black-box, step-level test of premise dependency: we intervene on a single premise by substituting its target predicate with a fresh symbol, re-run the model, and check whether each reasoning step's normalized conclusion (canonical predicate form) changes. W

Interventional Grounding Audits: Black-Box Premise-Dependency Tests for LLM Chain-of-Thought via Predicate Substitution

본문 미리보기

관련 글

OriginBlame: Record- and Token-Level Data Provenance for AI Training Datasets

Improving Molecular Property Prediction in Small Language Models Using Graph-based Tools

Multi-Agent Collaborative Reasoning with Tool-Augmented Evidence for Urban Region Profiling

SPINE: Bridging the Cyber-Physical Gap with Agentic AI