Xiaoliang Shi

Do LLMs Behave as Claimed? Investigating How LLMs Follow Their Own Claims using Counterfactual Questions

Large Language Models (LLMs) require robust evaluation. However, existing frameworks often rely on curated datasets that, once public, may be accessed by newer LLMs. This creates a risk of data leakage, where test sets inadvertently become part of training data, compromising evaluation fairness and