A preprint tests an LLM for detecting overlap among test questions, finding stronger psychometric alignment but a trade-off in simulated adaptive tests.