Buyun-Liang/SECA

This project helps evaluate the reliability of Large Language Models (LLMs) by identifying their tendencies to produce incorrect or fabricated information, known as hallucinations. It takes an original question or prompt and generates a slightly reworded, but semantically identical, version designed to provoke a hallucination from the LLM. The output is an altered prompt and the corresponding LLM response, highlighting any erroneous content. This is useful for researchers, AI safety engineers, or product managers who need to assess and improve the trustworthiness of LLM applications.