Summary:
- Researchers at the University of Southern California (USC) have developed a synthetic dataset called SUM (Synthetic Unanswerable Math) to help reduce hallucination in large language models (LLMs).
- Hallucination is a common issue in LLMs, where the models generate plausible-sounding but factually incorrect information. SUM is designed to help train LLMs to better distinguish between answerable and unanswerable math questions.
- The SUM dataset consists of synthetic math problems that are intentionally designed to be unanswerable, allowing LLMs to learn to recognize when a question cannot be answered based on the given information. This can help improve the reliability and trustworthiness of LLMs in real-world applications.