ROUGE scores demonstrate systematic length bias, where responses exceeding 100 tokens consistently receive scores below the 0.3 threshold, regardless of factual accuracy.

Authors

• Person: Not available Organization: arXiv
  Re-evaluating Hallucination Detection in LLMs - arXiv

Sources

• Re-evaluating Hallucination Detection in LLMs - arXiv arxiv.org arXiv via serper

Referenced by nodes (1)

• ROUGE concept