Third-party cyber evaluations involving OpenAI models
openai.com·14h ago
AI Research
arXiv CS.AI·May 28·5 min read
Verified Misguidance: Measuring Structural Citation Failures in Search-Augmented LLMs
TL;DR
Search-augmented large language models (LLMs) often misattribute citations, leading to misinformation. This study developed a framework to measure structural citation failures in these models.
✦ Why It Matters
Engineers can enhance LLM reliability by implementing improved citation validation techniques based on these findings.
Key Takeaways
- Implement CITETRACE to assess citation accuracy in your models.
- Monitor citation quality to prevent misleading information dissemination.
- Consider provider differences when evaluating model performance.
- Use the three-dimensional framework for comprehensive citation evaluation.
How It Works
CITETRACE traces the citation chain from user queries to generated answers, allowing for a detailed analysis of citation quality. The three-dimensional evaluation framework scores citations based on their alignment with user intent, the appropriateness of the source, and the fidelity of the answer to the source.