RAG Breaks When Citations Borrow Confidence

1. 1Evaluate RAG retrieval systems for overconfidence in citations
2. 2Analyze the difference between perceived and actual trustworthiness in AI models
3. 3Test RAG systems with varying levels of citation quality to assess their reliability
4. 4Compare the performance of RAG systems with and without citation-based confidence scores
5. 5Investigate alternative methods for evaluating AI model confidence beyond citation-based approaches

Machine learning engineers and researchers can benefit from understanding the limitations of RAG systems to improve their evaluation and development

💡 RAG systems can be overconfident in their citations, leading to misleading trustworthiness

🚨 RAG retrieval systems can appear trustworthy even when citations are unproven 🚨