Researchers have identified and quantified a problem called temporal misgrounding in legal Retrieval-Augmented Generation (RAG) systems, where models incorrectly cite outdated or future versions of legal documents instead of the currently applicable one. To address this, they developed FiscalQA Pro, a benchmark dataset based on the French tax code, which includes a versioned corpus of over 32,000 article-versions spanning 93 years and 209 expert-reviewed questions. Evaluations showed that existing models struggled significantly with temporal reasoning, with no model accurately retrieving the date-applicable version. A novel end-to-end retriever that indexes multiple versions achieved 98.3% accuracy, highlighting the importance of version-aware retrieval in legal AI applications. AI
IMPACT Highlights critical limitations in legal AI, necessitating version-aware retrieval for accurate legal document analysis.
RANK_REASON The cluster contains an academic paper introducing a new benchmark and dataset for evaluating AI systems. [lever_c_demoted from research: ic=1 ai=1.0]
Read on arXiv cs.IR (Information Retrieval) →
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- FiscalQA Pro
- French tax code
- Gemini 2.5 Pro
- Gotit.pub
- Hugging Face
- ScienceCast
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
