Researchers have developed NitiBench+, a new benchmark for evaluating Thai legal question-answering systems. This benchmark addresses the critical issue of temporal relevance in legal data, as laws change over time and older benchmarks often grade answers against current statutes rather than those in effect at the time of a court decision. NitiBench+ includes 1,670 items across 17 domains, providing both historical and current legal texts, and evaluates models on their ability to use the correct legal context. Initial testing revealed that using current law instead of historical law significantly degrades model performance, highlighting the need for temporal awareness in legal AI. AI
IMPACT Highlights the need for temporal awareness in legal AI models, potentially influencing future benchmark design and model development.
RANK_REASON The item is a research paper introducing a new benchmark for AI evaluation. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →