A statutory-survey evaluation shows how model errors, retrieval failures and incomplete human answer keys interact.
Legal AI research
Papers, evidence and connections across jurisdictions.
A direct human–model comparison separates resistance to prompt changes from agreement with ordinary speakers in insurance interpretation.
A preregistered, manually labelled evaluation of proprietary legal RAG products.
Separates externally checkable answers from contradiction-based hallucination estimates.
Re-examines GPT-4's bar-exam headline through comparison populations, score construction and evaluation transparency.
A legal training corpus and study of privacy and toxicity filtering grounded in legal context.