A new research paper explores the effectiveness of AI models in automating legal citation formatting, a task known as legal drudgery. The study found that current frontier language models only achieve 42.6% accuracy in zero-shot settings for Bluebook citation compliance and fall below human candidates in law review competitions. While retrieval-augmented generation offers only modest improvements, a neuro-symbolic approach combining natural language parsing with a deterministic rule-execution engine significantly boosts accuracy to 85.5%, suggesting a promising direction for legal AI development. AI
IMPACT Suggests that combining LLMs with symbolic rule engines is key to automating complex, rule-based legal tasks.
RANK_REASON Academic paper detailing a new benchmark and approach for AI in legal citation formatting. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
