The results mark a departure from earlier research on older models like GPT-4, which showed mixed or negative effects on top-performing students. The specific performance metrics for each task and detailed breakdowns of where the tools succeeded or fell short remain unpublished beyond the summary findings.
Practitioners should note the timing and scope of this research. The study tests newer AI architectures—retrieval-augmented generation for source accuracy and reasoning models for structured analysis—that address hallucination and reasoning deficits documented in prior work. With 55 percent of law schools now offering AI courses and firms deploying tools like Westlaw Edge and Lexis+ AI, this evidence of safer, more capable AI integration carries weight for legal education and junior associate training. Firms considering AI adoption for associate work should track whether these findings hold across different practice areas and experience levels.