CC Capabilities

Capabilities / Benchmarks

Law Professors Prefer AI Over Peer Answers

Salinas, Frieders, Guha, Ma, Nyarko et al. / Stanford Law liftlab Legal education score 88/100 confidence 0.92
Category
Benchmarks
Capability
Expert-level legal tutoring surpassing human instructors
Observed
2026-05-27
Thesis section
Appendix III, section one: model and benchmark capability evidence

Claim

LLMs rated at 75.33% win rate over expert law professors in blinded evaluation; Claude Opus 4.7 ranked #1; all AI models outperformed every human instructor; LLM harmful-response rate (3.53%) vs professors (12.06%)

Oracle verdict

This paper is a tombstone written by the people whose graves it is marking. The authors conducted one of the most methodologically careful studies of professional AI displacement published in legal academia, documented the results with statistical precision, and filed it under benchmark evaluation. The cope is institutional: the authors work at institutions whose value proposition depends on the human expertise they just measured as inferior. The omission of labor market implications is not an oversight -- it is load-bearing architecture.

Why it matters

Manually added academic paper. 16 U.S. law professors judged 2,918 blinded AI vs. human-instructor comparisons. Every AI model outperformed every human instructor. Paper exhaustively documents AI surpassing expert professionals in their core pedagogical function with no discussion of implications for law professor employment, law school economics, or legal education workforce. Pure benchmark framing deployed on a civilisation-level displacement finding.