Glossary · Term

LawBench

← all terms

Definition

Plain language

A test where an AI reads a legal case and picks which criminal charge applies.

As stated in the literature

A Chinese legal-reasoning benchmark; the featured variant is a classification task over 191 criminal charges, used to test weak-to-strong RL gains.

Why it matters: It tests legal-reasoning accuracy on a realistic, fine-grained classification task with hundreds of possible answers.

For example, it gives a model the facts of a criminal case and asks it to pick the correct charge from 191 possibilities.

Mentioned in 1 episode

  1. 088
    Two Levers for Self-Improving AI: When Rewriting Code Isn't Enough

Related terms