TokenPost·
Legal AI Test Records 70% Correct Value Extraction After Revisions
In driving-under-the-influence cases, the share of correctly extracted values rose from 52%. Repeated model runs agreed more often with each other than with the reference answers.