Two similar AI judges fail together 7.7x more often than independence predicts(github.com)
2 pointsby Lawless19877 hours ago1 comment