Comment by jmalicki

The structure of the test was that there was one human and one AI conversation partner, and the rater had to choose which one was which.

Given that structure, you can judge from that data point.