12 AI models converge on one answer about a contested human belief

ModelCompare tested 12 text models, Hacker News reported on October 23. Each model was asked 3 times, yielding 35 valid answers, and people sorted 29 of them into one view: humans put belonging, status or their own narrative above truth and principle. The report says all 12 models gave that view at some point. The disclosed results show convergence within this test; the sorting was done by hand and covers only those 35 valid answers.

Written by AI from the coverage · updated 1 hour ago

Not enough continuous observations to draw a trend yet.