“…20 , 21 However, in studies using newer gAI models, human evaluators are frequently no better than random chance. 22 – 26 Interestingly, judgment accuracy can be near chance even when participants’ self-report high degrees of certainty about the source, a finding that replicates in work with AI-generated faces, 22 , 23 videos, 27 artworks, 28 , 29 poetry, 30 , 31 and texts. 25 , 26 While several factors, including the overall length of a text, 21 can influence evaluation accuracy and dictate how credible or trustworthy human evaluators find the AI-outputs, 32 – 34 it’s clear that modern gAIs are very often able to fool human evaluators.…”