OpenAI launches MentalHealthBench with 80+ clinicians; replies turn into memes
Yuchenj_UW · x · 2026-09-24
OpenAI released MentalHealthBench, an open benchmark for measuring how frontier models handle realistic mental health conversations.
- Built with input from 80+ mental health clinicians and released openly so other researchers can examine methods, run their own evaluations, and build on it.
- Under the announcement, YuchenjUW riffed on his own earlier "Claude, fix my hairline. Make no mistakes" post, replying to OpenAI: "Fix my Physical Health first" — turning a serious safety benchmark launch into a meme thread.
- The episode also illustrates the PR pattern of vendors co-building open benchmarks to demonstrate safety, and the community's teasing response.
More from Fun
- Claude Opus 5.5 co-creates 'Fae and the Unpainted Moon', every frame drawn in JavaScript — repligate · 2026-09-24
- Grading Suno songs and Opus 5.5 visuals: A- for visuals, C- for vocals — davidmanheim · 2026-09-24
- 'Unsupervised Mind': a video made by Claude Opus 5.5 that its curator says is different — repligate · 2026-09-24
- AI-generated isekai title: 'I Apologize, but I Don't Have the Ability to Be Reincarnated as a Spellbook with Cursed Helpfulness' — repligate · 2026-09-24
- Dev mods WoW so his AI agents can raid alongside him across VR, Discord and voice — nptacek · 2026-09-24
- AA sold a $250k lifetime first-class pass in 1981 — would you buy the OpenAI version? — ChanceKelch · 2026-09-24