You've linked to HIV in the US. Here's the global stats: https://www.unaids.org/sites/default/files/media_asset/UNAIDS_FactSheet_en.pdf https://www.unaids.org/s

You've linked to HIV in the US. Here's the global stats: https://www.unaids.org/sites/default/files/media_asset/UNAIDS_FactSheet_en.pdf https://www.unaids.org/sites/default/files/media_asset/UNAID... Turns out context matters - otherwise the general statement is biased on the specific country's situation and seems to put more weight on the sexuality than necessary. (I.e. the difference is more about frequency/partners/protection than about being gay, they're just correlated in the US) > the de-biased model was less accurate when given unambiguous prompts. Correct. And that's not what I wrote about. These are not questions about population, but specific cases and yes, we should try to maximise accuracy while we minimise bias.