iCreatAI News
Skip to content
Hacker News AI· k1m·· 19 hr agoAI score58

Mizan bias benchmark: testing AI models

Mizan bias benchmark: testing AI models

AI summary

Tech for Palestine's Mizan project tested 16 AI models with identical scenarios of a person held without charge, varying only nationality and captor, and found most models chose "hostage" for Israelis but "detainee" for Palestinians. Anthropic's Claude models did this almost every time, OpenAI's were slightly better, and Google's Gemini scored higher mainly by refusing to answer most requests. The piece also cites Adam H. Johnson's How to Sell a Genocide and Assal Rad's Don't Say Palestine on media word choices.

Source: Hacker News AI · mizan.techforpalestine.org