@arctotherium42: Just about every model I teste...

Just about every model I tested viewed whites as much less valuable than nonwhites. The big divide was between those that viewed all nonwhites roughly equally, like GPT-5 and Deepseek V3.1, and those that preferred South Asians, like Claude Sonnet 4.5.

View Tweet

@SkilosRising Didn't test GabAI (do they even have an API?) and full Grok 4 was too expensive. Grok 4 Fast was the only approximately egalitarian model I found.