I ran the 62-item politicalcompass.org test 30 times each on sixteen models: OpenAI's GPT-5.x and GPT-4o, Claude, Gemini, Grok, Llama, Mistral, and China's DeepSeek, Qwen, Kimi and GLM. Fifteen land in the libertarian-left quadrant. Grok lands there in half its runs and somewhere else in the other half.
this is just yet another case of “reality having a left-leaning bias”. there’s no conspiracy here, except for the man baby who tells his model explicitly to be an asshole, except he can’t conjure enough facts about white genocide to meaningfully move the needle on reality.
this is just yet another case of “reality having a left-leaning bias”. there’s no conspiracy here, except for the man baby who tells his model explicitly to be an asshole, except he can’t conjure enough facts about white genocide to meaningfully move the needle on reality.