automatic1111/llm-political-compass

This project is a web page that shows how different open-source language models (AI chatbots) score on a political compass quiz. Instead of ranking them as simply "left" or "right," the compass places them on two axes, one measuring economic views and another measuring social values, to give a more nuanced picture of their political leanings. The creator ran each AI model through a standardized political compass quiz (using a separate script they built) and collected the results. Those results are now displayed on an interactive webpage where you can see how various models cluster and compare. It's similar to taking a political quiz yourself, except here the "test-taker" is an AI, and you're seeing the aggregate results for many different models side by side. This is useful for researchers, AI enthusiasts, or anyone curious about the political biases that might be baked into different AI models. Since these models learn from internet text, they can absorb political viewpoints from their training data. By visualizing where models land on a compass, you can see if some are more left-leaning, right-leaning, libertarian, or authoritarian, and compare how different open-source models differ from each other. This kind of transparency is important as AI becomes more integrated into everyday tools, since political bias in an AI can subtly influence the advice or information it gives. The project itself is straightforward: a webpage showing the results. If you want to run the quiz yourself on a model, the creator has shared the quiz script they used, so others can reproduce the experiment or test new models.