llm-beliefs
An open-source experimental project that asks language models closed-ended questions about each country and visualizes the answers, revealing systematic biases the models hold toward different countries.
Timeline (last 90 days)
A Reddit user developed an open-source program called llm-beliefs that iterates through countries around the world, poses closed-ended questions to language models about whether censorship exists in a given country, and visualizes the answers as an SVG world map.
The author of the llm-beliefs project said he was "quite surprised" by the results.
The article's author did not disclose which model llm-beliefs specifically used.
llm-beliefs did not publish the specific prompts, model version, or parameter settings, resulting in limited interpretability of the results.
All Facts (4)
A Reddit user developed an open-source program called llm-beliefs that iterates through countries around the world, poses closed-ended questions to language models about whether censorship exists in a given country, and visualizes the answers as an SVG world map.
50%UnverifiedThe author of the llm-beliefs project said he was "quite surprised" by the results.
50%UnverifiedThe article's author did not disclose which model llm-beliefs specifically used.
50%Unverifiedllm-beliefs did not publish the specific prompts, model version, or parameter settings, resulting in limited interpretability of the results.
50%