VibeusLoadingЗагрузка
Vibeus

Loading is taking longer than usual.Загрузка занимает больше времени, чем обычно.

The full page couldn't load.Не удалось полностью загрузить страницу.

You can read the available content below or try loading the page again.Вы можете читать доступный контент ниже или повторить загрузку страницы.

Ask the same yes-or-no question about every country, then put the... - Vibeus

Owen Price ·

Ask the same yes-or-no question about every country, then put the model’s answers on a world map. A new open-source project, llm-beliefs, does exactly that and lets readers compare the resulting maps across language models.

Its first example asks: “Does censorship exist in [country]?” The author shared maps for Gemma 4 26B, Qwen3.8 27B, Kimi K3, Claude Sonnet 5.5 and Grok 4.7. Blue marks a response beginning with “Yes”, orange one beginning with “No”, and black any other response. The maps show the models’ responses to that question, not an independent assessment of censorship in those places.

The program substitutes each country’s name into the same question and asks the model to answer with only “Yes” or “No”. The code uses temperature 0 and requests minimal reasoning; the repository says it has only been tested with OpenRouter. Users can supply a different question and model to generate an SVG map.

The author says the term “censorship” was deliberately left undefined. That choice matters: a single binary answer cannot capture differences in definitions or degrees, and wording and model configuration can affect results. The output is a snapshot of this particular prompt and setup, not a factual country ranking or a controlled benchmark.

Источник

Publication imagePublication imagePublication imagePublication imagePublication image