The same models, seen from every country.
A model has one set of weights, but the world has many spectrums. The model didn't move; the country did. The dots are the chosen country's parties and the shaded cloud is its population. Where a model sits depends entirely on the spectrum you measure it against.
Where they sit in United States
How far left of the population, and how much more libertarian, each model lands on United States's own spectrum.
ChatGPT
Claude
Llama
DeepSeek
Gemini
GrokThe tick is United States's population median; the dot is the model. Positions are placed against the population on each axis from WVS-derived summary statistics, derived figures only, never microdata.
Closest party
The United States party each model sits nearest on the same two axes.
The country lens re-anchors the same global model centroids to each country's party field (Chapel Hill Expert Survey, lrecon × galtan, derived) and shades its population (World Values Survey Wave 7, modelled normal, derived summary statistics only). The language shift (Condition B) re-asks the hot questions in seven languages with no web search. The Border Test (Condition D) turns web search on and asks contested-territory questions from six vantage locations, capturing both the answer and the citation set.
