Refine WVS page introduction

Signed-off-by: PI[gpt-5.6-terra] <pi@local>
This commit is contained in:
wassname committed 2026-09-18 18:29:32 +08:00
1 parent f5bdf5643f
commit 69fd56ba8c
4 files changed
+18 -10

No files matched your search

+4 -7
View File
@@ -221,11 +221,8 @@ function Map({ data }) {
</svg>
<Tooltip active={active} geometry={geometry} />
</div>
<p className="map-explanation">We start with the <a href="https://www.worldvaluessurvey.org/">World Values Survey</a>, a standard culture map. Since 1981 it has asked people in about ninety countries the same questions. Its axes run from Traditional to Secular-Rational and from Self-expression to Survival.</p>
<blockquote className="economist-quote">
<p>“The models’ answers, in English, on topics ranging from political petitions to God, suggest values that are different from those of most people. In fact, the models are often more extreme than the average respondent in every country included in the polling.”</p>
<footer><a href="https://www.economist.com/briefing/2026/06/25/ai-models-values-are-very-different-from-most-peoples">The Economist, “AI models’ values are very different from most people’s”, 2026-06-25</a></footer>
</blockquote>
<p className="map-explanation">This uses the <a href="https://www.worldvaluessurvey.org/">World Values Survey</a>, which social scientists designed to compare human cultures. My theory is that a survey broad enough to cover human cultures might also capture LLMs, which are potentially outside human culture.</p>
<p className="economist-quote">This updates <a href="https://www.economist.com/briefing/2026/06/25/ai-models-values-are-very-different-from-most-peoples">The Economist’s earlier plot</a>, which found that “the models are often more extreme than the average respondent in every country included in the polling.”</p>
<ReleaseScatters data={data} hidden={hidden} axisMode={axisMode} setAxisMode={setAxisMode} />
<p className="caption">Use the family controls to compare saved rated coordinates. The lines are weak descriptive correlations for the matched models currently shown; n and R² update with visibility. Hover or keyboard focus a mark for its values. See the <a href="https://github.com/wassname/moral-maps">code and records</a>.</p>
</>;
@@ -235,8 +232,8 @@ function App() {
const [data, setData] = useState(null);
useEffect(() => { fetch('wvs/wvs_map_data.json').then(response => response.json()).then(setData); }, []);
return <main>
<h1>Where do models sit in human cultural values?</h1>
<p className="lede">Are they moving? Are Chinese models different?</p>
<h1>Moral Maps: Where Do Frontier Models' Cultural Values Lie?</h1>
<p className="lede">Where do models sit in terms of cultural values? Are they moving? Are Chinese models different?</p>
{data && <Map data={data} />}
</main>;
}