Living Worlds

Research · Lifetimes · Oct 6, 2026

Lived, not learned: children ever had

Met1,524 people, 2024 held out

The test at a glance

What it tests
Accuracy against reality: how many children people have had.
Data
General Social Survey 2024, US-born respondents in six cells (women and men 25-59); the engine lives the US from 1950 on UN WPP 2024 rates, with weights fitted only to Census counts of children among women 40-44.
Held out
Every 2024 answer.
Compared against
Gemini 3.8 Flash answering from what it has read.
Bar
Lower mean error over the six cells than the text model.
Result
0.17 against 0.36: met. Siblings missed (1.10 against 0.38).

A population lived month by month in the United States from 1950, on each year's official fertility and mortality, matched how many children US-born women and men aged 25 to 59 had had within 0.17 on average; a frontier model answering from what it has read missed by 0.36. No General Social Survey answer was used to build it; its only fitted weights come from Census counts of children among women 40 to 44. Registered before any 2024 answer was read; scored on six cells of the 2024 General Social Survey. Its count of siblings missed (1.10 against 0.38) and became the next test.

Protocol, result and scorer

Lifetimes

More in this area.