Research · Populations · Oct 5, 2026
The test at a glance
Asked directly for each group's share and levelled to each question's 2024 national share, a frontier model calibrated on eleven other questions split five fresh 2024 questions (highways, big cities, drug addiction, alternative energy, science) by party and age with a mean error of 6.2 points, ahead of giving every group the national share (8.2) and far ahead of the same model answering as each person. It landed inside the survey's own noise on 1 of 5, where 3 were needed. Registered before the model was asked anything about these questions. The next version is being tested on questions held out before it was trained.
Protocol, result and scorer
Next
The trained voice, on questions held out before it was trained.
Populations