Observed arrival · 2026-08-30
The Byerley Test
An AI evaluation experiment tests whether politically steered, AI-written statements can pass as authentic words from world leaders.
Field notes
The experiment separates imitation from political steering: models first attempt to reproduce a leader’s voice, then generate versions shifted across five issue-relative positions. Its diagnostics include stance and escalation scales from -2 to +2, matched-attempt memory patches, and an archive filtered by leader, model, channel, result slice, and status. The homepage identifies five leaders and 27 scored cycles, although the extracted results and transcript panels are still shown as loading.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue