Taki AI

Taki AI is the model described in A hawk who flew on political winds: Lindsey Graham as having been trained only on text from before 1931. The episode uses it as a vivid case of training-data worldview: when asked about national pride or empire, it answers from a strongly British imperial perspective.

Taki matters less as a product page than as evidence for Language-Dependent AI Bias and AI Model Value Surveying. If a model’s text world is historically bounded, its answers can preserve values that later readers might treat as obsolete, harmful, or revealing of the corpus rather than of any deliberate policy choice.

Connections

• AI Model Value Surveying - survey method that makes Taki’s worldview visible.
• Language-Dependent AI Bias - broader training-data and language-dependence concept.
• AI Model Censorship - contrast case where model values are shaped by post-training refusal behavior.
• Sondre Solstad - source participant explaining why these tests matter.