Teaching Language Models to Verbalize Uncertainty

The episode’s core move is visual: uncertainty should attach to a claim, not to the next token alone. This page turns that distinction into diagrams, calibration plots, semantic clusters, and shift stress-tests.