AI's "OH MY GOD!" exclamations may actually help its reasoning

danintheory · x · 2026-09-30

Linguist Kyle Mahowald argues that exclamations like ARGH and OH MY GOD in AI agents' chains of thought aren't just imitation—these words likely play functional roles in reasoning: what comes after "oops" differs sharply from what comes after "aha". The argument draws on an Atlantic piece he co-wrote with Daniel Drucker.

Original post →

More from Models

Models channel →