Odd word choice sparks token watermarking speculation in LLM output

Aizkmusic · x · 2026-08-17

An odd LLM output behavior sparked discussion about token watermarking. The model completed a phrase with "slop" where "slack" was the logical fit, suggesting a deviation from standard probability. The user speculated this could be a watermarking signature and asked for expert analysis.

Original post →

More from Models

Models channel →