Discussion on Model Writing "Slop"

Afinetheorem · x · 2026-07-10

A reshared discussion argues that "slop" in model outputs isn't exclusive to the LLM era; empty phrases like "thread 🧵" were already common in earlier LinkedIn copy.

The core point is that some models are "greasier" than others, frequently using vague clichés, while some write much more cleanly. The issue is considered solvable through training and control methods.

Original post →

More from Models

Models channel →