LLM Outputs Are Getting Harder to Read, Raising Human Cognitive Load

dejavucoder · x · 2026-08-10

Developers point out that collaborating with LLMs demands enormous cognitive effort, as models throw huge walls of text requiring rapid human judgment. Furthermore, the generated "neuralese" is becoming increasingly unreadable, with outputs from models like GPT-5.6 and Opus 5 being notably tough or annoying to parse.

Original post →

More from AGI Musings

AGI Musings channel →