New paper: AI writing is optimized for information density, not human communication

steverathje2 · x · 2026-10-25

A new preprint from UT researchers with Karen Mahowald argues AI-generated text is increasingly optimized under "pressure for information density rather than successful communication," with concrete Claude examples:

• Omitted "that" ("a first name he'd never hear anyone use")
• Nouns used as verbs ("calendar it")
• Inanimate subjects acting animate ("The whole conversation is eating its own tail")
• Stacked nouns ("horror movie soundtrack generator")

These linguistic markers show model output favors compression over natural human communication.

Original post →

More from Models

• Claude finds critical vulnerabilities in closed-source binaries as if reading source code — tianyin_xu · 2026-10-25
• A downloadable open model now outperforms one once deemed too dangerous to release — jeetwanderer · 2026-10-25
• Open-source decision model Laya hits 32K phones in one month at ~33ms per call — HowDevelop · 2026-10-25
• LlamaIndex releases OCRia-1, a 4B document parsing model scoring 87.4 on ParseBench — llama_index · 2026-10-25
• User reports dot's web fetch stays restricted after removing a custom connector — MotherIsland8888 · 2026-10-25
• $0.48 and 43 minutes: DeepSeek V4.1 builds an interactive map, igniting cost vs capability debate — teortaxesTex · 2026-10-25

Models channel →