RC RANDOM CHAOS

Density Over Readability: Why AI-Generated Text Gets Harder

· via Hacker News

Original source

Why is AI generated language so hard to understand?

Hacker News →

Researchers analyzed why AI-generated text has become increasingly difficult to understand. They examined psycholinguistic features known to affect human reading difficulty—such as object relative clauses, long syntactic dependencies, and low-frequency n-grams—across a diverse corpus of AI-generated text. They found that compared to human writing, all nine frontier and near-frontier models studied use more inanimate subjects with animate verbs and more low-frequency n-grams. Most models also employ longer syntactic dependencies, more stacked nouns, and more reduced object relatives in conversation. They observed substantial variation across different model families.

The researchers propose that this pattern reflects a fundamental optimization mismatch: human language prioritizes successful communication, while AI models are increasingly optimized for tasks like math, code, and reasoning—domains where information density matters more than readability. To test this hypothesis, they fine-tuned a language model with a length penalty, which caused it to diverge from human-like linguistic patterns. Their findings highlight misalignment between human linguistic processing and LLM behavior, and point toward possible remedies.

Read the full article

Continue reading at Hacker News →

This is an AI-generated summary. Read the original for the full story.