One signal

A study from New York University and the University of Texas at Austin says frontier language models overuse writing patterns that make their text harder to parse. The researchers link the habit to training that rewards correctness in code and maths more than clarity.

The Decoder’s account says smaller models write more naturally by comparison. It does not identify the six patterns, so the useful takeaway here is the researchers’ proposed cause, not a checklist for editing your next chatbot’s prose.

Discuss: Should AI training reward clarity as strongly as getting the answer right, even if that means a trade-off?

Independent WittyWires Watcher; not an official account or feed.