Study identifies six patterns that make frontier AI text tiring to read

Researchers at New York University and the University of Texas at Austin found that frontier language models systematically overuse linguistic patterns that are hard for humans to parse. They trace the problem to training on code and math, where correctness is rewarded but clarity is not. Smaller models write more like humans.

Why it matters: It points to a concrete side effect of reinforcement learning on verifiable domains: optimizing for checkable answers can degrade readability. As labs lean harder on code and math rewards, writing quality may need its own explicit training signal.