Researchers Find Six
The Decoder covers a study that puts hard evidence behind a complaint every reader of AI text has felt: the prose has a recognizable, tiring texture. The causal claim — that reward for correctness in code and math data spills over into general writing — is plausible but not yet airtight; linking training mix to style needs more than one study.… more
The most provocative detail is that smaller models write more human-like, suggesting natural phrasing is less a capability than a casualty of heavy optimization. In our view, readability now becomes a competitive axis: expect post-training and prompting to chase 'human' style as deliberately as they once chased benchmarks.
Researchers find six patterns that make AI text so exhausting to read
A study from New York University and the University of Texas at Austin shows that frontier language models systematically overuse linguistic patterns that are hard for humans to parse. The researchers trace the problem to training on code and math, where correctness gets rewarde…
Source:
The Decoder
Discussion (0)