A recent study by New York University and the University of Texas at Austin has identified six linguistic patterns that make AI-generated text challenging for humans to read. According to The Decoder, frontier language models tend to overuse these patterns, which complicates comprehension.
The researchers attribute this issue to the AI training process, which heavily involves code and mathematical data. In these domains, correctness is prioritized over clarity, leading to language that is technically accurate but less accessible to human readers. The Decoder also noted that smaller AI models tend to write in a way that is more natural and easier for humans to understand.
For Japanese financial markets, where clear communication is crucial in FX, crypto, and equities trading, understanding these AI limitations is important as market participants increasingly rely on AI tools for analysis and decision-making.