A Hacker News post argues that large language models gravitate toward the rhetorical pattern "It's not X, it's Y," known as negative parallelism. The author notes its prevalence on LinkedIn and in AI‑generated text, and challenges the claim that the construction signals lazy writing. Instead, the piece treats the pattern as a cognitive tool for reframing assumptions.

The essay critiques AI‑detection services like Grammarly, which flag phrases such as "automated language production" as highly likely to be machine‑written. The tool then suggests human‑sounding alternatives, effectively rewriting the author's voice. To prove authenticity, the writer paid Pangram $20 for a verification report, highlighting the financial and reputational stakes of false positives.

Finally, the post explains how reinforcement‑learning techniques (RLHF and RLVR) amplify token sequences that resemble human reasoning, causing models to emit longer chains of contrastive language. This feedback loop explains why the "not‑X‑is‑Y" structure spreads, and it warns that over‑reliance on detectors may erode genuine rhetorical nuance.