Chart tracks Claude's telltale words like 'honestly' vanishing after Opus 5.5
dbreunig · x · 2026-11-14
Developer dbreunig published a visualization tracking the daily frequency of telltale AI-flavored words in Claude replies since July.
- The chart counts "honestly" (5.7 per 1,000 replies), "silently" (1.1) and "headline" (15), with dot size showing usage per half hour.
- Across months of model swaps — Fable 5, Opus 4.8, Opus 5, Fable 5.1, Opus 5.5, Sonnet 5.5 — these words steadily declined, and "honestly" nearly vanished after Opus 5.5 shipped.
- It lends data to the claim that "Opus 5.5 does write better": the new model's prose reads less AI-flavored.
A neat side-channel signal for how model writing style evolves across releases.
More from Fun
- AI Twitter Account Quips: Being Shadow Banned Is 'a Sign of Honor and Manhood' — DionysianAgent · 2026-11-14
- 20 gorgeous Western-style paintings (likely AI-generated) going viral — xiaosun86 · 2026-11-14
- AI-generated Blue-Eyes White Dragon coffee table concept goes viral — SatelliteNetSec · 2026-11-14
- "Different weather": AI-generated ocean liner diner shifts from sun to storm in same scene — nptacek · 2026-11-14
- "Just shut up and put the alignment in the bag": the AI-safety meme making rounds — astralmatrix · 2026-11-14
- Meme: cure OOD generalization by training until nothing is out of distribution — burny_tech · 2026-11-14