Opus 5.5 shows a recurring 'stamp' phrase pattern, dubbed the new 'delve'
dylfreed · x · 2026-10-01
After spotting it in several videos, dylfreed documents a recurring 'stamp motif' in Opus 5.5 outputs, calling it the model's version of 'delve' — a telltale phrasing habit that could become a marker for detecting AI-generated text.
More from Models
- One-line take: GPT-6.1 Sol is underrated, says AI commentator — mallow610 · 2026-10-01
- "Opus 5.5 is the cheapest model" — human hours saved beat token pricing — CamBrazy3 · 2026-10-01
- CUHK study maps when recurrence helps in looped language models, proposes history-state injection — CUHK-CSE · 2026-10-01
- Local 27B Face-off: Dirk-Qwen3.8 Beats Swift-1.5 on a 200-Question Personal Eval — norenEnmotalen · 2026-10-01
- Influencers hype Gemini 4 Argon: GOATED or hopelessly benchmaxed? — thatroblennon · 2026-10-01
- Reddit Pushback: OpenAI Users Subsidize Failed Experiments Like Atlas and Sora Via Price Hikes — dagerika · 2026-10-01