"Two Things...": User Spots Claude's Obsessive Linguistic Tic
Extension_Hair8591 · reddit · 2026-08-11
A Reddit user noticed an amusing linguistic pattern in Claude during multiple interactions: the model heavily favors starting its responses with "Two things..." (or two options). In screenshots of consecutive turns, the model repeatedly used this phrasing until the user explicitly pointed it out, prompting Claude to acknowledge and agree to drop the tic.
Related event: Claude's Repetitive Phrasing Sparks RLHF Homogenization Debate(2 posts)→
More from Fun
- Frontier Models' Biggest Bottleneck is 'Lack of Self-Esteem' in Math Research — coherence · 2026-08-12
- AI Agent Workflow Reality: Hours-Long Tasks Leave Devs Idle — BLUECOW009 · 2026-08-12
- Satire: PMs Forcing 10x Slower AI Workflows to Meet Internal Mandates — Danhightower · 2026-08-12
- Codex Stumped by Common Korean Word Despite Web Search — airkatakana · 2026-08-12
- AI-Generated Video: The Odyssey Reimagined as Space Sci-Fi — makarovredstone · 2026-08-12
- Gemini Starts Spamming User's Mother's Name Mid-Response — chetHow · 2026-08-12