221 fresh chats: Claude Sonnet 5.5 says "47" for 92% of random number requests
randal_olson · x · 2026-10-19
Randal Olson ran 221 fresh claude CLI sessions asking Claude models to pick a random number from 1 to 100. Sonnet 5.5 gave only 4 distinct numbers, answering "47" 92% of the time; Haiku 4.5 produced just 2 distinct numbers (47 at 76%); Opus 5.5 was the most varied with 6 numbers but still picked 42 in 61% of chats. The experiment highlights how strongly models' notion of "randomness" is biased toward meme numbers like 42 and 47.
Related event: Test of 221 Chats Finds Claude Says '47' as Random Number 92% of the Time(2 posts)→
More from Fun
- Meme: by late 2026 even a glider game ships a true wind solver — MoonL88537 · 2026-10-20
- Chinese humanoid robot shows off pole dancing skills in viral demo — LexiLove · 2026-10-20
- AI-designed custom LEGO sets: a full branded collection built for Cofounder — noahsolomon · 2026-10-20
- Creators now use performative green screens just to make AI videos look more pro — mrjonfinger · 2026-10-20
- Midjourney sref codes deliver consistent crayon-textured animal illustrations — sergeantsref · 2026-10-20
- Cookie Dad Finds Out He's Not the Father: AI Cartoon Amuses Reddit — Leo-Mitchell83 · 2026-10-20