February 25, 2026
Mark us down as… Marcus
I asked Claude for 37,500 random names, and it can't stop saying Marcus
Internet loses it as Claude keeps naming everyone “Marcus”
TLDR: An experiment asked Claude for random names 37,500 times, and “Marcus” dominated—one version even answered Marcus 100/100 times. Commenters turned it into meme fuel (Gary Marcus, Roman emperors), debated training‑data bias, and shared a simple fix: add random words to prompts to shake up answers.
The internet just discovered Claude’s favorite baby name, and it’s… Marcus. After a wild experiment asking the AI to pick a “random” name 37,500 times, the results came back hilariously one‑sided: Marcus showed up 23.6% of the time, and one model variant even said “Marcus” 100 out of 100 times. Cue chaos. Some commenters went full comedy mode, with one quipping that “Gary Marcus is living in Claude’s head rent‑free,” turning a longtime AI critic into the meme of the day. Others reached for history class: if your data loves Rome, maybe you get Marcus—think Marcus Aurelius and Crassus, says one user.
Then came the lore. “Marcus the Worm infected Claude,” joked another, linking a YouTube short like it’s a found‑footage horror flick. Meanwhile, data sleuths started tossing around theories about training sets and obscure sources, even dropping a link to Amara like they’d cracked a secret code.
Amid the memes, a practical thread emerged: several users fixated on a clever hack from the write‑up—if you want more varied answers, add a few random English words to the start of your prompt. It worked better than noise and doubled the variety in names. So yes, the experiment cost $27.58 and shattered the illusion of dice‑roll randomness—but it also sparked a perfect internet moment: half bug hunt, half Roman Empire fan club, all Marcus.
Key Points
- •37,500 prompts across five Claude models tested how LLMs handle “random” name selection.
- •“Marcus” was the most common male name, appearing 4,367 times (23.6%).
- •Claude Opus 4.5 output “Marcus” 100/100 times with a simple prompt, indicating determinism.
- •Nine parameter combinations yielded zero entropy (perfectly deterministic) outputs.
- •Elaborate prompts and random word seeds increased diversity; word seeds outperformed random noise.