I asked Claude for 37,500 random names, and it can't stop saying Marcus

GitHub - benjismith/ai-randomness: experiments invoking AI agents and asking them to act randomly!

github.com

February 25, 2026

1 min read

🔥🔥🔥🔥🔥

49/100

Summary

Experiments with AI agents assessed how language models manage randomness, with Claude selected a name at random 37,500 times. The most frequently chosen male name was "Marcus," appearing 4,367 times, and Opus 4.5 produced "Marcus" 100% of the time with a simple prompt, while nine parameter combinations resulted in zero entropy.

Key Takeaways

The name "Marcus" was selected 4,367 times (23.6%) in a series of 37,500 random name prompts across five models.
Opus 4.5 produced "Marcus" 100 out of 100 times with a simple prompt, demonstrating deterministic behavior.
Nine parameter combinations resulted in zero entropy, indicating perfectly deterministic outputs.
Using random word seeds was more effective than random noise for increasing diversity in name selection.

Read original article

View story →

Coverage & discussions

Hacker News · github.com

I asked Claude for 37,500 random names, and it can't stop saying Marcus ↗

Community discussions

Hacker News discussion ↗

Historical count: 5 · 5 saved in sample · Historical sample

Analysis ready

Community Sentiment

Mixed

Positives

Injecting randomness into prompts can enhance the diversity of LLM responses, offering a practical method to overcome limitations of temperature settings.

Concerns

The repeated mention of 'Marcus' highlights a potential collapse in the model's training, raising concerns about its ability to generate truly random outputs.