[HN]score: 0.27
GPT-4.1 Asked 10,000 Times to Pick 1–100: Distribution Is Not Uniform
May 25, 2026
Running gpt-4.1 10,000 times on the same "pick a number 1–100" prompt reveals a non-uniform distribution that mirrors human biases toward numbers like 37 and 73, confirming LLMs inherit statistical patterns from human training data.