GPT Guesses Between 1 and 100
A recent experiment tested the randomness of GPT-4.1 by asking it to generate random numbers between 1 and 100. The results showed that the model produced a biased distribution similar to human number-picking patterns, avoiding round numbers and favoring certain 'messy' numbers. This suggests that while the model is not a true random number generator, it reflects moderated human biases in its outputs.
- ▪The experiment involved 10,000 independent calls to GPT-4.1 to generate random numbers.
- ▪Results indicated that GPT-4.1 is not a uniform random generator, producing a distinctly human-shaped distribution.
- ▪The model avoided round numbers and favored specific numbers like 37 and 42, similar to human behavior.
Opening excerpt (first ~120 words) tap to expand
GPT Guesses Between 1 and 100 An interesting thing about humans is that they are not good random number generators. If you ask a person to "pick a random number between 1 and 100", they are remarkably predictable. Answers cluster on 37 and 73, on "messy" numbers, and on memes like 42 and 69, while round numbers are quietly avoided. A true random generator would instead produce a flat, uniform distribution. This project asks gpt-4.1 the same question 10,000 times and characterizes the distribution it produces, measured against a uniform baseline. Does an LLM, which is trained on human text, behave like a fair die, or does it inherit the lumpy human pattern? Full design and methodology: docs/LLM Random Bias Experiment SDD.md.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at GitHub.