GPT Guesses Between 1 and 100
Summary
The article chronicles an experiment to test whether GPT-4.1 outputs a uniform distribution when asked to guess a number between 1 and 100. It finds human-like biases in the model’s outputs, including spikes around 37, 42, and 73, and an aversion to round numbers, with an unusual exception at 69. The piece outlines the methodology, data pipeline, and implications for understanding LLM randomness and safety guardrails.