Chart shows LLMs asked to pick a number 1-10 almost always converge on 7

gerardsans · x · 2026-11-20

A viral chart visualizes what happens when you ask models to "pick a number between 1 and 10": humans spread loosely around 5-7, true randomness is flat, but the LLM distribution spikes sharply at 7 — a neat illustration that LLM "randomness" is highly deterministic sampling, compounded by training-data bias.

Original post →

More from Fun

Fun channel →