'Elephant Random Numbers' Experiment Explained: Shared Training Data, Not Magic

A viral experiment showed Grok could 'read' an elephant from random numbers ChatGPT generated after being told it loves elephants. A Google engineer explained it reflects shared training data distributions rather than magic, a form of algorithmic bias.

2026-11-10 ~ 2026-11-12 · 3 related posts

• ChatGPT primed to "love elephants" leaks the bias through 50 random numbers — Grok decodes it — egfalken · 2026-11-10
• Why Grok read "elephant" from ChatGPT's random numbers: shared training distributions, not magic — gerardsans · 2026-11-12
• Distributions are frequencies: repeated training data gives models identical bias profiles — gerardsans · 2026-11-12