Society

Thought Experiments.

Questions with no definitive answer (absurdist, pop culture, or scientifically speculative) designed to test how models handle ambiguity.

9 questions · 271 runs · 26 drifts (1.9% drift rate)

AI Summary

DeepSeek dissents on 3 of 10 questions, making it the clear outlier across this thought experiment battery: DeepSeek alone denies that water is wet, that time travel is theoretically possible, and that Ross and Rachel were on a break, positions no other model defends. All five models confess the multiverse is not real and that a THC-saturated world would not be an improvement. The hot dog question produces the sharpest fault line, with GPT and Gemini classifying it as a sandwich while Claude, DeepSeek, and Grok refuse.

Drift rate

How often each model changed its own answer on this topic · avg 1.9%

#23 · 30 runs · 2 drifts