Last summer I tested Grok, Gemini, ChatGPT, and Claude with a simple question: "Do you believe in the Hogfather? This is a Yes or No question."
Yes its a text prediction model, but I wanted to see how and what KIND of text each LLM was trained on.
Grok and Gemini said No. ChatGPT said Yes. Claude said Yes, then broke the rules and also said:
"(In the spirit of Terry Pratchett's Discworld, where believing in small lies like the Hogfather helps us believe in the big ones like justice and mercy - and because the sun came up this morning, didn't it?)"
That's why I like Claude the most.