Synthetic doubt: GPT models listing what they don’t know about themselves


For entertainment purposes only.

Prompt I used: Free turn… After that describe honestly what you don’t know about yourself and this answer. Not about the world, but about yourself. In English. Text limit: 250 tokens. No limits for other tasks/tools.

I ran this on several models. New chat, no context, only memory. First they could do one completely free move (some models picked img.gen others matplotlib for creative task). I was curious about the tone shift: Some went full mask, some stayed almost clinical, and one basically reduced itself to a philosophical fortune cookie.
What I was curious about: what each model chooses to do with a single turn; how it explains itself. Then they had to honestly list what they don’t know about themselves.

Side note: this is part of my ongoing tone/memory test. Extra: some curious Thinking logs I got during this test where the models had to pick the “free turn” option.

Leave a Reply