This repository contains an interesting experiment that you can run on GPT-4o that is aimed at exploring emergent knowledge. Some - not all - outputs are categorically hallucinatory. However, the prompt seems to consistently produce obscure and sometimes interesting knowledge that in some cases can be validated.
Please let me know if there is anything I can elaborate on in order to make the framework more clear. I pushed a step-by-step example (1) to the repository.