I've always been super curious how small language models model information and how accurate they really are. So I've created mindworld.space - a Llama-based Wikipedia where there is no search, and content is only traversed from other content.

Unlike most AI tools, hallucinations are encouraged (for fun). The model used is Llama 3.2 1B, which is 8x smaller than English Wikipedia, so it's not going to perfectly recite information.

Keen to hear your thoughts and what entities you can find, and any funny inaccuracies (i.e. Albert Einstein was married to Marie Curie).

Behind the scenes, I built it in a few hours using OpenRouter and SQLite, with no advanced features (such as entity de-duplication). Will open source over the next few days)

Mandatory Warning: It is inaccurate and just for fun.