Fascinating cross-model comparison using cartography
With the following procedure, we’ll be able to extract an (imperfect) image of the world as it exists in an LLM’s tangled web of internal knowledge.
The images due it justice, but essentially it asks for any given coordinate is the location land or sea. And the resulting images probably give cartographers conniptions. it both allows you compare contrast knowledge and make fun of it too?
ps reminds me of this cool site https://what3words.com and obligatory https://xkcd.com/977/