How Does A Blind Model See The Earth?
Large language models can’t see the world. Technically, they don’t even know the world exists. But given everything they’ve been trained on, what do they think the world looks like? This is a fascinating experiment: Give many different large language models coordinates of the world, one by one, and ask a simple question: Is there land or water? Assemble and get maps of the world. Of sorts.
From Weekly Filet #550, in November 2025.
π Some older links might be broken β that's the state of the web, sadly. If you find one, ping me.