A laboratory for exploring how things became thinkable
Imagine every possible way a sentence could go as a landscape.
How machines think is so complicated, we barely understand it ourselves.
Our minds reach for familiar metaphors. Imagine every possible way a sentence could go as a landscape. The unlikely continuations stand high as peaks; the plausible ones lie low in the valleys. A raindrop that lands there never sees the whole map, but gravity pulls it off the improbable heights and down into the more likely valleys below.
What if we could understand AI models through a kind of cartography? That is the aim of Raindrop: a laboratory for exploring how things became thinkable.