The world is being quietly rearranged by people who write very long documents.


The title they went with Do Neurons Dream of Primitive Operators? Wake-Sleep Compression Rediscovers Schank's Event Semantics Noisy translates that to

AI system rediscovers how humans actually think about events


Researchers used an automated learning technique to figure out what basic building blocks humans use to describe what happens in the world — and found that an AI system trained only on compression pressure (making explanations as short as possible) independently rediscovered the same primitive operations a linguist proposed by hand in the 1970s. This suggests that the way human brains organize knowledge about events isn't arbitrary — it's actually the most efficient way to compress and explain what we observe.
For decades, cognitive scientists have argued about whether the primitives they identified for understanding events are real properties of human thinking or just one linguist's intuition. This paper shows the same primitives can be derived from pure mathematical pressure to compress information, which means they're likely capturing something genuine about how efficient explanation works — not just Schank's arbitrary choices.

If you insist
Read the original →