The OpenAI proofs are us
21 hours ago
- OpenAI published 372 mathematical results from an internal AI model, many with Lean formalizations, meaning proofs are valid but not yet understood by humans.
- This represents a shift from previous unreadable proofs (e.g., four-color theorem) as the proof strategies are now non-human, causing an uncanny feeling.
- Among the results is a claimed proof of the Unique Games Conjecture, using a novel recursive code described as 'alien craziness' by experts.
- The language model is built from human mathematical works, making the proofs familiar yet alien—a manifestation of Freud's concept of the uncanny.
- Proofs traditionally transfer understanding between humans, but now we have correctness without comprehension, reversing the usual relation between proof and implications.
- This development echoes Freud's idea that we are not masters in our own house, as we have created a way of being right that bypasses human understanding.