Something interesting happened last night: Researchers started proposing tighter conditional bounds for integer multiplication, submitted by OpenAI yesterday, then a swarm of collaborators online pushed it even further.
Its so interesting that human researchers provided the insights to push to the out-of distribution solutions that no one model reached alone.
I just submitted the current new best conditional bound, conditional κ = 4.123863984e-5, and I have a very useful system setup for auto-research that requires human intuition to reach out-of-distribution solutions.
https://github.com/CrocSwap/integer-mult-bounds/pull/37#even...
Something interesting happened last night: Researchers started proposing tighter conditional bounds for integer multiplication, submitted by OpenAI yesterday, then a swarm of collaborators online pushed it even further. Its so interesting that human researchers provided the insights to push to the out-of distribution solutions that no one model reached alone.
I just submitted the current new best conditional bound, conditional κ = 4.123863984e-5, and I have a very useful system setup for auto-research that requires human intuition to reach out-of-distribution solutions.
I have a list of ideas to test so we may not be done yet :) Original paper from OpenAI: https://github.com/openai/math/blob/main/preprints/Integer-m...