Comment

Comment by electricships

After AlphaZero, people tried to train various combinatorial solvers

IIRC, they were unable to exceed Sota of SAT solvers (which admittedly are the results of decades of research)

maybe we just needed bigger networks, or transformers, or just more training compute

Discussion

Live from the official Hacker News API

A modern reader for Hacker News · SvelteKit + Skeleton