One of OpenAI's new maths papers settles a 1970s scheduling problem with an "efficient" algorithm that the entire universe isn't big or old enough to run. I dug through the release looking for results that might actually change something outside academia:
