Hard problems as a test of AI capability
A user urges people to give AI hard problems where they can judge success. He says a result he discussed would have taken him a day to produce and ranked near the upper end of his day-long professional work.
TLDR
A user says a result he discussed would have taken him a day of labor to generate and ranked near the upper end of his professional outputs that take a day. He urges others to test AI on hard problems where they can judge success. He also says that, two years earlier, he would have expected the capabilities he was discussing around 2029 or 2030, though he thought current approaches might plateau before then.
Combined views
133.2K
5 Sources, first seen 15h ago