No one is claiming an agent can do 50% of arbitrary tasks. It's just 50% of METR's benchmark set.
> I think you're overestimating, or oversimplifying
Yeah if you only read comments on HN but not the actual linked article you will get oversimplified conclusion. Like, duh?