> I think the cutoff date is closing in on our current now. If models cannot easily become bigger, they will likely advertise using "up-to-date-ness". Maybe they will be merely a few days behind. Or bigger models will make use of smaller but more up-to-date models

But they will still rely on assembly, C, Rust, Linux, HTML, TCP/IP... Doesn't matter how up to date they are, they rely on existing code they have been trained on, they can't just create new languages without the training data.