>That doesn't mean these kinds of systems can't be better and even more useful in 1-2 years.

It's also faulty to assume that improvements will happen linearly. We had a self-driving car mania in 2013-2017 with billions poured into it and people just forgot about it.

Also consider that if easy linear improvements were possible, they would've delayed launching it. AI (software in general?) seems to be the kind of thing where once R&D hits the exponential difficulty wall, they make it generally available.