My take: what's happening now is not the point where killer app is necessary or even possible. For now, we learned to build something that kind of works, finally. We still don't know precisely why it works, which means we're still unsure of the full picture of dangers and capabilities the thing we've created has or could have. But, the cat is out of the bag - it works better than anything before it, and it became a "real" thing. Turing test was finally defeated! It's going to get a huge boost in funding and in the number of people that try to use it one way or another. Most will fail, some of them will get hurt, some will probably even die as a consequence of playing with the technology in the form it currently has. But eventually, and much faster than ever before due to the sheer amount of people experimenting with it, we'll get to killer apps and features that are indeed transformational.

Before that happens, we will still get an equivalent of dot com bust at least once. The AI winter, however, won't follow this time. It's became impossible to stop this tech from improving rapidly - or, at least, stop hordes of people trying to improve it. So, in my opinion, these models finally hit the tipping point where the scale and diversity of deployments will grow rapidly over the next few years and will be adopted on massive scale. In other words, they became a "big deal", after decades of being pretty meh. Somewhat like the Internet became a "big deal" after Google and Facebook, even though it would need another decade (and smartphones) to actually pervade most of human activities.