alt Hacker NewsWorld models do not belong here, or at least, we're still some years from figuring out how they would. If I want to text my wife I can imagine just telling my phone that somehow rather than using the current UI paradigm, but what "world model" am I going to pull up that is helpful to do that? World models belong in their own stream, along with rendering in general (movies, etc.), games, VR, and other similar things that we do not today classify as UI, for good reason.
In fact I suspect "world models" may let us re-experience some idiocy from yesteryear we thought we had put behind us, like [1]. Can't wait to go "shopping" in a "world model" of a store again! However do I survive in 2026 merely zipping around the store buying my favorite items off of my favorite's list as fast as I can think of the items and using search on the thousands of available items rather than WASD'ing my way through a "model" of the store.
By contrast I think the browser is undersold. GUI toolkits existed before browsers, but they were all based on widget layouts. That is, the top level of the widget hierarchy would be some layout engine, which had components, which had subcomponents, which had a widget, etc. Some were more dynamic and relative, some used a lot of absolute positioning, but they were all structured in this way. Browsers introduced a new paradigm, where textual layout was the "top level" of the tree, and the widgets all fit within that. Prior to a browser, a Mad Lib-style game where you have text boxes interspersed in a bunch of text was quite difficult. Many GUI toolkits would have required an individual absolutely-laid-out pane for each game you want to play because it couldn't do its own layout on interspersed text and widgets at all; most if not all of them (perhaps Tk excepted, though I'm not familiar with what it could do in the 1990s as I picked it up later) would have made heavy weather of it if they could do it. (Although GUIs made heavy weather of things in general before browsers.) Now all the GUI toolkits have a lot more support for textual-layout like browsers and of course the browsers have carried on like crazy.
AIs-running-in-browsers seem a very powerful paradigm to continue forward with.
With regard to the "world model" I would see "augmented reality" being the major move forward there. It has consistently failed for a long time but I think there's a very plausible case to be made that it was just premature tech, that it doesn't work without the powerful AIs that only recently came out and are still pretty hard to stuff on to a realtime platform. That will start to enable some very interesting UI paradigms here at some point. But that still really doesn't replace WIMPs, it's more a new frontier entirely. Again, I don't necessarily want to "use augmented reality" to text my wife. AR may prompt for it in some particular circumstance, but if I'm originating one out of the blue I'm going to use a conventional UI to do that, not try to wrangle AR into it.