When computers first appeared, they were huge in size. And of course, they were incredibly difficult to use. Only specialists working in the field could operate them, or those who bought directly from companies and told them what kind of computer to build to order. Then came the so-called mainframes. There it became accessible to a wider circle. But still, the computers themselves were somewhere in a central location, and everyone worked on them remotely. And communicated through a terminal, meaning typing commands and getting something printed on screen in response, or later something printed on a printer.
Then personal computers appeared. When a person could already install something there themselves. They were very expensive and slower than those central ones, but gave much more freedom in what and how to do. Then these personal computers became cheaper, simpler to use, more accessible, and there became much more of them. Then these personal computers shrank altogether, became cheaper and turned into smartphones. We carry them with us, and their usage significantly exceeds the usage of personal computers.
And then AI appeared. And notice how resembels old paradime as if we are sent back.
Not even that it rolled back, but you could say a new branch of this whole path appeared — from a complex computer that only uber-narrow specialists use (this is pre-GPT-3), then a phase when it's like a mainframe, where we all connect to a central server and type something there, getting printed responses (meaning through all these ChatGPTs and the like). Now comes the phase like when there were first personal computers. Some are already running models on their own computers. But this is still very rare — difficult and expensive. The era when "mobile" devices for AI will appear is still ahead. And most likely, this branch, though it will be similar, will have many things different in both UX and UI.
I now prefer using those same LLMs on a computer more, my use of mobile devices compared to computers has decreased significantly, largely because of wanting to do something with AI. Well, and for other reasons too, of course. I still want computers to be smarter and more convenient. A unified OS across all devices. Like a cloud-OS that's fully synchronized and knows everything about me. Different types of devices are for different caliber tasks. Something's more convenient on the go, and thus portability will prevail there, something's more convenient to do on a huge screen, with a big keyboard, mouse, tablet. Something on a touchscreen, something without a screen at all, purely by voice. And so on. And as the main agent in all this — an AI that knows everything about me, but at the same time doesn't share my data with any third party without my explicit permission.
Currently on devices, this main "agent" is the file system and its interface, like Windows Explorer or Finder, or Springboard on iOS. They're the ones that kinda "know" about what we work with, storing and managing our files. But these agents are still extremely dumb, and therefore we have to do the same things many times a day. Even today's LLM models would handle that incomparably better.
So it is clear that the power of models is already sufficient for radical improvements, but specifically making products based on them lags far behind. And although it seems like there's now an endless bunch of different AI-based products — all of this is too little, and primarily because they do all this on a superficial level, not at the operating system level. And here, of course, it's a disgrace. Apple simply slept through this entire LLM revolution and still can't get their act together. Google only partially slept through it, already woke up, but institutionally they're weak at good productization. Microsoft is too big, unwieldy, greedy and very weak on taste, and apparently fell behind ideologically so they can't imagine how everything they do could be done differently. I won't even talk about Linux. Unfortunately, nerds aren't capable of thinking creatively with the user first. And these are the main players in the systems field.
Of course, others will appear, new ones, who will do everything better, but because this isn't coming from the current main system players, all this will happen noticeably later — there's just too much to do, and almost everything from scratch. But the "mobile" era of devices might go completely differently, and that's intriguing. I think it'll be even more interesting.