It’s a bit like when a human brain remembers facts, if you sat down and computed the hash values of the thing you were trying to remember, and traced out a DB lookup in your mind, it’d be a lot slower than just “oh yeah I remember that”.
To misquote a certain YouTuber: LLMs are just games, unlimited games, but no games.
Gosh. That would be a very silly use of an LLM.
Bad use of any AI, even image/video models which are certainly less wrong than this. Stuff like DLSS show we're getting stuff that could generate a game's graphics frame-by-frame, but even though it's possible to have an entire game environment (not just graphics, everything) done by an AI, such things demo better than they play due to temporal coherency.
That said, having an AI make the code for each thing as you need it, might work somewhat OK eventually. Wouldn't trust it today (security, data integrity, and a tendency to cheat on tests), but eventually it could happen.
A key point of DLSS (5) is that it is significantly guided by ground truth rasterization or raytracing info. It's not intended to replace graphics pipelines, it augments them to produce more detailed surfaces and interpolate frames more efficiently than calculating all of it traditionally is. But you have to calculate some of it to get these results, it can't be a purely AI solution without looking incoherent.
Doing it in a way that isn't described as "hilarious" is probably also possible in the time since that video came out. Doing it in a way which isn't described as "a security nightmare" or "liable to lose all your data", probably not yet. Not as vibe-coding, at least.
We have Jev to prove LLMs that output constrained decisions are useful. Why not make LLMs just output UI natively?
At a fast enough inference speed, you could just connect the network to it.
So an LLM would then just take the network, user input, and output a UI, with tool calls to deal with memory and other resources. In theory, you don't need an OS any more.
1) Have a single person vibe together a computationally inefficient solution in 1 month that runs transformers on some backend.
2) Hire a team to work for 2 years to design, perfect, optimize, plan, revise, and architect a solution.
The risk/reward for 1) is extremely better than 2). Given the turnover for entertainment in this day and age it doesn't make sense to do all this complex stuff.
are we not talking about the NES?
These are just machines for ripping off existing games/programs with mostly accurate simulations. You can't vibe together a transformer of something that doesn't exist with meaningful control over the results.
> hallucinate an entire operating system every single frame 1. I don't know that "hallucinate" is the right word here... any more than Windows 11 itself is a hallucination, for example 2. I don't personally think that this system is a million (thousands x thousands, from your attestation) times slower, but given Moore's Law, even that isn't the worst thing ever. this is the low water mark.
You're comparing apples to oranges there!
To hallucinate is arbitrary, you can use a 2B image param model, it will hallucinate an OS no problem* running real time!
What this post is about is completely different, which is actually emulating the complete system in an LLM - that is hundreds (or more) times slower.
*absulutely crappy quality tho, unusable OS in practice
Now is that a desirable product? Would it succeed on the merits in the marketplace? Maybe it would...
It's called a Nintendo emulator.
That's kind of the point here.
Code isn't going anywhere. It can be literally billions or trillions of times more efficient than an AI, and as AIs scale up that number just continues to grow and can completely plausibly hit quadrillions in our lifetime. What relationships humans have with code is a completely different question, but the idea itself isn't going anywhere. Anyone who thinks otherwise has failed to grasp just what producing a single token from an AI actually entails.
For example, Minecraft in Minecraft [0] is (while impressive) slow and inefficient too, but it doesn't tell us much about humans except that some of them are weirdly ambitious in their projects.
But at some point we’re hitting a wall here. If we want our software to undo 100,000x more processing powers, we need that llm software emulation layer.