Microsoft has Copilot Plus PCs loaded with AI, and rumors are that Apple is all in on AI, too, but if you don't want AI in everything you do, there is another option: Linux.
At least with the more advanced LLM’s (and I’d assume as well for stuff like image processing and generation), it requires a pretty considerable amount of GPU just to get the thing to run at all, and then even more to spit something out. Some people have enough to run the basics, but most laptops would simply be incapable. And very few people would have resources to get the kind of outputs that the more advanced AI’s produce.
Now, that’s not to say it shouldn’t be an option, or that they force you to have some remote AI baked into your proprietary OS that you can’t remove without breaking user license agreements, just saying that it’s unfortunately harder to implement locally than we both probably wish it was.
That’s true but if you don’t mind the fact that the AI can’t learn anything new you can actually go hardware optimization routes and get pretty good performance. We’re starting to see AI chips being made. They will do for AI what GPUs did for graphics.
However these hardware optimized chips are only for running the AI you still need GPUs for training it. I could see a situation where new models are trained by big companies and then the results are sold to individuals who then buy the packages and install them on local chips.
They’re available on the open market but you have to buy them as integrated systems since no especially available motherboard has a socket for them, don’t even think there’s a standard for a socket. They come soldered to the board which isn’t the best because when a better version comes out you basically have to throw everything away and start again.
But in a few years I suspect we’ll have proper socketed versions.
I don’t want to avoid it. I just want it locally
So… Heu… How to say that… Linux still the solution for this.
I’m sure I can install a local AI on a Windows PC as well. Linux is not the solution to every possible problem in the universe. Oh indeed many of them
Hum did you try to solve this with Linux ? Kidding. Yeah but this included with the previous privacy claims make it a good solution.
deleted by creator
At least with the more advanced LLM’s (and I’d assume as well for stuff like image processing and generation), it requires a pretty considerable amount of GPU just to get the thing to run at all, and then even more to spit something out. Some people have enough to run the basics, but most laptops would simply be incapable. And very few people would have resources to get the kind of outputs that the more advanced AI’s produce.
Now, that’s not to say it shouldn’t be an option, or that they force you to have some remote AI baked into your proprietary OS that you can’t remove without breaking user license agreements, just saying that it’s unfortunately harder to implement locally than we both probably wish it was.
That’s true but if you don’t mind the fact that the AI can’t learn anything new you can actually go hardware optimization routes and get pretty good performance. We’re starting to see AI chips being made. They will do for AI what GPUs did for graphics.
However these hardware optimized chips are only for running the AI you still need GPUs for training it. I could see a situation where new models are trained by big companies and then the results are sold to individuals who then buy the packages and install them on local chips.
interesting. are these ai chips actually being released on open markets yet, or are thongs still in development phases?
They’re available on the open market but you have to buy them as integrated systems since no especially available motherboard has a socket for them, don’t even think there’s a standard for a socket. They come soldered to the board which isn’t the best because when a better version comes out you basically have to throw everything away and start again.
But in a few years I suspect we’ll have proper socketed versions.
https://jan.ai/ Jan works fine on W11