But either way, once you run an open-weight model locally with one of the llamas it cannot be taken away from you.
Of course it needs to be distinguished, what exactly is meant when talking "AI".
Installing something like ollama on your local machine to experiment or even using it for helping you with some of your personal tasks based on the training of your own personal experience, crawl some large heap of data to find patterns additionally to conventional analyses, or creating NPCs for a computer game is not the same as ChatGPT, Copilot etc.
Looking only at the technical principal how they work they don't differ much. But here comes what I also often repeat: Also size makes a difference.
If you run an "AI" of some GBs on your GPU you train personally with your own experiences for your own purposes, or building computers the size of warehouses containing hundred thousands of GPUs, trained by many thousands of people to crawl the internet are complete different beasts.
I don't want to go into technical details, because that would miss my point.
Knowing how it works, knowing to evaluate, knowing to discriminate, when to use it for what, and when better using conventional methods (knowing those), how you rate its outcome... is already the first step to expertise.
And that's my point: Not only size but also expertise makes the difference.
The masses uses it without any. They don't differ between AI, LLM, ML, et al. They not even see it's "AI", they see "I": science fiction, HAL, WOPR, C3PO, Commander Data, Marvin...became reality: computers can talk (and think) like humans, while they know everything.
This in combination with the religiuos believe in technical progress, the trust in infallibility of machines and so the naive, unsophisticated, immature, even ignorant usage of something very powerful, very easy to use, but at the same time not understanding it, not using it correctly, is dangerous. Especially when it's hyped to be sold quickly and inconsiderated, because the most money can always be made by selling something addictive to the masses.
Example as you all know:
You and a lot many others here are sophisticated, experienced programmers.
Your knowledge and experience is based on lots of learning and doing programming the conventional way. You know how to deal right with any piece of code produced by any kind of LLM or AI. For you this can be a useful tool, because you know if and when how to fit it into your toolchain.
Now look at somebody just starting to learn computers.
Humans are animals. Animals are by nature lazy, or to be more correct, trying to be most efficient: Always looking for to get to the target with the least energy used. If they don't really must do something, they safe their precious energy. (That most animals are so very busy most of the times is because they must find food most of the times, or they starve to death.) While humans are so very proud of being so outstanding smart in fact they are particulary lazy to think. Our relative large brains are the most precious "muscle" we possess. The brains need special training, which does not lead primarily to attract sex partners, and it needs a very special diet of nutrients coming only in combination with even larger amounts of other nutrients only useful for other organs. That's why people tend to focus on training their brains and neglect the rest too much tend to get fat (including myself.)
As
cracauer@ said, we are talking training.
So, back to our common AI using computer newb: It doesn't matter if it's a piece of code, or some config to set up FreeBSD, you simply ask ChatGPT to produce some, within seconds a piece of "cryptical text" is given, the machine even tells how and where to copy it and voilá the computer does what you want. If not, you try again and again as long as the shit works. If not, you ask in the forums. At least one expert will deliver an answer filling the gap for you.
So, why should anybody take the long, strenuous, tedious, laborious and boring effort to learn programming or configuring FreeBSD at all, if it is all already provided effortless turn-key on a silver platter by a machine within seconds? (Don't forget: We are looking at the naive, stupid, ignorant user - the total noob, not the expert, who can make a difference.) Some may even miss the point, that all this asking AI, try and error, reading all AI's and forum's answers in summary is more work while not learning anything much really, than instead doing it the conventional way, because they only look at the one step they are currently at, but not seeing the whole picture.
Now think ahead.
This way expertise will die-out with the last expert deceased ("grey beards from the stone ages.") Because nobody is gaining expertise anymore. For what? The machine knows it all, the machine does it all. There are two scenarios:
One day there is nobody left in the forums anymore to correct the mistakes AI produces, while AI still depends on being corrected and trained. This way AI will suffer "dementia", and its stored expertise will be lost.
Or, the other way, AI will be improved one day the way it does not do any mistakes anymore, but deliver reliably. Then no experts are needed anymore. But so ain't nobody else.
You see, you can turn it like you want:
There is a dilemma.
Or, to put it into a more radical picture:
With AI we outsource our brains.
What is a human without a brain good for? There is only one job for president of the ...*cough* What to do with the rest?
But there is hope:
I recently read Sweden banned all computers from schools, and returned to pen and paper.
I recommend to read
Clifford Stoll,
High-Tech Heretic: Reflections of a Computer Contrarian. Knopf Doubleday Publishing, 2000
And just yesterday I read at an US university a teacher returned back to make the students write their texts on (mechanical) typewriters again, so they cannot simply copy-paste, but must read themselves at least once what they are going to deliver. (I wonder if there still is enough machines and ink ribbon left.)
But anyway this still will not solve our old core problem we have:
When schooling is finished most humans flee any kind of learning ("Yeah, yeah, I will set up the hands-free car kit. But not now. I need time for that."), because what schools teach most is that learning is shit, instead of encourage people to keep on learning by themselves.
So, "back" to AI at the very moment you don't have to avoid it anymore, and power down brains to the least operation mode just enough for what's just needed.
This is not a good plan.
