Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
The truly ironic thing with all this AI crap is that the more the big AI companies build out their data centers and suck all the available RAM available, the harder and more expensive it’s going to get for consumers to afford computers to utilize these companies AI.

The level of shortsightedness is astounding, albeit hardly surprising.
1788125818966.png


If these trends continue...AYYEE

Also, not much of a computer is necessary to use web apps.
 
Making Apple look like the company is still clueless about AI years after Federighi toyed with GitHub Copilot at the end of 2022 isn't even remotely believable anymore. It's more likely this is some strategy to boost Ternus' profile by distancing him from out-of-touch leadership.

Exactly. The story makes little sense if taken credibly and literally. I think there's more to it than what was reported, and a juicy tale from someone who has a trial balloon to float or an axe to grind makes more sense to me.
 
Last edited:
Apple is not immune for GPU, CPU or modems. They are still built in the same places. I am not sure whether a memory fab facility would be the answer as it requires different processes to build. 2nm vs other nm as examples.
Any idea can always be nixed with "is not perfect/immune etc.". Apple could have one of these ghoul companies build a facility that only makes RAM etc. for them, or they could do their own thing.
 
The Mini's are not even good for AI though..

It's simply all the stupid YouTuber's hyping them but at this point we should know better than to trust them.

You're way better off building a desktop with RTX.
Even a cheap 12gb vram RTX will destroy an M5 Pro.

Even then your local AI options are extremely limited in reality.
24gb vram gets you into a bit better league but still very limited.
After about 30gb vram You're shooting straight up to around 128gb to be in the next tier but that's a whole different price level and makes sense for very few people to do that rather then just paying for cloud.

We need to reach the point that most people have come to this realization after testing and then the bubble of buying this hardware like Mini's will pop.
The more tech savy have already gotten there. It's just the less tech savy and more gullible people believing what they've heard.
A lot of them won't even end up using local AI after buying. Another good chunk will not realize how much better of they would be with using cloud.
True bubble and if the tech doesn't catch up to the lie of local AI fast enough then it will pop.

Then there's the whole issue of cloud AI pricing that could change a lot.
Another time bomb. Already been bursting with many companies seeing the massive bills for API usage and canceling or extremely restricting budget for AI.
Providers need to very quickly figure out how to make their Frontier models run more efficiently.
"The lie of local AI" is already pretty disqualifying. The rest is even wilder. Come back with some actual results and after some spellchecking.
 
Seems like most people here haven't heard of AWS Bedrock which is understandable. But it is possible to run frontier models in the cloud without any data going to the model provider.

I think Azure probably has a similar offering at this point.

Besides for most companies large enough to think about this stuff, what matters more is the terms in the contracts. If they signed a contract with Microsoft for Teams meeting summarization, you think they're really gonna accept terms where Microsoft is allowed to train on all of their meeting recordings?
 
Macs are now the cheapest and best way to run LLMs locally. The problem with cloud providers is privacy, an issue that doesn't exist with local LLMs. And they've gotten really good; you don't need frontier models to do the basic stuff like summarize your emails, build a website, or analyze a financial statement using a set of criteria.

The field is really crazy, and improvements to efficiency and throughput are everywhere. Keeping track of everything is well-neigh impossible.

But local LLMs? They're great.

I wonder... are we seeing the first signs of a turning point where consumers and companies will move away from cloud services, including AI, and instead have more powerful hardware on-site (again)?
 
  • Like
Reactions: Smartuser
Seems like most people here haven't heard of AWS Bedrock which is understandable. But it is possible to run frontier models in the cloud without any data going to the model provider.
And I had assurances from all kinds of companies that they weren't going to be hacked.

We were talking about situations in which companies or people want certainty that their data does not leave their computers. I'm not seeing where AWS Bedrock would be a solution to that problem.

I wonder... are we seeing the first signs of a turning point where consumers and companies will move away from cloud services, including AI, and instead have more powerful hardware on-site (again)?
I'd need to see actual data. But in any event, people have always moved from mainframe-y or centralized things to those they can own and control themselves.
 
I keep wondering when some company - a new start-up, perhaps? - comes out with a product/device that allows one to use dramatically less RAM while getting even better performance than we do now and costing dramatically less than we pay now. Yes, I’m sure keepers of the status quo will say that’s impossible. But almost nothing really is impossible. I mean, if the market is grossly distorted and isn’t providing us with what we need now, then history tells us something is likely to come along to change that situation. I’ll bet the answer is simpler than we might think.
Its apps and websites. They havent needed to care about RAM for awhile so there is no need to optimize. Now there is a big of a crunch but the developers have little incentive to change anything because "this will end soon"
 
Price Price Price helps with the local ai stuff maybe.

Ai on premises while they are building out 1Trillion$ plus around the world I dont think so Tim!
Over 2000 large and small Ai Data centers on the books to build out latest study for the world.

Most USA states have 3 Ai data centers on the books to build out now in each state average.
The government knows more about you then you do so it will all then be in the Ai data centers for good or bad.
Look your self up will be so easy.
Better be good boys and girls its coming the Matrix will know everything.
 
  • Haha
Reactions: August West
If they had a clue those leaked servers would already be products piling out the door to those enterprises instead of the desktop market begin destroyed (ram shortages excepted).

By the time they do react the bubble will have burst.

Lets hope post bust turns into useful things like .com did.
 
Price Price Price helps with the local ai stuff maybe.

Ai on premises while they are building out 1Trillion$ plus around the world I dont think so Tim!
Over 2000 large and small Ai Data centers on the books to build out latest study for the world.

Most USA states have 3 Ai data centers on the books to build out now in each state average.
The government knows more about you then you do so it will all then be in the Ai data centers for good or bad.
Look your self up will be so easy.
Better be good boys and girls its coming the Matrix will know everything.
Do you want a tinfoil hat for your 14th birthday?
 
For most applications cloud AI is cheaper. Someone pointed out that the Apple Lease for the biggest studio costs as much as a claud pro subscription, or something like that, and is slower.

But you know, if you want to mess around with your models, or use abliterated models, or mess with runtimes local is the way to go. I'm a happy gemini pro user, but my 128GB studio buy is paying off because I can do a bunch of experiments locally.
 
1. That $20K is cheaper over time

People really need to smarten up and stop trying to use this math because it doesn't math. At least not for the vast majority.

Only very specific use cases will that work.
First big hurdle is you have to be running something a local model can do just as well as a cloud model can.
For most this doesn't pass, meaning you can't even do the task the local way.
Done.
 
  • Like
Reactions: Biro
Yea the Mac mini or even the Studio isn't as fast as the Data Centers, so what? Its for home use, the Atari 2600 wasn't as capable as an arcade cabinet when it came out. People didn't care. Eventually the home gaming systems became on par and then superseded the arcade cabinets. The same thing is happening. As I said, you young guys just don't have the perspective. But this isn't new, don't be scared. The PC didn't crater the economy because they replaced most white collar workers, because it did, look at movies from the 60-80s and you'll see a ton of employees working in offices. Those jobs are long gone but people and the economy adapt. We will adapt to AI as well. Its ok guys.
 
  • Love
Reactions: Smartuser
For the price of the Mac Studio I’ll stick to the grace blackwell Nvidia stuff I can get from work for less. 2 chained with 128gb fiber can discover other ones on my network so can go super fast and then distribute if needed.
 
In the name of pure “line go up!” mentality, Apple continues to push for ever-increasing AI usage and demand, which will inevitably cause costs for customers to continue to soar, who will in turn also continue to demand MOAR AI and “line go up!” shareholder mentality, but also wonder why their new Mac costs $7,999.

It’s going to crash eventually. A new CEO will hopefully be looking ahead 10-15 years.

No one, wonders why costs went up. Everyone knows… it’s an industry-wide issue due to memory shortages. In fact, RAM prices increased by 400% late last year and Apple just recently started increasing their prices in June.

It costs me nothing to use Apple’s AI, it’s perfectly capable of running on my original M1 Mac mini, a system that’s nearly 6 years old now.
 
  • Like
Reactions: Smartuser
I do wonder if they've made the problem worst. We were banning on the ai companies going bust because the public will never pay. Now local/Siri/Gemini/Bixby are good enough they've pivoted to corporate clients and are actually making money.
 
Yea the Mac mini or even the Studio isn't as fast as the Data Centers, so what? Its for home use, the Atari 2600 wasn't as capable as an arcade cabinet when it came out. People didn't care. Eventually the home gaming systems became on par and then superseded the arcade cabinets. The same thing is happening. As I said, you young guys just don't have the perspective. But this isn't new, don't be scared. The PC didn't crater the economy because they replaced most white collar workers, because it did, look at movies from the 60-80s and you'll see a ton of employees working in offices. Those jobs are long gone but people and the economy adapt. We will adapt to AI as well. Its ok guys.
I was going to bring up arcade machines and home consoles, but figured hardly anyone would know about arcade machines.

In terms of home AI, we're approaching the ColecoVision era 🙂.
 
  • Like
Reactions: norbinhouston
People really need to smarten up and stop trying to use this math because it doesn't math. At least not for the vast majority.

Only very specific use cases will that work.
That's the point I'M making. The statement I was refuting was "why would anyone spend $20K", not why would everyone. No-one is claiming it's a blanket solution for all people.

There is absolutely a use case for a $20K Mac Studio as a local LLM host that saves money long term AND preserves privacy. In fact, Apple won't to be able to build enough of them even for just this targeted market segment.
 
  • Like
Reactions: Smartuser
For the price of the Mac Studio I’ll stick to the grace blackwell Nvidia stuff I can get from work for less. 2 chained with 128gb fiber can discover other ones on my network so can go super fast and then distribute if needed.
It's an honor to see you post here, Mr. Huang.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.