Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple Will 'Watch Everything Burn' When AI Bubble Bursts - Ed Zitron
- Thread starter MacRumors
- Start date
- Sort by reaction score
There's one more piece of the puzzle as to why LLM AI is indeed a fragile bubble. It's the technology itself. No one stops to wonder why these massive energy sucking, water consuming data centers are needed in the first place.
LLM is a brute force mechanism that simulates "intelligence," but it has none and never will. All LLM AI does is catalog massive amounts of data and use massive amounts of energy to brute force probability calculations for what word should come next. It takes material (data) that has already been created and pieces it together into collages. There is nothing original. There is no converging on sentience. Its hype is built entirely around the human tendency to anthropomorphize anything that can remotely be perceived as human, and words in a string that looks like a sentence looks human, even when it isn't. It takes LLM AI massive catalogs of data and massive amounts of computational power to create simulated output that is less reliable than what can be produced by a college sophomore powered by three slices of pizza.
That's not sustainable. Adding more computational power is already yielding diminishing returns, and the feedback loop of unreliable AI output going back into the cataloged data on which LLM agents are built only makes it worse.
Now, bring in the argument that consumers are realizing that LLM AI output is middling and unreliable, and they're not going to keep paying for it.
-Pop-
LLM is a brute force mechanism that simulates "intelligence," but it has none and never will. All LLM AI does is catalog massive amounts of data and use massive amounts of energy to brute force probability calculations for what word should come next. It takes material (data) that has already been created and pieces it together into collages. There is nothing original. There is no converging on sentience. Its hype is built entirely around the human tendency to anthropomorphize anything that can remotely be perceived as human, and words in a string that looks like a sentence looks human, even when it isn't. It takes LLM AI massive catalogs of data and massive amounts of computational power to create simulated output that is less reliable than what can be produced by a college sophomore powered by three slices of pizza.
That's not sustainable. Adding more computational power is already yielding diminishing returns, and the feedback loop of unreliable AI output going back into the cataloged data on which LLM agents are built only makes it worse.
Now, bring in the argument that consumers are realizing that LLM AI output is middling and unreliable, and they're not going to keep paying for it.
-Pop-
thebarrkbyte
macrumors member
Totally agree with him. Needs $1.5 Trillion IN PROFIT not revenue to be a business.
And what does it do other than summarize things? I have yet to see anybody explain what it does other than condense. I have used it to write code, and it sucks. I spent more time fixing bugs it made than the time it saved.
And it is ridiculously expensive for the amount of times it gets things dead wrong.
And what does it do other than summarize things? I have yet to see anybody explain what it does other than condense. I have used it to write code, and it sucks. I spent more time fixing bugs it made than the time it saved.
And it is ridiculously expensive for the amount of times it gets things dead wrong.
That's because in the end, most AI as we think about it and utilize it will be merely a feature, not a standalone product. Is that worth billions upon billions of investment?
Prophets of doom always extrapolate from today’s prices as if compute costs were carved in stone. Meanwhile the price of a given level of intelligence keeps falling, quietly, whether or not it fits the narrative.
Sure, turn an agent loose and it’ll incinerate money for hours. That’s a choice made by companies competing on demo quality, not a description of what normal people do with these things. Normal people ask questions and build small apps, and that gets cheaper every year until one day it’s just… profitable.
The crash and the technology are separate stories. Conflating them sells better, though.
Sure, turn an agent loose and it’ll incinerate money for hours. That’s a choice made by companies competing on demo quality, not a description of what normal people do with these things. Normal people ask questions and build small apps, and that gets cheaper every year until one day it’s just… profitable.
The crash and the technology are separate stories. Conflating them sells better, though.
The AI bubble is definitely going to burst, and I'm glad Apple didn't throw itself in whole hog, whatever I may think about their current state of things (which I, personally, am not a fan of for the most part), and this allowing for the fact that advanced algorithms are not something we can objectively afford to ignore on a broader scale.
After the past couple of tech bubbles, though, you'd think people would learn, but no. The gold rush paradigm is actually pretty predictable, and still many sink their teeth into it. I don't have a whole lot of sympathy left for any of those people, honestly.
After the past couple of tech bubbles, though, you'd think people would learn, but no. The gold rush paradigm is actually pretty predictable, and still many sink their teeth into it. I don't have a whole lot of sympathy left for any of those people, honestly.
Yeah, if that’s what he’s saying, then he doesn’t understand the technology at some very basic level as the device actually ASKS you to remove the device before an update. So, either he lost the ability to communicate simple ideas effectively at the end, there, OR he doesn’t understand the relatively simple tech of the Vision Pro. And if the doesn’t understand something so simple that all other companies are copying Apple’s interaction model, one wonders about their understanding of the rest.
jazzwill
macrumors newbie
Yeah, OK Ed
Just like the Internet is a fad.
Just like the Internet is a fad.
The K-T extinction event didn't wipe out all life on earth, it just made the world impossible for dinosaurs to survive. When the AI bubble bursts, it will wipe out trillions of dollars and plenty of companies, but that's a financial crisis, not a technological one.
AI as a technology is here to stay. The insane investment and speculative valuation of companies like OpenAI is what will evaporate once it's apparent that there is nowhere near enough revenue to keep all these data centres profitable.
As many people have pointed out, the railway boom of the late 1800s is a useful parallel, as is the dotcom bubble and the fibre-optic craze. Many people and companies were ruined financially, but the technology persisted and most of it eventually became useful.
jazzwill
macrumors newbie
Agreed. Can't read anything at MR about AI that does not have at least 10 'slop' comments or 'who wants this?'
Technology is not going to be a panacea for everyone. It's like everyone who uses tech has forgotten about ITERATION!
AI, Vision Pro, iPhone, Apple, Siri (to name a few) will be wildly different in a year, 5 years, 10 years.
But, unless you paid hundreds of thousands of dollars (along with those two other people) generating that massive amount of Matlab code, there’s no way what you did was profitable for those that own and operate Claude. (And with the revenue hole that generation created, the Matlab license doesn’t offset that. Far from it.) And, if individuals are unwilling to pay significantly more for the service, and companies are curtailing their efforts to spend more, profit and loss will work its magic soon enough.
I share these thoughts.
One thing seems certain and that's the timeline, a degree of unpredictability prevails.
This continues to be a very interesting story to follow.
Pets.com was just early (which is the same as wrong). Chewy is basically the same thing and has a current market cap of ~$9B. 🙂
With results like the one above, it causes one to wonder, just happened to Apple testing and quality control?
Once upon a time, Apple delivered reliable hardware and software.... I miss those days.
I believe AI, as a technology, is here to stay and will become an established in one form or another.
At the same time, I believe the current valuations of AI companies are far too high. Investors are pouring money into them partly out of fear of missing out, even though no one really knows yet where AI will create the most value or which business models will ultimately succeed.
To me, it feels similar to the dot-com bubble. After that bubble burst, information technology and the internet went on to create enormous value but not because the initial hype was justified. The real value came later, through genuinely useful products, services, and business models that the technology made possible. That value was created gradually over many years.
I think we’ll see a similar pattern with AI. First, the hype will fade and valuations will likely come down significantly. Then, over the years, we’ll see steady and sustainable value creation from companies that develop truly innovative products and services where AI is an enabling technology rather than the entire story.
At the same time, I believe the current valuations of AI companies are far too high. Investors are pouring money into them partly out of fear of missing out, even though no one really knows yet where AI will create the most value or which business models will ultimately succeed.
To me, it feels similar to the dot-com bubble. After that bubble burst, information technology and the internet went on to create enormous value but not because the initial hype was justified. The real value came later, through genuinely useful products, services, and business models that the technology made possible. That value was created gradually over many years.
I think we’ll see a similar pattern with AI. First, the hype will fade and valuations will likely come down significantly. Then, over the years, we’ll see steady and sustainable value creation from companies that develop truly innovative products and services where AI is an enabling technology rather than the entire story.
These data centers are in a unique predicament. They are built up super big because the models use lots of processing, but I can easily forsee two types of breakthroughs that will quickly render them obsolete. A combination of new thinking on the core of the software coupled with brand new hardware that's highly optimized just for these models (no more GPUs). When that happens you can then either open a brand new data center for a lot less money or you can replace old data centers with much more capable ones. That's a big expense but it will be warranted given the savings in the long run. When that happens, were does all the old hardware go? Can it be repurposed or will it be old and obsolete and trashed?
I don't disagree. It was the exuberance surrounding early e-commerce that ended up killing it. Had Pets.com just taken a measured approach (like Amazon did), they would likely still be around today and in Chewy's place. But, instead, they bought into their own hype and imploded in three years.
That's the point all of us are making about the AI bubble. AI is real (Deep Blue's development started 41 years ago; it beat Kasparov 29 years ago), but the hype is just that—hype. It's sound and fury signifying nothing, and the more money that's thrown at it, the larger the fire will be when the hype all burns to the ground.
Chewy.com is a meme stock just like many of Ryan Cohen's other companies. It is ripe for the AI bust.
In some cases, crypto mining hardware is being reused as Steam Machine clones (see what can be done with the Asrock AMD BC250). One would hope the current hardware can be repurposed and reused in some way. I have an image in my head (that I hope never comes to pass) of squatters in a dark data center surrounded by racks and racks and racks of useless hardware, though.
You understand that you're describing OpenAI here, right? There is no possible path to profitability for that company once the VC money runs out. That's been Zitron's whole point for months now. The larger the models get, the higher the inference costs. When companies receive the *real* bill for their token usage, they sprint for the off switch. Meanwhile the open-weight models are a fraction of the cost and are proving to be "good enough" for most situations.
The frontier models are like Formula One racing teams. They're truly impressive, but 99% of us don't have much use for a McLaren or a Ferrari when a Yaris or a Chevy Spark will do.
That wasn’t AI. Well, you could call it AI but it was a powerful computer brute forcing all possible chess moves (not exactly but almost). Which is quite different than LLM-s.
AI had to start somewhere, and it beat Kasparov. Deep Blue had to crawl for LLM's to reach this toddler stage.
So, he doesn't actually mean software update.
The interpretation is from the literary sense of "update" as "Change".
[The AVP (Apple Vision Pro) needs to be changed so that it is so invisible you can wear it all the time like contact lenses, or a pair of correction glasses, or perhaps a neural link.]
The interviewee was deliberately trying to be "edgy" but failing so because anyone even tech adjacent thinks of update as a software update whereas writers think of update as change.
It's really annoying to read things like this and I believe the editor should have known better, but likewise they probably believed that this was "edgy" as well, and those of us that noticed are now in for the detriment of it all.
These days you have to be edgy to piss people off and farm engagement.