▲ 43 points
back
86 comments
Ed Zitron also said that consumers were losing interest in ChatGPT in January 2024, that AI models were reaching their upper possible limits in Feburary 2024, that ChatGPT and similar products were "not particularly useful for anything", that hallucinations could not possibly be reduced in models, etc.
He's a churlish AI skeptic that isn't worth listening to.
If you're wondering why these "churls" are increasingly popular, it's the reality that far from being mobbed by skeptics, the entire market economy is being taken for a ride in the name of AI. It's very much like the febrile environment that developed about Crytpo, in which anyone who expressed concern that maybe this tech wasn't going to revolutionize money, programming, and society itself was a "churlish skeptic".
I certainly wouldn't credit Ed Zitron with timing his predictions very well, but that doesn't make them wrong, and it certainly doesn't support the AI maximalism we're all forced to endure.
Agree. I don't pay attention to people trying to call the top, but I do pay attention to profitability/sustainability concerns.
There's a huge crowd of "wishful thinkers" that want to see AI collapse and will cling to any rationalization that it will do so. This is Zitron's audience.
How exactly is AI "succeeding?" By the numbers, every AI company is on a fast track to going bankrupt at record speeds. All while burning the planet down and making electronics unavailable to literally everyone. This is not sustainable.
Keep a close eye on the "Is AI Profitable Yet?" website. The numbers will open your eyes. Please, enlighten me on how AI is successful.
People still believe that the model intelligence has plateaud and I feel like they live under the rocks. I get the insane capex spend, I get the hype, I even get certain company will go under but to ignore the capability jump in a year is absurd.
The only honest benchmark is "has it cured cancer yet?". It has not, nor moved close to doing so.
People are doing great at marketing the diminishing returns as breakthroughs though.
It hasn't plateaud but the generational leaps are definitely getting smaller.
He's been mostly wrong about the potential for AI, but mostly right about the financials. Don't listen to him for opinions, but don't discount his factual reporting because his opinions are off.
I am very impressed with LLMs, but.....
I don't think we are seeing productivity gains due to LLMs in the general economy yet.
Even in software development, are we seeing an increase in the number of products or features shipped?
For someone like Ed, it feels like way too much effort to wade through the muck.
I personally can't stand anything that whiffs of "outrage content" . Doubly so if they are stupendously wrong on a regular basis. Triply so if they never own up to those bad calls.
If someone else wants to get their hands dirty, more power to them. I'll wait for anything useful to filter through other sources.
Life is short, and there's so much other high quality material I'd rather give my limited attention to.
Zed publishes usage metrics for AI and they are trending down:
This is not a complete picture. I use Zed daily and I have the AI features enabled, but I never use the built in agent in Zed. I reach for an agent in a separate terminal session.
Any YouTube video with him is filled with fawning comments. People love it when others tell them what they want to hear. He'll be able to do his schtick for a long time, well after it should be obvious to everyone the AI is going to change things. I'm not enthusiastic about AI and wish we could pump the breaks somehow but at some point you have to recognize there's something there. I don't think it's helpful that the face of AI criticism says a lot of nonsense.
And you're an equity partner for a PE/VC that funds AI. And?
Hah, you know they feel threatened when capitalists have to whip out the classist insults.
> that AI models were reaching their upper possible limits in Feburary 2024,
I’m curious, removing coding as a criterion what is more impressive about the current models than say gpt 4o? Give a prompt example. Keep in mind most consumers of AI are likely not using it for coding so this is relevant.
I doubt anyone could give a not coding example where it’s meaningfully better with current frontier than 4o.
Take humaneval. 4o gets 90, gpt 5.6 gets 94%. So what?
https://openai.com/index/hello-gpt-4o/
If an iPhone had a 4o quality model that could run locally frontier models would be finished.
Exactly. In all facets of life, it's amazing that people continue to listen to those with a strong track record of confidently wrong predictions.
This seesawing of perspectives on apple’s ai strategy reminds me of that Alan Watts story:
Once upon a time there was a Chinese farmer whose horse ran away. That evening, all of his neighbors came around to commiserate. They said, “We are so sorry to hear your horse has run away. This is most unfortunate.” The farmer said, “Maybe.” The next day the horse came back bringing seven wild horses with it, and in the evening everybody came back and said, “Oh, isn’t that lucky. What a great turn of events. You now have eight horses!” The farmer again said, “Maybe.”
The following day his son tried to break one of the horses, and while riding it, he was thrown and broke his leg. The neighbors then said, “Oh dear, that’s too bad,” and the farmer responded, “Maybe.” The next day the conscription officers came around to conscript people into the army, and they rejected his son because he had a broken leg. Again all the neighbors came around and said, “Isn’t that great!” Again, he said, “Maybe.”
I think people, especially in the valley / sf, truly under estimate how much the average person dislikes AI anything. If you know anyone who does anything in marketing outside of tech they'll tell you if people see the word AI it generally means they'll hate it with out even knowing what it is.
Local models running on Apple hardware are definitely a thing, a low-hanging fruit they are harvesting right now while every pure-play AI company is losing money.
It's not a huge thing outside the consumer/novelty sphere. Apple missed the train on the trillion-dollar ARM datacenter rollout, and will suck those lemons for the next decade.
Ironically for home-lab users (ie those who primarily have access to consumer hardware), apple silicon is probably the best value (and also the most obtainable for running some of the smaller sized models. You can get unified memory of up to 192 GB in the mac studios. The equivalent for the "open market" is probably AMD's halo strix which offers up to 128GB unified memory, and the speed of the memory is much slower.
I guess the Nvidia Jetson AGX Thor also exists, but good luck getting your hands on that... and regardless that's also 128GB.
Put it this way, with 192 GB VRAM you can run deepseekv4... with 128gb only with lots of quantization
Edit: WTH! I was curious and i went to apple's website to price out a high end unit just to price comparison, and it looks like the high ram units are gone now, and they max out at 96GB?? Guess i missed that...
There are three companies I am confident will survive an AI bubble popping: Google, Microsoft and Apple.
A lot has been made of Google's recent earnings being cash flow negative (I think for the first time ever?). Building data centers is something Google is very good at. They've also partially or wholly insulated themselves from a collpase by using Special Purpose Vehicles ("SPVs") where the collateral for loans is simply the GPUs. Even the physical building and the land its own is leased from another subsidiary. Google is still a cash cow. I also think they may break the NVidia mohnopoly with their own TPU hardware given enough time.
Microsoft too is a cash cow and seemingly hasn't bet the farm on Copilot.
And Apple is a real dark horse here. They, like Microsoft, don't seem to have bet the farm on AI but also, they're in a unique position to drive a move to local LLMs with their own silicon. Their ARM CPUs use a shared memory architecture. Nvidia strictly segments the market by limitng VRAM on consumer GPUs. Apple can get around that. They just don't have the raw FLOPS to compete yet. There have been rumors they're targeting the M7 series in 2028 for a big jump in AI performance.
This market is going to boil down to performance-per-Watt (IMHO) and that's going to be interesting to watch as NVidia moves up to an annual product cycle. what will the new generation do to existing deployements when hardware 1-3 years is at a severe PPW disadvantage? we'll see.
Personally I just want to be able to buy 64GB of RAM for $200 and an SSD for under $100 again. You know, like 2025.
> Microsoft too is a cash cow and seemingly hasn't bet the farm on Copilot.
I think a good follow-up question is: which Copilot?
Anyway, Microsoft seems to be leaning hard into the whole "you can't get fired for buying Microsoft" thing that used to be IBM and AWS. Normies all around the world still insist on using Microsoft Windows and Office.
Anecdotally, they supposedly lost a lot of money on GitHub Copilot's inference costs, at least until June. So they'd be far more equipped to survive an AI crash now.
Isn't saying that Apple will survive the AI bubble a bit like saying Walmart will survive the bubble? Apple isn't really involved.
> Zirton misses a few points…
What! Ends with a cliff-hanger!? I NEED TO KNOW WHAT POINTS HE MISSED!!!
(but not badly enough to buy a $100/mo subscription).
Zitron reminds me of that lawyer character in Silicon Valley (HBO). A ~decade later, turned influencer.
[dupe] Discussion on source: https://news.ycombinator.com/item?id=49070427
AI may burst in terms of the subscription model pushed by large corporations here in the US. AI is here to stay as a technology, there is no question at this point that it is powerful effective and a transformative development in human history. Not sure where we go from here but AI is not going away.
Hosted by Apple or not, but isn’t the biggest feature of iOS 27 LLM integration?
The biggest feature is stability, but Apple Intelligence has an excellent toolkit and API.
Timing is the key: if the burst happens, let's say, in the next 12-18 months apple is very well positioned, but every month without the bubble popping add pressure on apples executives and, at some point, the pressure will be so high that they will have to do something (most probably something really silly from the business perspective).
Anyway, I think they are really well positioned for local inference, if local inference will be a thing they will be fine.
Apple never needed to participate in the AI race to zero. As they were already at the finish line.
The past decade has shown people will still buy iPhones even with Siri being the worst assistant of the major players, even with Liquid Glass UI debacle, etc
Their hardware is just that good and the fact that is great for LLMs is just icing
Is this investment advice? Otherwise, who cares? Don't tie your identity to being a fan of a company. Apple clearly did not intend to fumble their attempts at AI they way they have. Arguments that "Actually, Apple is the smartest at AI" are boring.
Ed’s interview covered many different topics. As an aside, he was asked what he thought of Apple’s position. The juicy side quote is now what defines the interview.
It's so popular to think this is a bubble. Some aspects likely are, but the zeal for this to be a bubble is disturbing to me. I get that there will be winners and if you're not one of them, you might be more likely to want this thing to pop. In my mind, that's not the right disposition, you should want the gains to be spread as equitably as possible.
The losses are being spread equitably to anyone who has wanted to buy computing equipment over the last few years. Similarly anyone who wants to buy electricity in a region impacted by data center build outs.
AI is like Robin Hood, just proxied through China lol
if they simply stayed out of it, they would get my business from here on out. too late for that though.
"stayed out of it" needs to be considered relatively though, doesn't it? I don't see Apple spending billions in capex on hyperscaler DCs, and apart from the weird product direction that is Image Playground most of their on-device AI is for genuinely helpful uses like smart removal from images and grammar improvements.
they never had my business; their style over all else along with the combination of planned obsolescence and conspicuous consumption are obscene misuse of the human heuristics.
Google will win the AI war. I really don t see how things can be different. The true question is how much the involvement of Sergei Brin will have been a major part of that win.
[but now that I work in a big corp, I can understand how one smart guy saying “shut the f..k up, managers. I show the path” can be extremely disruptive in those kind of environments. For the better!]
In my personal experience, Gemini seems to have the best image editing and creation capabilities but when it comes to coding, Claude is my favorite. However, Claude is slow and errors out for me often. Grok is also very good when it comes to coding but I seek to run into their rate limits much easier, so I don't use it.
I only use free web version for all btw.
Stock valuation has nothing to do with AI being the most transformational tech ever devised. Like who can't grasp that
It's absurd to call LLMs the most transformational tech ever devised. At a bare minimum, computers in general and electricity both beat out LLMs, as they are prerequisites.
I don't think LLMs can approach, in terms of transformational power, writing, printing, or even the personal computer, but I get what you're saying.