While I generally agree that local models will gain popularity and Apple is generally well situated to capitalize on this, articles like this are not the way to convince anyone.
Plots with no axis labels or units displayed? In fact if you're literally just comparing two things, why not just write down the two numbers? Or better yet strengthen your argument by comparing more than just literally two options?
And the pictures. You seriously couldn't be bothered to just go find a picture of four Mac minis stacked on top of each other, and had to genai a picture of four vaguely-mac-mini-shaped blocks?
So the argument of this _very_ Claude written piece is that.. On apple devices you have GPUs with fast-enough unified memory and can link them together?
Okay, fair, but why would I get an Apple device when I can get a Linux device with some monster motherboard that has a dozen memory slots?
I submitted this ( I am not the author ) because I thought it was interesting take, I honestly didn't think it was AI, maybe AI polished, but not AI. Was surprised when it was flagged and I saw over on HN Removals . Ran through Claude and got this back:
"Overall, my honest read: this feels like a human-written opinion piece (possibly AI-assisted for polish/editing, which is extremely common now), not a fully AI-generated article. The specific personal history ("I've been saying this in group chats for months"), the running joke about being right before everyone else, and the reader engagement in the comments (with the author replying) all point to a real person with a strong personality behind it — though I can't rule out that Claude or another model helped tighten the prose."
It is interesting that everyone was so very certain it was AI written. Just read your comment about it being Claude based, now I am trying to figure out if Claude is trying to recuse itself in some way by giving me those results.
Still feel like the point it interesting and going to keep diving into it and I am surprised it was flagged. May have to up my reading on Removals more and more.
I don’t think that a generative model is a good judge to assess if a text was written by itself.
This article has a super stong
“written by AI” smell to it. Once you notice all the em dashes and “it isn’t X, it’s Y”patterns you can’t unsee it.
That plus the sloppy images, make it really hard for me to finish reading.
Looks like Mac Minis had a 64 GB model (hard to get now) for roughly $1,600. You could get a 3 of these (192GB) & thanks to the Thunderbolt 5 connection, they out performed any single board AMD for a similar price that had more memory. If you tried to do so something similar with Nvidia GPU you're looking at a lot of watts, cooling & a much faster experience but only 48 GB of memory.
AMD's Strix Halo has a 256-bit memory bus. Apple's Max class chips have a 512-bit memory bus, Ultra class chips have a 1024-bit memory bus. Unless you want to go for datacenter hardware, AMD's offerings simply aren't in the same league in terms of memory bandwidth.
if the author wasn't too lazy - this article could've been way shorter.
but of course author chose to use a.i to write this.
& bury very important points which are 1. Chinese Memory Manufacturers 2. Apple Unified Memory + their own chips which are a result of their 'car' effort.
> Every historical technology wave ends the same way: the mainframe becomes the PC, the timeshare becomes the laptop, the cloud becomes the edge. The only people who ever argue otherwise are the people selling mainframes.
Perhaps AI serves also as an attempt to reverse course on this: raise prices on general-purpose computing hardware (excluding phones, which are increasingly locked down) so that it becomes again inaccessible to consumers.
This article goes to show that you use AI to produce reasonable sound argument but under the hood the assumptions are still not very defensible.
The way I see it, is that they missed the boat on how powerful those models will become meanwhile other players are integrating deeply into their ecosystem. It might go down as the fall of the empire similar to Nokia, Kodak etc. It is still too early to tell.
If this article was true then Cerebras would be worth at least 2.5T and Nvidia half of what’s worth now. The market seems to be paying a premium for training chips rather than inference. Oh, and there’s also a bubble as you can see in Shiller’s PE ratio here https://www.multpl.com/shiller-pe
My comment won’t add anything to the discussion. As far as I know I have heard same idea * several times from George Hotz blog posts. I guess that’s what tiny corp wants to do.
The author seems unaware that products such as nVidia’s DGX Spark and AMD’s Strix Halo exist at a similar price point per gig of unified RAM, and I don’t see either company offering less unified memory as time goes on. I think he’s right about Apple doing a good job right now, but wrong about its competitors.
I like Apple and think there is an AI bubble, but this article is kind of dumb.
Apple is not doing anything magical with hardware that other companies, including Nvidia, cannot do for roughly the same price/performance. The M series is amazing, but there are comparable products out there now. If you want to say eventually demand for Nvidia will fade because demand for training frontier models will decrease with time, fair, but that doesn’t suddenly make Apple the king of AI.
Also, for new better AI models to happen someone will need to train those new better models and that isn’t going to happen on 5000 Mac minis duct taped together.
This is actually a stronger argument for Apple than the article makes.
The king of AI isn’t the shop with the biggest/bestest model, or whose machine has the bandwidth and unified memory.
It’s the company who has the trust of a massive number of users.
Apple was smart enough to give OpenAI the finger on deep integration, and focus on building out the part that everyone really wants.
Whoever provides the llm inference, do it with the privacy protection, and with access to the personal data you want AI to have access to, but without sacrificing the former. They’re literally the only company on earth in a position to do this.
We’ll see if they’re going to pull it off, very soon. If they have, they’ll have managed to negotiate and capitalize on this crazy hype cycle better than anyone, by far.
They could become the trusted gatekeeper of AI services, doing the one thing that none of the rest can ever do… and have done it without lighting trillions of dollars on fire fighting an unwinnable war.
Any interaction with the iOS 27 beta siri compared with ChatGPT Live quickly disproves this drivel. Also OP seems unaware of their bandwidth contstraints
On the deep partnership with Google for Gemini, it will be interesting to see how Google and Apple respond to GPT Live.
This does really feel like the first milestone for voice assistants that are equivalent to talking to a human. Not suggesting it's perfect - but the tech is only going to improve from here.
I wonder if the author will leave a comment justifying the tediously vanilla LLM-generated prose by stating English isn’t their native tongue and that the original truly was the fruits of their own rigorous thinking and writing efforts.
It would be nonsense of course, but it’s a tale that seems to work for a significant number of readers here.
While I generally agree that local models will gain popularity and Apple is generally well situated to capitalize on this, articles like this are not the way to convince anyone.
Plots with no axis labels or units displayed? In fact if you're literally just comparing two things, why not just write down the two numbers? Or better yet strengthen your argument by comparing more than just literally two options?
And the pictures. You seriously couldn't be bothered to just go find a picture of four Mac minis stacked on top of each other, and had to genai a picture of four vaguely-mac-mini-shaped blocks?
So the argument of this _very_ Claude written piece is that.. On apple devices you have GPUs with fast-enough unified memory and can link them together?
Okay, fair, but why would I get an Apple device when I can get a Linux device with some monster motherboard that has a dozen memory slots?
I submitted this ( I am not the author ) because I thought it was interesting take, I honestly didn't think it was AI, maybe AI polished, but not AI. Was surprised when it was flagged and I saw over on HN Removals . Ran through Claude and got this back:
"Overall, my honest read: this feels like a human-written opinion piece (possibly AI-assisted for polish/editing, which is extremely common now), not a fully AI-generated article. The specific personal history ("I've been saying this in group chats for months"), the running joke about being right before everyone else, and the reader engagement in the comments (with the author replying) all point to a real person with a strong personality behind it — though I can't rule out that Claude or another model helped tighten the prose."
It is interesting that everyone was so very certain it was AI written. Just read your comment about it being Claude based, now I am trying to figure out if Claude is trying to recuse itself in some way by giving me those results.
Still feel like the point it interesting and going to keep diving into it and I am surprised it was flagged. May have to up my reading on Removals more and more.
I don’t think that a generative model is a good judge to assess if a text was written by itself. This article has a super stong “written by AI” smell to it. Once you notice all the em dashes and “it isn’t X, it’s Y”patterns you can’t unsee it. That plus the sloppy images, make it really hard for me to finish reading.
This made me go down a rabbit hole.
Looks like Mac Minis had a 64 GB model (hard to get now) for roughly $1,600. You could get a 3 of these (192GB) & thanks to the Thunderbolt 5 connection, they out performed any single board AMD for a similar price that had more memory. If you tried to do so something similar with Nvidia GPU you're looking at a lot of watts, cooling & a much faster experience but only 48 GB of memory.
> only 48 GB of memory.
You can layer the LLMs across multiple GPUs, you can absolutely have 192gb of Nvidia dGPU memory with the right MOBO and CPU combo.
Because bandwidth between those sticks and the GPU isn't fast enough for fast inference.
Even with an integrated gpu? Surely it'll be faster that using thunderbolt cables
both AMD and Apple hit the unified memory goal. but its the bandwith thats bottlenecking their performance and onboarc memory is even worse.
Is Apple in a better position than Amd as this article seems to claim?
AMD's Strix Halo has a 256-bit memory bus. Apple's Max class chips have a 512-bit memory bus, Ultra class chips have a 1024-bit memory bus. Unless you want to go for datacenter hardware, AMD's offerings simply aren't in the same league in terms of memory bandwidth.
I see, interesting thank you
depends: if they can keep getting larger memory, yes
if the author wasn't too lazy - this article could've been way shorter.
but of course author chose to use a.i to write this.
& bury very important points which are 1. Chinese Memory Manufacturers 2. Apple Unified Memory + their own chips which are a result of their 'car' effort.
emdashes?
> Every historical technology wave ends the same way: the mainframe becomes the PC, the timeshare becomes the laptop, the cloud becomes the edge. The only people who ever argue otherwise are the people selling mainframes.
Perhaps AI serves also as an attempt to reverse course on this: raise prices on general-purpose computing hardware (excluding phones, which are increasingly locked down) so that it becomes again inaccessible to consumers.
This article goes to show that you use AI to produce reasonable sound argument but under the hood the assumptions are still not very defensible.
The way I see it, is that they missed the boat on how powerful those models will become meanwhile other players are integrating deeply into their ecosystem. It might go down as the fall of the empire similar to Nokia, Kodak etc. It is still too early to tell.
If this article was true then Cerebras would be worth at least 2.5T and Nvidia half of what’s worth now. The market seems to be paying a premium for training chips rather than inference. Oh, and there’s also a bubble as you can see in Shiller’s PE ratio here https://www.multpl.com/shiller-pe
> * The endgame is intelligence as a commodity.
My comment won’t add anything to the discussion. As far as I know I have heard same idea * several times from George Hotz blog posts. I guess that’s what tiny corp wants to do.
i like the tiny vision but it's yet to break out from its research project phase
The author seems unaware that products such as nVidia’s DGX Spark and AMD’s Strix Halo exist at a similar price point per gig of unified RAM, and I don’t see either company offering less unified memory as time goes on. I think he’s right about Apple doing a good job right now, but wrong about its competitors.
I like Apple and think there is an AI bubble, but this article is kind of dumb.
Apple is not doing anything magical with hardware that other companies, including Nvidia, cannot do for roughly the same price/performance. The M series is amazing, but there are comparable products out there now. If you want to say eventually demand for Nvidia will fade because demand for training frontier models will decrease with time, fair, but that doesn’t suddenly make Apple the king of AI.
Also, for new better AI models to happen someone will need to train those new better models and that isn’t going to happen on 5000 Mac minis duct taped together.
It is hard to be the king when it only dominates the market share in a few rich countries.
The king of AI is yet to ship a modern AI assistant.
This is actually a stronger argument for Apple than the article makes.
The king of AI isn’t the shop with the biggest/bestest model, or whose machine has the bandwidth and unified memory.
It’s the company who has the trust of a massive number of users.
Apple was smart enough to give OpenAI the finger on deep integration, and focus on building out the part that everyone really wants.
Whoever provides the llm inference, do it with the privacy protection, and with access to the personal data you want AI to have access to, but without sacrificing the former. They’re literally the only company on earth in a position to do this.
We’ll see if they’re going to pull it off, very soon. If they have, they’ll have managed to negotiate and capitalize on this crazy hype cycle better than anyone, by far.
They could become the trusted gatekeeper of AI services, doing the one thing that none of the rest can ever do… and have done it without lighting trillions of dollars on fire fighting an unwinnable war.
Has Apple built the part everybody wants? I'm a Mac used and I'm yet to see it. Instead I got transparent buttons I didn't ask for.
Any interaction with the iOS 27 beta siri compared with ChatGPT Live quickly disproves this drivel. Also OP seems unaware of their bandwidth contstraints
On the deep partnership with Google for Gemini, it will be interesting to see how Google and Apple respond to GPT Live.
This does really feel like the first milestone for voice assistants that are equivalent to talking to a human. Not suggesting it's perfect - but the tech is only going to improve from here.
What does the quality of Siri vs ChatGPT have to do with the argument made in the article?
Apple hires people who don’t know how to integrate existing models even well behind the state of the art
Unbearable AI slop, flagged.
I wonder if the author will leave a comment justifying the tediously vanilla LLM-generated prose by stating English isn’t their native tongue and that the original truly was the fruits of their own rigorous thinking and writing efforts.
It would be nonsense of course, but it’s a tale that seems to work for a significant number of readers here.
The retort to this slop runs on usb-c power and was not made by Apple