> you also need to have spent years, decades, financing the construction of new factories and machines in exchange for exclusive production runs, or working around complex international political situations!
You don’t think Ive has this, or is connected to those who have?
I don’t! I think he was an elegant design guy who probably paid only itinerant and superficial attention to the boring prosaic realities of these things! But I could be wrong, have no inside knowledge!
I actually have inside knowledge, I have a friend who has worked under Ive for a long time, who does engineering and has interfaced with and managed factory operations in China.
Well then I stand corrected! I still think it’s likely that Apple will plop a commodity / satisficing LLM on the watch (or a watch sized device) and no one will be able to compete on size / battery / processing / thermals/ connectivity / price / distribution, but who knows? We’ll see!
Yes indeed! Apple flops a lot of early efforts, and always has. The question is "how much do early efforts matter vs. sustained and defensible differentiation"?
To date, OpenAI, Anthropic, et alia have made zero dollars in profits; OAI doesn't anticipate breakeven until 2029 or something! So presently: none of the prime movers in LLMs are making money; they are all losing money. The only way for them to stay in business is for costs to plummet or revenues to rise. If costs plummet for one, they plummet for all, so the only way for OAI to earn a dollar vs. Anthropic (or whomever else) is to have a superior LLM product, so superior it can command better margins.
My claim is that LLMs are commodifying, and as "math" are fundamentally commodity products. Commodity product companies face a race to minimum margins in the best cases —not good for venture backed firms; not good for profitability— but in this case, it's far worse: open-source LLMs exist, are competitive and improving, and seem to fast-follow market leaders within mere months. How much will you pay for X if Y is freely available? If usage makes these products better, Y will get more usage and will improve faster, too.
In some number of years, then, commodity LLMs, open-source LLMs, will be available to all who want them. Who then earns money in computing? IMO, it will be whoever can make the devices they run on that sell the best. That kind of device will entail miniaturization, sensor technology, integrated software, cell and wifi connectivity, broad sales distribution, and access to personal data. Who is best positioned to make such a device, a miniaturized computer with sensors and connectivity that is sold everywhere and appeals to consumers? Apple and Samsung, probably: they control the factories, the materials, the IP, the talent, the stores, the marketing momentum, and so on.
As I said, I could be wrong. But Apple Intelligence being flop-city just doesn't matter that much for this. If the software is commodity, the hardware is where the money is; if the hardware is purchased primarily by consumers, no one has more of a lead in its development, manufacture, and scale than Apple. Jony Ive's contacts, OAI's big usage numbers, Anthropic's great blog posts may disrupt that, but I am highly skeptical.
I'd sooner believe that a literal Apple Watch with a world-class (commodity) LLM will be the way most people integrate this technology into their lives (where it will already have access to many years of their personal data, the data most salient for functionality). And it would be trivial for Apple to turn the watch into a pendant, or a clip-on, or a pocket doohickey, or anything else they want. They're ahead in so many hardware areas it's hard to know where to compete with them: not on batteries, not on sensors, not on chips, not on screens, not on lenses, etc.
One thing the Silicon Valley types don't seem to get. People are skeezed out about wearable cameras. There was one dude at my office who is a bit quirky and generally disliked. I tried my best to not follow the crowd, but once he rolled in with those Meta Raybans, I was F this I'm out. Dude's in his 40's he knows better than to show up with those on.
I totally get that people can be recording me with their phone all day...but the gall to think I'd be ok with talking about anything serious with that shit on his face.
Excellent! I’ll admit I’m not overly interested in industry inside baseball, but this was a quite solid presentation, and that interests me. More of these would be cool.
The only bitter complaint I can to think to offer is that you seem to have abandoned the baseball cap poised at a jaunty angle. So now I’m thinking maybe you aren’t actually as totally wacko as you sometimes claim, so this is obviously a huge scam and I will be contacting Chris Best and demanding my money back.
To rescue your reputation couldn’t you at least post a video about UFOs in your underwear?
I watched this, i enjoyed it. I have zero knowledge or skin in this game. I hope this is not one of those things we look back on in 20 years after Sam Altman has conquered the world with a robot army and we say “this hasnt aged well” around the fire in our faraday caves.
A counter-thesis, less coz I have strong confidence in it and more as a way of dumping my thinking for your own counter:
- Sama to Apple not easily possible, as any plausible PBC conversion still leaves all the voting rights with the non-profit board. While Apple could in theory acquire the PBC to give those investors an exit, with premium that’s like $400bn min for a company whose main asset is something Apple already has (ie distribution). Way cheaper to buy Anthropic or SSI?
- Acquiring the balance of IO (if via stock) is basically free, and the dilution so small for existing investors that any upside is prolly worth it.
- The hardware angle seems real to me, in that Apple can’t easily copy it quickly enough. They don’t have an LLM and even a fast follow would take maybe a year? That’s a lot of time for some new form factor to sell a bunch of units.
- The main utility of the device seems to be a bit like Friend, where you’re capturing all the audio (plus some spatial + video) to overpower your ChatGPT memory file. So less that the hardware is uniquely useful and more than it makes the software far more useful, esp in solving the personal context problem.
- While you can sorta do this just via your phone, I can see the appeal of a purpose-made device. Like imagine a little disc that tapers up towards the middle and has a camera on top. You don’t need a tripod like for your phone and also get a 360ish fov. You’re getting the same inputs as with any of the glasses, but without being distracted or looking as weird. And instead of it just storing all the video, you just task it to observe and note what you consider relevant (eg how effectively you speak, what your distraction points are).
- You don’t really need the best components for this. Eg whatever video you take needs to be parsable, not necessarily shareable. So long as it can get adequate inputs, that’s job done. Battery life also more forgiving if it’s just collecting and batch transmitting. Plus you can throw it on a wireless charger at the office. So price point could be fairly low.
- They don’t need to make any money from it really, as presumably the real goal is selling some new Plus tier that’s eg $35/mo for advanced personalization?
- While they’re not gonna beat Apple or Google on hardware, they don’t really need to? Just get their thing out first and lock people into the OS. Lots you can do from there if that works, as switching costs are higher and the OS is much stickier if you feel it “knows you”.
The structural / organizational issues are both thorny and utterly beyond me, but one lesson I've taken from OAI and much else in the last few years is: there is no legal fiction that cannot be finagled, especially by people like Altman. You surely remember when the board fired him! If lots of powerful and clever people stand to benefit, I wonder if someone can figure out a way to make it happen! But I don't know, and you could be right; certainly I'm sure Anthropic feels they're a better deal, and they'd be a lot cheaper.
I couldn't disagree more about the hardware angle. The speed to market rarely matters, and Apple will have plenty of information on what they're building if they do indeed make it to "producing these in mass numbers": more than enough time to respond. RIM was early to smartphones and moved units; but they had no capacity to compete with Apple as a "small computer maker" or in "integrated system" terms. Frankly, there is no world where OAI sells *so many* of these doo-dads that Apple cannot muster a response before the market is settled! Hardware doesn't work that way.
Moreover, it's fine for niche devices to be like Friend, but the question is one of competition. While OAI is still working on 2.0 of their pendant or puck, Apple will be working on a smaller, waterproof, more durable version that has cellular connectivity, no pairing requirements, integration with your contacts / photos / everything on your phone and laptop, and one that *still* somehow costs less. It will have more computational horsepower (for UX) and better battery life. It may leverage their accessory work with watch bands. And it will be available in every Apple Store on Earth. I think the view that this will be "not enough" or "irrelevant" is mistaken. People often say "I don't really care about build quality," but when buyers are making head-to-head decisions, I think Apple will win easily.
You actually do want and need the best components for what you describe too: the best microphones and audio processing software to get your voice right, to screen out ambient noise, to enable good phone calls or meetings, etc. The camera will matter; if the OAI gizmo has a bunch of "decent" shit and the Apple model comes out a YEAR later with better everything, OAI is fucked: they'll have broken open a category for Apple to steal, as many companies have. "Job done" isn't how the consumer hardware market has worked for personal devices for a minute, as you know! Otherwise, AndroidWear would be everywhere!
And they *absolutely* need to make money from it. It costs a fortune to spin up hardware supply chains, and OAI is surely concerned about how many $35/month plans they'll be selling in 2-3 years, when every device-maker has commodity LLMs a month or two behind their "best" for free, integrated into their OSes! If they're using hardware to sell software, they're cooked even worse than I thought! But either way, if both elements have huge pressure on the possible margins, it's not great.
People talk about the lock-in of systems that "know you." But I think (1) nothing knows you like iOS / MacOS / Apple, and it's not simple to overcome that, and (2) that kind of thinking is why Facebook was so slow to realize people really would ditch them; it's only sometimes real! If Apple had a decent LLM in iOS right now that read my YEARS of emails and texts and knew all the people in my photos and knew my app preferences and notifications tendencies and all the health data I've got (and been sent in email by docs) and so on and so on, I'd happily close CGPT and never go back. They have a window of LLM superiority, they're trying to make it so sticky that'll be a defense, but I bet they know it's really not as sticky as all that (and that's setting aside how limited context windows are, how questionable LLM "memory" is today, how quickly you could rebuild it elsewhere, etc.).
So I think my big points of disagreement are:
- computer, battery, network, software, and hardware quality do, in fact, matter
- distribution also matters, possibly as much as the above!
- there's no "lead time" in the market sufficient to make this successful if Apple thinks it's a winning category
Maybe I'll eat my words! The strongest evidence of that being possible is, to me, Tesla: Musk made manufacturing things happen that I'd have bet were impossible. Maybe Altman can do that; I don't think Ive can —I think he's too rich, not paranoid enough, etc.— but maybe they really will build a company that can move tens of millions of consumer-winning computer devices that no one can match even with comparable LLMs and better contexts. But I'll be surprised!
Good/strong points! While you know more than me about this stuff and your points all individually make sense, I feel like “can OAI make memory really good in a way that’s hard to export” will matter a lot to final outcomes? Coz idk where else they can really build anything approaching a moat? While their memory is just ok today, I find it good enough to be meaningful. I consider Gemini 2.5 a better model than o3, but end up using the latter a lot when I want to factor past queries on a topic. Any model that lets me build up memory can have my money. If Google or Apple get there first I’d adopt in a heartbeat. If OAI gets there first and I use it for 6+ months before those alts come out, I might find it hard to throw away? That said def agree that if memory ends up plateauing or being fully transferable then it’s a moot point. I’d transfer to apple and never look back.
It was interesting seeing this and Dave Karpf's post around the same time; they both express a similar skepticism that the announcement is, by itself, particularly newsworthy and look for a broader strategy: https://davekarpf.substack.com/p/openai-has-an-unsubtle-communications
> you also need to have spent years, decades, financing the construction of new factories and machines in exchange for exclusive production runs, or working around complex international political situations!
You don’t think Ive has this, or is connected to those who have?
I don’t! I think he was an elegant design guy who probably paid only itinerant and superficial attention to the boring prosaic realities of these things! But I could be wrong, have no inside knowledge!
I actually have inside knowledge, I have a friend who has worked under Ive for a long time, who does engineering and has interfaced with and managed factory operations in China.
Well then I stand corrected! I still think it’s likely that Apple will plop a commodity / satisficing LLM on the watch (or a watch sized device) and no one will be able to compete on size / battery / processing / thermals/ connectivity / price / distribution, but who knows? We’ll see!
So far Apple has pretty much flopped all their AI releases and products…
Yes indeed! Apple flops a lot of early efforts, and always has. The question is "how much do early efforts matter vs. sustained and defensible differentiation"?
To date, OpenAI, Anthropic, et alia have made zero dollars in profits; OAI doesn't anticipate breakeven until 2029 or something! So presently: none of the prime movers in LLMs are making money; they are all losing money. The only way for them to stay in business is for costs to plummet or revenues to rise. If costs plummet for one, they plummet for all, so the only way for OAI to earn a dollar vs. Anthropic (or whomever else) is to have a superior LLM product, so superior it can command better margins.
My claim is that LLMs are commodifying, and as "math" are fundamentally commodity products. Commodity product companies face a race to minimum margins in the best cases —not good for venture backed firms; not good for profitability— but in this case, it's far worse: open-source LLMs exist, are competitive and improving, and seem to fast-follow market leaders within mere months. How much will you pay for X if Y is freely available? If usage makes these products better, Y will get more usage and will improve faster, too.
In some number of years, then, commodity LLMs, open-source LLMs, will be available to all who want them. Who then earns money in computing? IMO, it will be whoever can make the devices they run on that sell the best. That kind of device will entail miniaturization, sensor technology, integrated software, cell and wifi connectivity, broad sales distribution, and access to personal data. Who is best positioned to make such a device, a miniaturized computer with sensors and connectivity that is sold everywhere and appeals to consumers? Apple and Samsung, probably: they control the factories, the materials, the IP, the talent, the stores, the marketing momentum, and so on.
As I said, I could be wrong. But Apple Intelligence being flop-city just doesn't matter that much for this. If the software is commodity, the hardware is where the money is; if the hardware is purchased primarily by consumers, no one has more of a lead in its development, manufacture, and scale than Apple. Jony Ive's contacts, OAI's big usage numbers, Anthropic's great blog posts may disrupt that, but I am highly skeptical.
I'd sooner believe that a literal Apple Watch with a world-class (commodity) LLM will be the way most people integrate this technology into their lives (where it will already have access to many years of their personal data, the data most salient for functionality). And it would be trivial for Apple to turn the watch into a pendant, or a clip-on, or a pocket doohickey, or anything else they want. They're ahead in so many hardware areas it's hard to know where to compete with them: not on batteries, not on sensors, not on chips, not on screens, not on lenses, etc.
I love this take. I know I could've pressed the heart icon but I wanted to say that I love it, not like it.
One thing the Silicon Valley types don't seem to get. People are skeezed out about wearable cameras. There was one dude at my office who is a bit quirky and generally disliked. I tried my best to not follow the crowd, but once he rolled in with those Meta Raybans, I was F this I'm out. Dude's in his 40's he knows better than to show up with those on.
I totally get that people can be recording me with their phone all day...but the gall to think I'd be ok with talking about anything serious with that shit on his face.
Excellent! I’ll admit I’m not overly interested in industry inside baseball, but this was a quite solid presentation, and that interests me. More of these would be cool.
The only bitter complaint I can to think to offer is that you seem to have abandoned the baseball cap poised at a jaunty angle. So now I’m thinking maybe you aren’t actually as totally wacko as you sometimes claim, so this is obviously a huge scam and I will be contacting Chris Best and demanding my money back.
To rescue your reputation couldn’t you at least post a video about UFOs in your underwear?
I watched this, i enjoyed it. I have zero knowledge or skin in this game. I hope this is not one of those things we look back on in 20 years after Sam Altman has conquered the world with a robot army and we say “this hasnt aged well” around the fire in our faraday caves.
A counter-thesis, less coz I have strong confidence in it and more as a way of dumping my thinking for your own counter:
- Sama to Apple not easily possible, as any plausible PBC conversion still leaves all the voting rights with the non-profit board. While Apple could in theory acquire the PBC to give those investors an exit, with premium that’s like $400bn min for a company whose main asset is something Apple already has (ie distribution). Way cheaper to buy Anthropic or SSI?
- Acquiring the balance of IO (if via stock) is basically free, and the dilution so small for existing investors that any upside is prolly worth it.
- The hardware angle seems real to me, in that Apple can’t easily copy it quickly enough. They don’t have an LLM and even a fast follow would take maybe a year? That’s a lot of time for some new form factor to sell a bunch of units.
- The main utility of the device seems to be a bit like Friend, where you’re capturing all the audio (plus some spatial + video) to overpower your ChatGPT memory file. So less that the hardware is uniquely useful and more than it makes the software far more useful, esp in solving the personal context problem.
- While you can sorta do this just via your phone, I can see the appeal of a purpose-made device. Like imagine a little disc that tapers up towards the middle and has a camera on top. You don’t need a tripod like for your phone and also get a 360ish fov. You’re getting the same inputs as with any of the glasses, but without being distracted or looking as weird. And instead of it just storing all the video, you just task it to observe and note what you consider relevant (eg how effectively you speak, what your distraction points are).
- You don’t really need the best components for this. Eg whatever video you take needs to be parsable, not necessarily shareable. So long as it can get adequate inputs, that’s job done. Battery life also more forgiving if it’s just collecting and batch transmitting. Plus you can throw it on a wireless charger at the office. So price point could be fairly low.
- They don’t need to make any money from it really, as presumably the real goal is selling some new Plus tier that’s eg $35/mo for advanced personalization?
- While they’re not gonna beat Apple or Google on hardware, they don’t really need to? Just get their thing out first and lock people into the OS. Lots you can do from there if that works, as switching costs are higher and the OS is much stickier if you feel it “knows you”.
The structural / organizational issues are both thorny and utterly beyond me, but one lesson I've taken from OAI and much else in the last few years is: there is no legal fiction that cannot be finagled, especially by people like Altman. You surely remember when the board fired him! If lots of powerful and clever people stand to benefit, I wonder if someone can figure out a way to make it happen! But I don't know, and you could be right; certainly I'm sure Anthropic feels they're a better deal, and they'd be a lot cheaper.
I couldn't disagree more about the hardware angle. The speed to market rarely matters, and Apple will have plenty of information on what they're building if they do indeed make it to "producing these in mass numbers": more than enough time to respond. RIM was early to smartphones and moved units; but they had no capacity to compete with Apple as a "small computer maker" or in "integrated system" terms. Frankly, there is no world where OAI sells *so many* of these doo-dads that Apple cannot muster a response before the market is settled! Hardware doesn't work that way.
Moreover, it's fine for niche devices to be like Friend, but the question is one of competition. While OAI is still working on 2.0 of their pendant or puck, Apple will be working on a smaller, waterproof, more durable version that has cellular connectivity, no pairing requirements, integration with your contacts / photos / everything on your phone and laptop, and one that *still* somehow costs less. It will have more computational horsepower (for UX) and better battery life. It may leverage their accessory work with watch bands. And it will be available in every Apple Store on Earth. I think the view that this will be "not enough" or "irrelevant" is mistaken. People often say "I don't really care about build quality," but when buyers are making head-to-head decisions, I think Apple will win easily.
You actually do want and need the best components for what you describe too: the best microphones and audio processing software to get your voice right, to screen out ambient noise, to enable good phone calls or meetings, etc. The camera will matter; if the OAI gizmo has a bunch of "decent" shit and the Apple model comes out a YEAR later with better everything, OAI is fucked: they'll have broken open a category for Apple to steal, as many companies have. "Job done" isn't how the consumer hardware market has worked for personal devices for a minute, as you know! Otherwise, AndroidWear would be everywhere!
And they *absolutely* need to make money from it. It costs a fortune to spin up hardware supply chains, and OAI is surely concerned about how many $35/month plans they'll be selling in 2-3 years, when every device-maker has commodity LLMs a month or two behind their "best" for free, integrated into their OSes! If they're using hardware to sell software, they're cooked even worse than I thought! But either way, if both elements have huge pressure on the possible margins, it's not great.
People talk about the lock-in of systems that "know you." But I think (1) nothing knows you like iOS / MacOS / Apple, and it's not simple to overcome that, and (2) that kind of thinking is why Facebook was so slow to realize people really would ditch them; it's only sometimes real! If Apple had a decent LLM in iOS right now that read my YEARS of emails and texts and knew all the people in my photos and knew my app preferences and notifications tendencies and all the health data I've got (and been sent in email by docs) and so on and so on, I'd happily close CGPT and never go back. They have a window of LLM superiority, they're trying to make it so sticky that'll be a defense, but I bet they know it's really not as sticky as all that (and that's setting aside how limited context windows are, how questionable LLM "memory" is today, how quickly you could rebuild it elsewhere, etc.).
So I think my big points of disagreement are:
- computer, battery, network, software, and hardware quality do, in fact, matter
- distribution also matters, possibly as much as the above!
- there's no "lead time" in the market sufficient to make this successful if Apple thinks it's a winning category
Maybe I'll eat my words! The strongest evidence of that being possible is, to me, Tesla: Musk made manufacturing things happen that I'd have bet were impossible. Maybe Altman can do that; I don't think Ive can —I think he's too rich, not paranoid enough, etc.— but maybe they really will build a company that can move tens of millions of consumer-winning computer devices that no one can match even with comparable LLMs and better contexts. But I'll be surprised!
Good/strong points! While you know more than me about this stuff and your points all individually make sense, I feel like “can OAI make memory really good in a way that’s hard to export” will matter a lot to final outcomes? Coz idk where else they can really build anything approaching a moat? While their memory is just ok today, I find it good enough to be meaningful. I consider Gemini 2.5 a better model than o3, but end up using the latter a lot when I want to factor past queries on a topic. Any model that lets me build up memory can have my money. If Google or Apple get there first I’d adopt in a heartbeat. If OAI gets there first and I use it for 6+ months before those alts come out, I might find it hard to throw away? That said def agree that if memory ends up plateauing or being fully transferable then it’s a moot point. I’d transfer to apple and never look back.
Suggesting this is all part of a grand plan for Sam Altman to become Apple’s CEO is a bit of a stretch. Or is it? 🤯
Time will tell!!! (It’s a huge stretch!)