You can also link directly to a web installer so customers don’t even have to ever see the Store app. The Store provides you a web link for a tiny exe installer, like this:
https://developers.openai.com/codex/app/windows
I would be interested to find out a bit more about your experience with the MS App store. Do you get much traffic? Does it convert? What price range is your product? Is it a game or utlity?
> This is absolutely correct. Instead of maintaining any sort of ABI and API stability, Apple offloads a constant burden of maintenance updates across thousands of developers, just to keep existing apps from breaking every year with a new iOS version. This takes time which could be spent in more productive ways such as fixing bugs, adding features, or developing new apps. It seems like the wrong trade-off, since stability would offer huge, multiplicative benefits across the whole ecosystem. Apple does seem to want apps to die to mitigate the glut of shovelware in the app store, but there has to be a better way (human curation still seems like the only reliable approach for app surfacing and discovery.)
I keep trying to explain this to people but it's hard enough to describe the issue, even harder to get people to care, and an impossible battle to change Apple. I don't actually think they're doing this to kill old apps. I think it's a very cynical and calculated plan to require developers to actively maintain their applications, *thereby requiring the use of subscriptions as the only viable business model for developers.* That is Apple's primary revenue stream by far, and they're making far more money now that we have to subscribe to workout apps instead of buying them once and using them for years.
Apple has been dropping older subsystems and backwards compatibility layers long before app subscriptions were the default way people got paid for software. The 68k -> PPC transition happened in the mid 90's and 68k support was dropped entirely somewhere around OS 8 and the start of the iMac era. The Carbon framework might have been the most long running one, going from about 2000 to 2012 for deprecation and basically ending once the 64 bit transition happened around 2018. The PPC -> Intel transition, including the original Rosetta emulator was ~2005-2011. The app store itself only debuted in 2011.
I do agree that Apple does this in part to force developers to either stay active and maintain their apps or stop shipping for the platform, but I personally posit that the move of more and more apps to subscription models is simply due to how many more apps are connected and user expectations for update timeliness (and the devaluing of updates both by increased popularity of "free as in beer" open source apps and also the distribution of no-cost OS updates by Apple. People expect more for free and expect it as soon as someone notices a problem. I think the idea of not only waiting a year or more to have new features or some bugs fixed and then on top of that having to shell out more money for that is just not something people are as wiling to do. So subscription models become necessary to fund the continuous work that goes into keeping up with all the new trends. Apple's dropping of old libraries and frameworks is part of that churn, but it's only one part in a sea of other pressures driving the subscription model.
>thereby requiring the use of subscriptions as the only viable business model for developers. That is Apple's primary revenue stream by far, and they're making far more money now that we have to subscribe to workout apps instead of buying them once and using them for years.
Apple earns almost twice as much revenue from selling iPhones ($210B per year) than it does from "Services" ($109B), and "Services" includes far more than just App store commissions.
Thanks for the correction. I should have said profit, and I should clarify further: gross profit for the App Store and global iPhone production is similar ($84B and $88B), but the App Store and other services have nearly double the profit margin and is growing much faster than physical products.
I interpret their argument differently. We know that bullying leads to harmful outcomes. We know that punishment reduces the frequency of undesirable behaviour. So we know that this policy will lead to an aggregate reduction in harm. The question is whether it could lead to some degree of harm to the bully. In the absence of compelling evidence of that, the policy itself seems merited.
For the record, bullying is a complex problem to solve, and no nation or policy or tactic has the silver bullet.
There are famous stories of prisoners in Japanese prisons during the World Wars being healthier than their captors because they ate refined, white rice instead of the brown rice for the prisoners.
Costo makes almost no money on food sale, and almost all of its profit from the membership fee. It's required for their business model, which is VERY friendly for its employees. This is an example of capitalism done right.
This is such a misconception/oversimplification that drives me nuts and it is so common. It isn’t at all supported by their filings.
In past years their profits match the approximate amount of revenue from their memberships (which, I suspect is the origin of this idea), although recently their profit exceeds their membership revenue by a fair margin (total membership revenue was less than half of operating income for 2025). Membership is not what causes them to be profitable since the “cost” of the membership is operating the rest of the business. You can’t separate retail sales from membership sales since they mutually require each other in the business model. If they got rid of their retail sales (which are, in fact, more profitable than membership sales on their own) then the membership would be worthless. The reverse is not true.
It's true that in recent years profits have exceeded membership revenue, but that still comprises 67% of total profit. It is still a requirement for their business model to operate, unless you're arguing that they would be comfortable with significant losses in the lean years, which I would contest.
That’s a deceptive 67%. You divided the pretax membership revenue by their post tax profit. If we use the same math we see that they get almost 70% of their profit from sales at the same time.
It falls to less than 50% when you properly compare it to pretax profits.
It also doesn’t factor in the 2% rebate that a huge number of members receive.
Costco’s total net sales for fiscal 2025 were roughly $268 billion. With executive members driving ~73.6% of that, and a 2% rebate on qualified purchases, a rough back-of-envelope estimate would place total rebates issued in the range of $3–4 billion, but Costco does not publicly break this out as a discrete figure in its financial statements.
In any case, I’m not saying it isn’t a significant source of revenue, I’m saying that “they only make money/they make most of their money from memberships” is a deceptive framing that is oft repeated. It makes it sound like they are selling things at cost when they aren’t. Their margins are in line with the grocery industry as a whole. Their pricing is an effect of low opex and volume. They make most of their profit from selling things. If they didn’t charge for membership they wouldn’t lose money.
There is a trust component for sure, but a business requires assessing the value of time against revenue. I can say for our org that using an off the shelf solution like Clerk saves us time and money and we believe the risk is very small relative to the savings. Maybe the cost for you is not large right now, but when you've got 20 enterprise customers all asking for specific OIDC integrations configured with Private Link, custom domains, and private clusters, an auth solution starts looking mighty fine.
Isn't it widely speculated that these are distilled from current frontier models? Distillation is far less compute intensive than primary training. That said, if distillation produces something almost as good for a fraction of the cost, Jensen's point may stand.
You can't really distill a model without access to the internal weights. You could train on chat logs, but that's absolutely not the same thing, it doesn't even come close to comprehensively "extracting" the model's capabilities. And everyone does that in the industry anyway ever since ChatGPT was first released, some versions of Opus even claimed to be DeepSeek if you prompted them in Chinese.
Calling it distillation does however make normies go along with it when they inevitably add all the Chinese labs to the entities list to pad Dario and Sam’s pockets.
Weights are not required for distillation. I'm not sure how you came to that belief. Distillation is training a student model to minimic a teacher model output.
Anthropic, for example, posted a 2026 disclosure (https://www.anthropic.com/news/detecting-and-preventing-dist...) which singles out DeepSeek's distillation activity. They detected over 16M actions over 24,000 fraudulent accounts. That's just what they detected.
We are also technically a statistical process generating one part of a word at a time when we speak. Our neurons form the same kind of vectorised connections LLMs do. We are the product of repeated experiences - the same way training works.
Our brains are more advanced, and we may not experience the world the same way, but I think we have clearly created rudimentary digital consciousness.
My experience as well. This is even worse than just having a mediocre model, because I can work around that. The inconsistency means it produces different outputs for the same prompt, and I can't rely on that as a business tool.
reply