I think most people just assume it's magic, and are too awestruck by the hype to think critically.
Financially this feels similar to Uber's business plan in the 2010s; undercut the market with unsound pricing propped up by venture capital (PE was literally subsidising taxi fares; they admitted this and their intention to readjust, but no one seemed to care) then stop manipulating the market and allow fares to even out at (gasp) what it cost to get a cab before Uber.
The difference here is that the LLM market is human productivity; enormous subsidies are afforded to Anthropic, OpenAI etc. in the form of VC or compute credit, but eventually those debts will be called in, the free-to-use aspect will vanish because it's simply not profitable, and we'll be left with several premium products that only a few people will actually pay for, and even then that may not be enough to cover their costs. That's when the bubble will burst.
There’s the scenario where LLMs get more efficient in size, and to get 2026 SOTA performance you will be able to get it from consumer grade laptop.
Sure with a 1000B parameter you will get better performance but the average person will have it write some python script, not derive new physics equations.
So in a sense the demand for LLM intelligence with reach a plateau (arguably we are there today for avg person) so there will not be any subsidy required, because the avg person will not need the latest and greatest.
There’s not the same demand pattern for something like uber.
> There’s the scenario where LLMs get more efficient in size, and to get 2026 SOTA performance you will be able to get it from consumer grade laptop.
But isn't that bad for the AI companies, too? Because then people just run an ~2026 SOTA performance open source model on their laptop for free and not pay any subscription.
Regular folks will not pay Anthropic, but NSA, NASA or research labs might.
I’m not implying this will be a good time for AI companies. I am saying AI as a technology can provide value without it being controlled by only 3 companies.
In a hypothetical future with 2026 level LLMs on a (high end) consumer laptop, I still think that majority of buyers would prefer to pay 20 USD/month for a service. Just for the convenience and flexibility.
> In a hypothetical future with 2026 level LLMs on a (high end) consumer laptop, I still think that majority of buyers would prefer to pay 20 USD/month for a service. Just for the convenience and flexibility.
$20 a month is a lot of money, I don't think the "convenience and flexibility" you get would actually be worth it, unless you've 1) got money to burn, 2) lack the skills to install software, 3) the open source community totally fails to develop a reasonable installer. The LLM service would probably be akin to a scam preying on ignorance, like those companies that will rent you a water softener for like $100/month.
It is a lot compared to what? I believe that a LLM capable laptop will cost considerably more than something that is good-enough for non-LLM productivity tasks. At least within the next 5 years. Say that it would cost 600 USD more, that would buy 30 months of subscription. It is this kind of scenario I think many people will favor the subscription.
Anyone calling anything an anti-pattern without evidence always sounds to me like 'I don't like how you do this, but I need to find a more cerebral way of describing that so I don't sound like a child.'
I feel like you're giving certain entities too much credit there. Yes text is generated to do _something_, but it may not be to communicate in good-faith; it could be keyword-dense gibberish designed to attract unsuspecting search engine users for click revenue, or generate political misinformation disseminated to a network of independent-looking "news" websites, or pump certain areas with so much noise and nonsense information that those spaces cannot sustain any kind of meaningful human conversation.
The issue with generative 'AI' isn't that they generate text, it's that they can (and are) used to generate high-volume low-cost nonsense at a scale no human could ever achieve without them.
> Life’s too short to go through it hating others
Only when they don't deserve it. I have my doubts about Google, but I've no love for OpenAI.
> Plagiarism has a particular definition ... no-one is required to append a statement crediting every text he has ever read
Of course they aren't, because we rightly treat humans learning to communicate differently from training computer code to predict words in a sentence and pass it off as natural language with intent behind it. Musicians usually pay royalties to those whose songs they sample, but authors don't pay royalties to other authors whose work inspired them to construct their own stories maybe using similar concepts. There's a line there somewhere; falsely equating plagiarism and inspiration (or natural language learning in humans) misses the point.
ChromeOS is a good idea if all they do is surf; security risks are pretty low. iPhones are also fairly safe, but the older they get, the slower they'll get and the more you have to support them. I've had to explain to my dad that eventually the software on the phone will be so old that new apps can't go on it. He looks at me in confusion and aggression and asks why, as though I'm responsible for Apple's planned obsolescence.
My family was ultimately able to convince my grandmother to get rid of her computers altogether, when her dementia really kicked in. I think we were lucky as she never really got on with computers, and would tell anyone who'd listen how computers 'came in' to her office the year she retired (in the 90s) and so never needed to learn.
> I've had to explain to my dad that eventually the software on the phone will be so old that new apps can't go on it. He looks at me in confusion and aggression and asks why
I’m with dad on this one! I have so much perfectly good hardware that becomes increasingly useless because Apple stopped providing updates, and 3rd party software developers insist that I’m running the absolute latest OS. I’m at that point yet again with an iPhone 7 that works flawlessly but when I go to get new apps, “SORRY LOSER. You need bleeding edge iOS for your bank’s stupid app!”
I’ll channel your dad’s aggression if I ever meet a Schwab software engineer in person. I complain to every Apple employee I know, too, but they all look at me like I’m crazy for not just buying a new phone every year.
You fail to understand how software development and maintenance works.
OFC you need an updated system that has all known security holes fixed to run homebanking apps.
Also, as a dev I would only support one config and not a myriad of different devices and operating system versions (APIs). Livelong. For 3$ purchase price. On all devices. For the whole family.
And imho Apple devices are supported much longer than most Android ones...
An analogy is not a valid argument. Analogies are useful for illustrating a concept, but a waste of time when trying to support a claim. Also, your analogy does not seem analogous to me.
Sir this is a Wendy's. We actually can not serve you lobster. Our kitchen doesn't support cooking lobster so sorry you will have to go find somewhere else.
> You fail to understand how software development and maintenance works.
OK, dude. It's not like I've been in software for 25 years, 15 of them being on mobile, both on the OS side and on the third party app side. But, yea, I fail to understand.
On the 3p side, I've heard all the lame excuses, and they are almost all excuses rather than reasons. The big one is that supporting older versions blows up the test matrix (the number of device/OS/API level combinations that need to be validated). I can understand if you're a hobbyist, you might not be able to afford to buy the "myriad of different devices" and aren't staffed to test your app on each one. But if you are any kind of serious business, you signed up for this investment when you decided to write apps. Throwing a couple of older devices running older OSes onto that test plan should not blow your budget, and if it does, you probably shouldn't be writing that banking software to begin with. Also, almost all of your testing is automated, right? (Please say Yes). So it's not like you need to hire more human testers as your test matrix grows. If you are unable to support more than the bleeding edge OS, it makes me wonder what kind of fly-by-night developer shop you are.
The second excuse is about valuing developer convenience over users. "Oh, the new OSes contain cool new stuff that we want to take advantage of, and it's a total bummer to maintain the code paths that support 'legacy' devices." You already have the code that runs on the previous OS or API level, you're choosing to get rid of it, deliberately throwing users under the bus, so that you can clean code up or at least not have to maintain it. Bad tradeoff IMO.
There are a whole bunch of other little excuses for no longer targeting older systems, and most of them boil down to either cost, laziness, or a skill issue. None of them respect the end user.
I have a little more sympathy on the OS side. Sometimes a major step forward on the hardware (particularly the CPU architecture) might make it really tempting to cut off previous versions. I still think both major mobile OSes cut off old devices way too early. I have PCs from the early 2000s that can still run modern Linux distributions, so support for old hardware is usually technically possible, just inconvenient and costly. I'm not asking companies to support devices from 20 years ago, but they could.
Your "security holes" excuse is ridiculous: All major OS vendors already provide security updates to at least 5 previous major OS releases. There is no security reason for an app developer to support only the bleeding edge latest OS.
Part of it is also Apple. They don't hesitate to break things across OS versions, whether it's on iPhones or Macs. Unlike Windows which will run basically anything ever built for Windows. Also AutoLayout is bad at adapting to new screen sizes. The iPhone app I built in high school targeting a 3GS was more futureproof than a lot of newer apps cause I just used C macros to calculate UI sizes/positions.
End result, even a simple "fart button" app has probably broken several times.
> Throwing a couple of older devices running older OSes onto that test plan should not blow your budget, and if it does, you probably shouldn't be writing that banking software to begin with. Also, almost all of your testing is automated, right? (Please say Yes).
Yeah but a lot of apps will require the very latest iOS for no reason other than the dev happened to build it that way per Xcode defaults, and some apps will also require you to use the latest version of the app at all times.
The replies you're getting are a bit reminiscent of the "guns don't kill people, people kill people" defense of firearms - like, yes that's true, but the gun makes it a lot easier to do.
Sure, maybe? But if you were gonna stack rank death machines in order of death (in the US at least) and ban them, it'd go something like:
Drugs and alcohol first (or drugs first and alcohol second if you split them apart), then pistols second, cars, knives, blunt objects, and rifles.
We tried #1 already, it didn't really work at all. Some places try #2 (pistols) to varying degrees of success or failure. Then people skip 3, 4 (well except London doesn't skip 4), 5, and try #6.
And underlying that all is 50 years of stagnating real wages, which is probably the elephant in the room.
---
I'd posit that using an LLM to respond to a 10 page long ranting email is missing the real underlying problem. If the situation has devolved to the point where you have to send a 10 page rant, then there's bigger issues to begin with (to be clear, probably not with the ranter, but rather likely the fact that management is asleep at the wheel).
Is that even true? I feel like a lot of unhealthy foods are easy to eat with your hands, and a lot of healthy foods are hard to eat without a fork or a spoon
The story isn't about LLMs doing LLM stuff. It's about lawyers using LLMs as a shortcut for proper legal work, laboring under the delusion that it is entirely accurate, honest and 'intelligent', and the ramifications for the legal system.
Financially this feels similar to Uber's business plan in the 2010s; undercut the market with unsound pricing propped up by venture capital (PE was literally subsidising taxi fares; they admitted this and their intention to readjust, but no one seemed to care) then stop manipulating the market and allow fares to even out at (gasp) what it cost to get a cab before Uber.
The difference here is that the LLM market is human productivity; enormous subsidies are afforded to Anthropic, OpenAI etc. in the form of VC or compute credit, but eventually those debts will be called in, the free-to-use aspect will vanish because it's simply not profitable, and we'll be left with several premium products that only a few people will actually pay for, and even then that may not be enough to cover their costs. That's when the bubble will burst.