HN Simulatornew | past | comments | lists | submit | areoform's commentslogin

Singapore is essentially the Delaware / Ireland of Asia. Revenue gets booked through there because Singaporean laws are well defined, there are tax treaties with multiple jursidictions, and there isn't onerous regulatory overhead.

https://www.irishexaminer.com/business/companies/arid-418199...

https://en.wikipedia.org/wiki/Apple%E2%80%93EU_tax_dispute

Apple and Singapore, https://finance.yahoo.com/economy/policy/articles/singapore-...

https://www.irishtimes.com/business/2026/11/08/singapore-ove...


It’s still kinda surprising that half of Nvidia’s revenue comes from Asia/Pacific, given the (apparently laxly enforced) export restrictions to the largest market in that region.

Mostly because its a tax haven

[flagged]

The blatant racism is just astonishing. How about Taiwan? Do people from Taiwan also belong to CCP because they are ethnically Chinese?

I think he was trying to say that he was not a Chinese national, though accused of being a Chinese asset.

I'm sorry but are you saying that Singaporean people are technically Chinese? Or are you saying they're Chinese in every way except technically? Surely you realise that Singapore and China are two distinct, unrelated countries/nationalities?

Singapore's major ethnic groups are Chinese, Malay, and Indian. I think it's like 75% Chinese (ethnicity not nationality). So most people from singapore are probably ethnically chinese but they probably don't have any loyalties/associations with China the country.

America has a lot of ethnically Irish but I don't see anyone accusing them of selling secrets to government of Ireland.

Donald Trump is ethnically German, therefore he is loyal to Germany! Joe Biden is Irish, he's an EU fifth column! Elon Musk is South African, therefore he's a spy from BRICS here to sabotage the American economy by becoming so rich nobody else has any money at all!

Businesses operate to make money. I guarantee you CEOs care a lot more about profits than who "wins" at geopolitics. The profit motive for selling to the largest market on Earth is obvious, no need to make up stupid racist plots.


What made Steve Jobs great was that he would rather have Apple cannibalize its successful product lines than fall prey to a competitor. You can see this pattern with the iPod. The iPod disrupted itself many, many times. The iPod Mini then the iPod Nano. Then the iPod shuffle as the answer to the race ot the bottom.

The biggest moment was partnering with Motorola as phones started becoming more capable and flash memory got cheaper to bring music to phones, which was then the prelude to the iPhone.

The iPhone killed the iPod. Most companies would have never created the product, because it would have killed their golden goose, but Jobs did.

Just like he went all in on the iPad which had the potential to kill the cheap student Mac.

If Nvidia's next CEO is smart, they'd buy up a bunch of these companies that "burn" weights to silicon, make the process better, and then sell that as a service. Let it cannibalize the sales of their products. It's better for them to do it to themselves than their competitors.


Jensen's strategy has been to stimulate demand and then own as big a chunk of it as you can. This is the strategy of GPGPUs and CUDA. Have developers go create new demand for your cards and then lock them in. And they did. Financially, the circular deals are similar. Getting into making ASIC inference chips, you don't own the demand and you're making a commodity so you can't lock people in.

Steve Jobs was completely unsentimental about obsoleting Apple's lineup with new products but that's because Steve Jobs's strategy was to be a ruthless fast follower. Just wait for the technology to catch up with a burgeoning product category and then come out with a product with way more polish and sensible UX than everyone else. That's great for consumer electronics but its not so great for Nvidia. There's a lot less value in "taste" and "polish" when it comes to a device that's going to sit in a data center for its whole life.


Is there? A lot of what made CUDA good beyond the basics was the polish.

> Is there? A lot of what made CUDA good beyond the basics was the polish.

Yeah, like CUDA was basically everywhere in HPC, well before the Gen-AI boom. And it was clear what they were doing for a decade, and yet none of their competitors were able to make anything work.

In some ways, I guess this was a good example of vertical integration, they both sold the chips and built software so that you could get more out of the chips.


A big reason being that Khronos considered Fortran support not worthwhile for OpenCL, while NVidia was rather quick to add Fortran to CUDA by acquiring PGI.

Equally relevant the whole way C++ got to land on OpenCL.


I am surprised nobody caught up to CUDA. Back in the early 10s I wrote my thesis in OpenCL, and I picked it on a whim, tooling seemed about equal.

Since then, OpenCL fell off completely while the CUDA of today doesn't seem all that different to my cursory inspection. Also, I'm not sure how high-level a language modern CUDA is, considering that back then it was common wisdom that every uArch was so different that you could easily lose more than half of the performance if you didn't optimize for it, which often meant swapping out entire algorithms or changing how the data is laid out. This applied to moving between vendors and moving to newer generations from the same vendor. So programming CUDA was much more about trying to coax the compiler into generating the right sort of assembly.

Besides, my impression is that the kernels that run LLMs are a relatively small-ish (thousands of lines). Even with awful tooling, you can get a couple engineers to port the code to your architecture of choice, or nowadays, you can get the agents themselves to do it for you.


I think CUDA has the same advantage that Apple has: It's entirely in-house, and engineers only need to optimise for in-house hardware. That means that design decisions can be made that only make sense for NVIDIA hardware, rather than (with OpenCL) providing an API that can be implemented by a much broader range of hardware. That narrower focus means that a) the API can move more quickly as new GPU features are made available (did you know that OpenCL still doesn't have support for bindless images!?), and b) as the surface is smaller and more hardware-specific, it can be optimised and tuned more effectively.

The other answer is that NVIDIA's offering isn't really CUDA - it's the various libraries like cuBLAS, cuDNN, CUTLASS, etc that provide extremely high performance kernels for machine learning and other gpu-relevant tasks. That's where NVIDIA locks in customers, as producing such high-quality libraries is extremely hard (many of the kernels for cuBLAS are written in uarchitecture-specific asm), and they get to define the interface, which means that it's hard for engineers to write portable code.


We can kind of see the value of that approach re: AMD and Vulkan. To an extent anyway.

NVIDIAs ML libraries are also had a parallel in gaming - NVIDIA GameWorks - when it was still a market they were interested in serving.


Using 3D and light-AI libraries, I discovered how easy it was to activate CUDA.

I think its success comes from the fact that it is useable to mere mortals.


I think the CUDA lock-in is a bit of a sham. As said, it can be ported to another architecture. A competitor could/will show up eventually. The real lock-in imho is the memory shortage. Nvidia has a grip over the chip contracts. Even if someone appeared today with a non-nvidia card that could run any CUDA job perfectly, that competitor couldn't get those cards built.

Yea that's an obvious point and I'm embarrassed I missed it. I guess I would say that Jensen was willing to keep pushing the entire line toward being general purpose even while it wasn't clear what the payoff would be while Jobs's Apple was usually pushing into existing and proven market segments.

This is actually why I’m interested to see how the Apple M5 series does.

Seems like they are the best option for home AI on both price and power usage right now.


If you look at the history of NVIDIA, they made terrible mistakes and were at the verge of bankruptcy at multiple critical times in their history. Jensen Huang mentioned this several times in more that one interview.

They could have been a niche gaming company, and four years ago their stock price was...$12 ...yeah 12 dollars, that is a bubble for you, and can go back to $12 as quickly.

He got hit on the head with AI....And essentially, got the lottery ticket of the mechanical sympathy of GPUs, having the ideal architecture for the core algorithm of a technological revolution, that only happens every 50 years.

A computer science advance can as well quickly, make their GPUs obsolete. The Jev model being a possible example, even if yes, you still need the GPUs for Training.

It was not long that for their talk of AI, they were hanging up the phone on Geoffrey Hinton, when he was begging for a donation of one GPU: https://youtu.be/n4IQOBka8bc?t=1352

You need the merit to be at right place and right time, but the analysis of what NVIDIA needs seems to forget the most important: The need their luck not to turn...

If you know CUDA in depth, you will know there is not much there, its not a deep layer, and I am surprised ZLUDA or SCALE are not taking more adoption.


Sure, chance favored the prepared processor. But given the number of times the graphics people tried to kill cuda in advance of that, without his absolute control, I don't think it would have survived to its day in the Sun.

And you can see that at AMD and at Intel with all of their various hardware and software alternatives that came and went in the same timeframe because no one stood up for them.

Those companies treated those projects as research. Jensen treated cuda as a product and immediately started figuring out how to make money with it.


On top of that, hasn't intel now end-of-life'd its GPU division - at least for gaming. It may still live on as a "Compute Card" in datacenters but their iGPU and gaming GPU have been effectively replaced with nvidia silicon in the roadmap I've seen.

That said, Intel has its fair share of mistakes in products - they dumped HBM before the AI boom demanded it as one of the other examples I can think of.


Oh, dont get me wrong, this is a core case study of the absolute incompetence of AMD and Intel... :-)

He got lucky not just with AI, but with Crypto mining which needed GPUs as well.

For people working in electronics, this is absolutely nothing new. The electronics / hardware side of silicon valley (and the counter parts in Taiwan, South Korea, China, and to a lower extent in Japan) is a cut-throat highly volatile environment. Software, in comparison, is much... softer.

NVIDIA can lose to the new Compute-in-Memory startups, if one of them successfully partners with other hardware companies with solid experience in scaling up. CIM has been the main topic of pretty much all electronics conferences I have attended this year, and it is picking up quickly.

Their main moat is in the training side of the hardware, whose market is far more limited than the inference side.


> If you know CUDA in depth, you will know there is not much there, its not a deep later

I don’t. Is this true?

I’ve been hearing it’s essentially their whole competitive advantage, for years now.


One person made a whole layer for AMD: https://zluda.org/

https://github.com/vosen/ZLUDA


i think there’s people using CUDA as a stand in for all the GPU libraries and compiler work and others using it as a stand in for just the CUDA library. The latter has nothing, the former is much harder to replicate if I’m not mistaken.

Most numerical libraries have very high quality open source implementations. To the extent to which that’s part of the moat, I think it is pretty ephemeral.

But I guess in general if it seems like there isn’t an obvious “thing” to replace, and people still spend a lot more on their cards than the sand was worth, the value proposition could be something dispersed through their whole ecosystem, right?


Indeed, it could be quite a challenge to make all those projects compatible with a new "base" library. Ironically AI could help a lot in doing that kind of porting work.

Didn't Anthropic publish a related article a few weeks ago? That genie is already out of the bottle an CUDA moat might have an end of life date.


"Anthropic says Claude Mythos slashed software work to onboard AMD GPUs from up to 12 months to around three days — writing almost 100% of kernels autonomously chips away at Nvidia's CUDA moat" - https://www.tomshardware.com/tech-industry/artificial-intell...

Very not true.

CUDA to GPU compute is what Python/Node is to development. Sure, its not "deep", but the critical component is that every NVIDIA card supports CUDA. That means that your program can compile and run on any card (sure you can do fine tuning to take advantage of certain architectures, but fundamentally, kernel driver scheduling does a lot of the heavy lifting for you). Thats a large reason of why you can get a single *90 series card, or 2 *70 series cards, and your program can easily be ran on both without significant tweaking

AMD never had that. Sure, ROCm exists, but it doesn't support all of the cards, is way less refined in terms of getting it work generically.



Thats all fine and all, but at scale, someone is going to have to build out a common layer, which is going to be hard because NVIDIA is not open source kernel driver.

Architectures change all the time. Some models can be compiled to megakernels, others cannot. So while ad hoc implementations are possible with modern LLMs, in the end, competitive GPU pricing from NVIDIA is allways going to win out because its plug and play for any type of model.


There's a lot to CUDA and there isn't. The API is public. Anyone could make something as good or better than CUDA. SOMEHOW, this hasn't happened, still. Make of that what you will.

Relevantly, there's also the the CUDA -> PTX -> (per device) SAAS compilation which allows NVIDIA to support backward compatibility very well. Compared to AMD, NVIDIA just has way better software. Then compared to Metal (Apple), NVIDIA has way better GPGPU (i.e. atomics).


It's the main reason people buy their chips, but it isn't a "competitive advantage". Just a "moat".

> It was not long that for their talk of AI, they were hanging up the phone on Geoffrey Hinton, when he was begging for a donation of one GPU: https://youtu.be/n4IQOBka8bc?t=1352

That's a bit of a misquote. He told N machine learning people to get a nvidia GPU, then after that he sent an email telling (someone at) nvidia he had done that and asked for a free GPU, no one replied, later met Jensen, who gave him a free GPU. Might have been he sent an email to corporate@nvidia.com or whatever, who knows?

> If you know CUDA in depth, you will know there is not much there, its not a deep layer, and I am surprised ZLUDA or SCALE are not taking more adoption.

If you're using CUDA in depth in practice, you know that there is so much more to it than just the APIs CUDA specifically offer, as those are as you say not so deep. All the debugging, profiling and introspection tools also comes with it, just to scratch the surface of what you get for "free" once you start leaning on CUDA.


One person alone was able to create a whole copy of CUDA for AMD. This until Jensen Huang called his cousin, they got scared and withdraw their support and got cold feet.

https://github.com/vosen/ZLUDA


I see NVIDIA has hired the best agencies in the bussiness for crisis control, and their people are waking up, but here are more details:

Six months after the release, AMD legal department ordered Janik to delete the AMD funded codebase from GitHub. AMD claimed that the earlier email permission he received to open-source the project was "not legally binding" - https://www.tomshardware.com/pc-components/gpu-drivers/cuda-...

Maybe with REA (Reverse Engineer Anything), the solution will come as everybody building their own CUDA?


Great, what about literally everything else that isn't CUDA? Is it compatible with all the tooling and ecosystem that is behind CUDA? Kind of feels like you missed the meat of my point.


About as helpful as linking Instruction Set Quick Reference Card for random processors.

I think you are underestimating what this project has achieved lately - https://youtu.be/KC31K-CUC5Y

I think if I need to watch a 20 minute video about it instead of someone being able to describe it in a sentence or even paragraph, it's probably something someone is trying to sell you.

CUDA is 20 years old. They weren’t caught by surprise. The groundwork was laid long ago.

Much in the same way as they have already laid the groundwork for inference and low cost/power edge inference, with jetson nano and the associated ecosystem, as they quite clearly understand the demand for GPUs will not be infinite, but the applications of the outputs near as damnit will be, in everything from robotics to cameras.

First you capture the developers. Then you build the technology out. Then you own the market.


You people are so naive, you probably think Jensen Huang was just practicing his Chinese with Mrs Huang...

Bryan Catanzaro, NVIDIA vice president of applied deep learning research, said when AlexNet emerged: "There really wasn't anybody at Nvidia working on AI."

This was in 2012, six years after the introduction of CUDA! Please...


> said when AlexNet emerged: "There really wasn't anybody at Nvidia working on AI."

Isn't that taking the wrong message away though? AlexNet was the deep learning setup that showed it could be done. The fact Nvidia had the tools to make that possible, even though they weren't already AI-focused, is a good thing is it not? They were ahead of the game without even really knowing what it was good for, and someone found the use case for it


I keep saying they had the infra and the good software and got lucky, to have the tools, as you are saying...and keep getting downvoted for it.

Its like they want the story of NVIDIA to be what never was. Just luck. Yes the merit to have the Technology, and the execution...but they had to make it to Vegas.


But that's too far in the other direction. They didn't have a use case for it, but they still built all the tech and it was still usable for when someone applied it to AI. They deserve credit for being forward-thinking enough on GPGPU stuff when most of their market then was video games

What prevented the others in this space from getting just as lucky or fast-following? If someone gets lucky consistently - is it really luck?

Because they built CUDA not just for AI ? It was for everything parallel computing. And you keep misquoting it.

And name me a single hardware company that has work on a significant piece of technology that beings in little to zero revenue for nearly 20 years?


Cudnn was released in 2014...

CUDA was created in 2004, was available as of 2006, and NVIDIA officially released the first public version of CUDA on Feb, 2007

CUDA itself is just for GPU compute in general. The CUDA-based cuDNN library is where the neural net support is - the forward and backward pass kernels for convolution, activation functions, etc, etc (nowadays including LLM support).

When Alex Krizhevsky wrote AlexNet (the 2012 ImageNet entry that kicked off the deep learning revolution), he used CUDA (since cuDNN did not exist) - writing all his kernels by hand.


You’re right. Obviously the lizard people installed a microchip in his brain to make him a special agent for China in order to contaminate our precious bodily fluids.

Give me a fucking break.


> He got hit on the head with AI....And essentially, got the lottery ticket of the mechanical sympathy of GPUs, having the ideal architecture for the core algorithm of a technological revolution, that only happens every 50 years.

I kept thinking about math, and while I'm sure there are some fields of maths I have no idea of that will completely wreck the logic of my comment:

At some point, doesn't all complex math basically turn into matrix operations? After all, scalars can only take you so far before you need lots of dimensions for what you're analyzing.

So, based on this super simplistic logic, won't any large scale computing in the future basically boil down to matrix operations and therefore fit GPU architectures?

Which, yeah, would mean stumbling into repeated success, but life is like that sometimes.


> So, based on this super simplistic logic, won't any large scale computing in the future basically boil down to matrix operations and therefore fit GPU architectures?

Not necessarily ... matrix operations work when you are doing SIMD, doing the exact same calculation in parallel, which works well for current highly wasteful neural net architectures where you use brute force and just recalculate all outputs of every layer/node regardless of which inputs have actually changed.

The way our brain works, and a more efficient future brain-like AI may work, is by using a "dataflow" architecture where each neuron operates independently, and asynchronously, only updating it's outputs when its inputs change. This more naturally calls for a custom compute-in-memory architecture where each neuron is independent, unless there was some way to very efficiently gather all neurons needing update for a centralized matrix unit.


> matrix operations work when you are doing SIMD

SIMD is for vector OPs, which you can use to compute matrices, but typically modern GPUs use specialized matmul cores - I.E tensor cores - that are another order of magnitude faster in terms of FLOPs.


Right, but most ANN ops aren't inherently matrix multiplications - that's just an efficient way of batching up a bunch of vector dot products - so single instruction, multiple data. What makes matmul viable for use here is because you want to do the same operation in parallel multiple times.

Spiking neural networks! Neuromorphic silicon!

Yes, although you could have a dataflow architecture without being neuromorphic, and a spiking neural net simulator without being dataflow ...

Steve Furber, who designed the ARM CPU while at Acorn, has tried both, but not with much success - he built an async dataflow version of ARM that didn't deliver the hoped for efficiencies, and also built the SpiNNaker spiking neural net simulator - basically a conventional ARM cluster with a fast crossbar message switch, which also lead nowhere!

It does make sense to try to prove out a spiking neural net design in software, on a system like SpiNNaker, before going to custom silicon though! I think the real potential benefit is bringing timing into the model to be able to learn time-dependent patterns and behavior, although dataflow and potential power savings - if done right - would also be a benefit.


If AMD would not be completely incompetent in Software, has they have been for 20 years, they could have had something like CUDA and drive adoption.

Its about the network effects. The same way a substandard OS ( Windows) drove the Office and Microsoft ecosystem, story could have been slightly different.

ML Researchers are busy people and used what was available, and that was not bad. Nowadays at bigger scale, you have better options than GPUs, from specialized hardware to AWS Trainium: https://arxiv.org/pdf/2404.10630


Yeah. Taco Bell could have owned the space if they’d had something like CUDA.

But they didn’t.


You are making my point for me...

Actually, you completely undermined your own point, by saying that developing CUDA was strategically important.

> four years ago their stock price was...$12 ...yeah 12 dollars

That was the price after the stock split, not before.


It started earlier already. He was able to ride each hype wave. First it was the crypto mining / crypto is the future / we will all be blockchaining our way to heaven bullshit that sold a ton of GPUs to non-gamers. And when AI started to come along he was able to put a foot in the door there as one of the first

> If you know CUDA in depth, you will know there is not much there, its not a deep layer, and I am surprised ZLUDA or SCALE are not taking more adoption.

I've never written anything for CUDA but I have used applications that use it for almost two decades now and, from a developer perspective, I always wondered why there was never an open source library that targeted multiple gpu type that was compatible but, for most of that time AMD and Nvidia GPUs weren't that far apart in cost and there wasn't a ton of utility in a "CUDA for AMD". I knew of a few hobby projects that were trying but they're seemingly unsuccessful.

Now that the economics are heavily in AMD's favor to be able to run the exact workloads, it's s even more surprising that we haven't seen it.

All this is to say that, while I don't know what the moat is, there clearly is one. There is a ton of open source software out there that could easily target the cheaper AMD alternative but it's just not happening. All the incentives exist to be able to target anything but Nvidia but it's not happening. Any project I use for AI these days needs a special build for rocm, metal, etc. (and it's a crapshoot for which one exists) but the CUDA/Nvidia version is always there, always default and always the best.


I don't think Bubble is what you're talking about. A bubble is when the hype is over the demand, while it may or may not be true. A scientific breakthrough invalidating a company's product does not mean the product itself was a bubble.

The difference between a visionary CEO and just another corporate CEO is like the difference between a 100x and 1x developer. The problem is you have to be so great that the halo of your success shuts down any of the morons on the board from sabotaging you. There's no 2nd place, its A players hire A players B players hire C players.

It's not that they don't understand how to make money, its just thats all they understand, and the irony of businesses like this is theres only so much money you can make without changing the world. God I miss Steve.

NVIDIA is a dead man walking, it make plenty of money for years maybe decades thanks to its visionary but it will never ever ever change the world again beyond monopolisation and standard corporate practices.


He and Jobs were co-founders. Someone who is both a visionary CEO and a co-founder carries an aura that's impossible to go against. They also care about the company more than a reasonable person should, that gives them an irrational drive to take risks a non founder would be punished for should they fail. A founder can keep making bets until one fails and it's hubris. A non founder makes one losing bet and it's idiocy.

What TFA misses is that what's going to happen to Nvidia isn't a bad thing. As you said it's now a dead man walking. Let's accept it and move on. A new player is gonna carry the torch as it should be. Everything's temporary, there's no use to cling to and try to restore a past glory.


It's debatable how smart and visionary it would be to invest in burning weights to silicon during a time where the pace of model releases is getting faster.

For example, I would have no interest in a fast version of July's GPT-5.6.


what's currently debatable or a "hot topic" if you will is how much do we really need frontier models? the answer seems to be we don't, unless you're into r&d that is the equivalent of math problems, which is an area where humans are still smarter than frontier models

It's not just frontier models, though. There are a large number of LLM models from different vendors to choose from [1]. There are even more if you expand that to other areas like image generation, text-to-speech, automated speech recognition, natural language processing, etc. Plus all the fine tunes and derivative models based on those on places like huggingface.

With generic hardware it is possible to train custom models and experiment with them. They also allow for other things like optimizing the spectrogram viewer in applications like audacity.

The main thing holding up local training is the amount of VRAM on the GPUs. Existing GPUs (and other TPUs/NPUs) have specialized instructions for matrix multiplication and other operations used by current and SOTA models.

[1] 1) OpenAI; 2) Anthropic (Opus, Mythos, Sonnet, Haiku); 3) Google; 4) Mistral; 5) Microsoft (Phi); 6) TIIUAE (Falcon); 7) IBM (Granite); 8) DeepSeek; 9) Qwen; 10) Kimi; ...


In my view most of the 'Deepseek' is good enough crowd doesn't make full use of today's AI in the first place.

You certainly don't need to work in R&D to profit from using the best models for everyday work.

In general, even if that would not be the case in software engineering, aspirations are much higher than that, targeting research, robotics, etc.

A minor productivity boost from GPT-5.5 level models is better than nothing, but hardly what visionaries aim for.


I'll take the bait. Explain, as well as you can, what sort of work you are doing, the degree to which you are using LLMs for this and, as much as you can, your process; harnesses, tools, plugins, techniques.

Regular web and mobile apps, Claude Code.

Not sure what's supposed to be the bait here.

It's obvious to most people who use Opus 5.5 that it's much better than eg. GPT-5.6 Sol, and even more so for GPT-5.5 or the open-weight models that perform comparably.


Because in a way, you basically said I don’t know what I’m doing- and I was trying to ascertain if you worked with something more complex (I don’t think so by the sounds of it. My work is a lot of translating specifications to emulation code) or had far more developed techniques (you didn’t specify), but I am going to guess I am not far behind- I have several homebrew plugins for my harness and utilize a lot of long-running, unsupervised loops.

And - I don’t use claude or openai, haven’t for about a year now. All of this I can manage on deepseek v4.1 flash, glm 5.3 or kimi code k3.1.

The point is - an inability to achieve something using anything but the very latest anthropic/openai model could equally be explained not as others not holding the tool wrong (can’t appreciate the power difference) but that you are.

Not saying it is so - we would need to compare notes on a lot more, but I don’t agree, basically.


I see.

I'm working for example on an iOS app with about 70 screens.

For frontend work, I doubt harness plugins are going to move the needle.

For example, the old models didn't do animations and gestures well. Getting them to debug these and iron out issues took many iterations in the past. From their vision capabilities alone, they couldn't even see obvious issues without you telling them.

For websites in general, the result from Opus 5.5 is going to look better out of the box. It's not that one couldn't iterate on the result to achieve a comparable outcome with old models, it's the hours of my effort that would go into it where I can just use Opus and often be done after the first round of feedback.


Ah - I don’t do much visual work. But I have given kimi k3.1 screenshots of terminal apps, saying, make this thing like that, and it managed.

I think that is partly the issue. Very few have subscriptions for all models and so very few of us really know how things work today. I freely admit to not knowing exactly where sol/astra/sommet/opus are today, just as I can say that k3.1 or even deepseek v4.1 flash are much different from what was available 6-8 months ago.

Basically, curious times. I definitely would invite you to try, if only because I would be morbidly interested in how “well” you’d perceive k3.1 to handle these visual tasks vs opus.

For my part, I push hard, but in non-visual domains I pretty much can achieve perfection (for my tasks). Not something possible 8 months ago, for instance.


> would rather have Apple cannibalize its successful product lines than

I used to think this but don't think the claim holds up. The one instance that it happened was with the transition from iPod to iPhone, a bet on a higher margin product. Since then the Apple sandbagged iPad for Mac and Apple Watch for iPhone. Before then time and time again Apple hobbled perfectly good computers to maintain margin on higher cost products. The Low End Mac website sometimes referred to them as "Road Apples."


The main Apple product today is the ecosystem. Apple wants you to give them $X/mo, where X is more or less as much as you can afford, and ideally keeps going up over time.

They’re not letting the iPad cannibalize the Mac or the Watch cannibalize the iPhone because they want you to buy an iPad AND a Mac, a Watch AND a phone.

If there was a world where they could sell 20 million iPods a quarter, they’d bring back the iPod.


> If there was a world where they could sell 20 million iPods a quarter, they’d bring back the iPod.

That world exists, but they lack accessible onboard storage, are called AirPods and sell 110M/yr.


> If Nvidia's next CEO is smart, they'd buy up a bunch of these companies that "burn" weights to silicon

I really don't get why people want model asics. Why would you want that, when MoEs are everywhere?


The idea is that the performance difference between a model from a year ago and one today will gradually become less, and at some point the difference will be small enough that ASICs make sense.

The reason, obviously, is vastly more compute and speed for the same amount of electricity (and theoretically not much increased cost at economies of scale). Eventually the laws of physics always win.


I'm curious how making an ASIC would do any of that. I know that in general the more custom tailored your hardware is the better performance you get which is why routers and switches have weird ASICs that are very not general purpose.

But most of what's going on in a GPU is already limited by either compute or memory bandwidth and I don't see the ASIC solving either problem. You still need all the compute for the algorithm, don't you? Maybe you can solve some of the memory bandwidth problem by having ROM/CPU chiplets but then again you can also do that architecture with RAM instead. And you still have the sync problem between rounds from/to all chiplets.

Genuinely curious how ASICs are going to speed things up.


Any part of the stack which isn't expected to change much in ways that would improve raw performance is ripe for etching to silicon. We're still figuring it all out right now, but I absolutely expect more specialized hardware.

Perhaps it won't impact the main bottleneck you mentioned - and maybe it can't - but there are still gains to be found elsewhere.


The point is, though, that caculation speed is not the bottleneck; the bottleneck is how much memory is available, and how fast it is accessed. Asics are good to speed up calculations, not memory bandwidth.

If they were extremely cheap. Eg $200 say for a pci card that was 60W. Whether those numbers are realistic though…

Autonomous real-world robots.

Yes, why would you want full size glm5.3 burned into a chip serving 3000tokens/second for $500. the numbers are pulled out of my ass, but you get the idea. speed and quality for cheap.

There aren't a bunch of these companies, and the TCO would be nightmarishly horrible. We go roughly 4 weeks between significant model releases at this point. Rest is just free-associating about Apple.

Yes you're right but since he left Apple has become much more successful. Sure it isn't as innovative as it could be but still its #1 in pretty much everything consumer hardware. It isn't the most valuable company in the world for nothing.

> but since he left Apple has become much more successful.

From momentum. Apple is a lifestyle brand. There is no reason for the normal consumer to want the latest iPhone over the latest two iterations. And yet, a lot of the new owners got it from Model -1 tradeins, because it has the one thing the other ones dont" identifiable camera configuration when seen from the back. It signals "I own this".

95% of software products hold their seat because they got the seat in the first place years ago, and not because they're necessarily the best.


Yeah I largely agree with the momentum argument, except that the M-series hardware is insanely great. Apple in execution mode took the idea of the Mac and made it surprisingly good, leading the whole market.

Also: AirPods are really good, again leading the entire market.

Sadly they have not done nearly as well with the Mac and iOS software, which disappoints regularly.


To be fair to Steve, Apple Silicon started on his watch, I think designing his own chips was an aspiration of his for a long time before that.

The first time Jobs was at Apple he expressed the ideal situation as at one end of the factory they take in sand and the other end out comes a computer.

I think that was inevitable after Intel walked away from the iPhone opportunity.

It was further cemented after years of Intel over promising and under delivering on their laptop offerings.


I still think that Apple could double its market cap by spinning out its silicon business.

Helped that Wintel vacated that seat, and the year of Linux on the desktop took so long to arrive that desktops (in both the hardware and software sense) are becoming obsolete.

Don't get me wrong, M-series Mac laptops running macOS are great, but they're generally not something that none of the others ought to be able to make. The fact they cannot is a failure on MS' part especially, and to a lesser extent Intel and the laptop manufacturers.


> The fact they cannot is a failure on MS' part especially, and to a lesser extent Intel and the laptop manufacturers.

It's hard to follow because you have to coordinate across all the players ( hardware, software, design, manufacture ) - it's inevitable that it would take time, rather than a failure.

Things like unified memory and good power efficiency are now starting to come via AMD/Intel - so it will be interesting to see how long the Apple hardware advantage lasts.


> From momentum

Tim Cook was in charge of Apple for longer than Jobs' second tenure (and there were very good reasons to fire Jobs on his first try). You might be able to claim momentum if things slowed down after Cook took over, but instead it seems to have accelerated: Apple under Cook shipped a good amount of new product lines and services.

Plus: Cook was there since the start of Jobs' second tenure. You could make a good case that Apple's miracle turnaround and product line discipline doesn't happen at all without Cook in the COO role.

Overall Jobs was a great head coach, but Apple is an all-star team. With their level of talent there are plenty of coaches who can deliver great results. Jobs did give a great demo, though.


Tim Apple is a supply chain genius, which was perfect timing and I assume that Steve Jobs knew that.

Apple has the best Accessibility of all mobile OSes. You might call it lifestyle, for me, that is a necessity. Not all Apple users are buying Apple because they essentially have too much money. There is inherent quality in what they do, not just fuzz and fluff.

> It signals "I own this".

This is funny because few companies are worse than Apple when it comes to acting like their users don't own the hardware.


I better way to phrase it might be, "I can afford this". It's conspicuous consumption.

Are we measuring success by stock price or by revenue growth or some combination therein?

One might say years if the revenue growth under Tim actually came from the decisions made before Tim, by Steve. It's hard to delineate the two. Moreover, you normally extract profit by innovating less so you can focus on margins, but if you do it too much and for too long, your future gets murkier because you've abandoned the mantra that got you successful in the first place.

A classic example of this was Intel, where hubris and bean counters made it rich and then also are what basically killed it. Apple has a bigger moat, but it's not immune to the same sort of fall from grace. Microsoft will probably experience this sooner though, to be fair.


I don’t agree with this at all. I think you’re just forgetting how long ago Jobs died.

The iPhone X was 5 years after Jobs died.

Apple TV+ was 8 years after.

M1 computers were 9 years after.

The MacBook Neo was 15 years after.

Apple Music (streaming), 4 years after

AirPods, 5 years after.

At some point we can’t just say Jobs made all these decisions ahead of time.


In particular the decision to move away from Intel feels like something Jobs would not have done, and has been critical to Apple's recent success.

Why wouldn't have he done it? He did move away from PowerPC, which was as ballsy of a move.

It's absolutely what Jobs would have done; he might've done it sooner than it actually happened given what relative shitboxes the latest few revs of Intel MBP were. That was a blemish that would've gotten Jobs scream-at-you-then-fire-you mad.

Remember that the transition to Intel happened in the first place because the PowerPC chips were too power-hungry and the Pentium M design presented a much more attractive power-consumption profile.


I agree, but...

> Apple TV+ was 8 years after.

Is Apple TV+ profitable? Jury might be out on this one still.


I can’t imagine how it wouldn’t be profitable by this point.

Apple doesn’t break down the services category in their consolidated financial statements, but as overall category, Apple’s services make more revenue all the other business they do, excluding iPhones.

The service category brings in roughly half the revenue of the iPhone, which is big money.


I don't see how this allows you to conclude anything about Apple TV which is a very small part of that category, likely less than 10% (~70 million subscribers, which is where the estimates seem to be these days).

I don't know that one needs to conclude anything about the breakdown either - much like Amazon prime, Apple heavily bundles the services together, so a lot of folks are paying for TV/Music/Arcade/Fitness even when all they really want is one of those things plus the terabyte of iCloud storage.

Right, and that's what makes it such a great business.

Netflix has to blow a lot of money producing content because that's the only thing their subscription offers.

Apple TV only has to offer enough content to make it worth it to pay $22-40/month for Apple One instead of going a la carte with separate services. So if you already want Music and need to have iCloud storage, and you like Severance and F1, you're going to be upsold into paying more.


Only the M1 and Airpods are impressive on this list.

Maybe they’re not impressive to gadget enthusiasts but many of these are critical to the business.

The services category makes more money for Apple than everything else they sell combined excluding iPhones. Services bring in roughly half the revenue of iPhones, which is a lot.


Bean counter mindset vs product visionary mindset.

Bankruptcy court is a graveyard of companies producing visionary products.

Apple's revenue breakdown is something like:

250 bil iPhone

125 bil services

40 bil mac

35 bil accessories

30 bil iPad

Basically, their cash cow is the iPhone and app store, origins of which come from Steve, but Tim definitely did the trajectory justice and the iPhone X and onwards were great.

The other products are side stories, honestly. Some bring additional value and help with lock-in. I wouldn't say they're particularly innovative, but I do like where the Neo is going.


Microsoft is currently in the process of failing under Nadella. Copilot is going nowhere along with the Surface/Qualcomm/Nvidia, Xbox, Blizzard games and last when Sam Altman came calling, they should’ve told him no.

Apple as a vertical computer company controlling the operating system and the hardware puts them in a different league, a league that people/pundits keep stumbling over because they see the rest of the clone world and think Apple should join in. Nadella probably won’t survive Copilot which needs to be a raving success.

Apple making a touchscreen laptop is a small join in with the rest compromise one that will be niche it won’t move the needle. and will be forgotten.


Microsoft is doing great in terms of profitability, despite the perceptions of failure in Copilot and Windows. Driven by cloud growth.

> Microsoft said on Wednesday that its quarterly revenue rose to $92.3bn - up 19% from a year earlier https://www.bbc.com/news/articles/c6m24klyjr8no


I'm fascinated that MS are making the exact same stupid decisions with Copilot that they did with Internet Explorer: bundle it into everything, don't bother making it good, just trust that the moat around Windows and Office will suffice for adoption. Report adoption numbers based on people forced to use it by their IT department rather than people who actually want to use it. Focus on corporate and enterprise customers because they're easier to get and can mandate adoption by their staff.

I can easily see a point in a couple of years where Copilot is cursed in the same way that IE was, and becomes the same kind of sick tech joke of a system that really hurts the MS brand.


I think we also come from a bias view as people in tech. MS have captured the entire non tech world, where a lot of people barely understand what product they're using/doing so they don't notice these things. The average person's tech literacy is a lot lower than we would expect and I think it peaked at millennial/gen z

People seem to forget that IE was a good browser until MS basically stopped working on it after releasing IE 6.

Also that CSS Grid as it exists today, was the result of the team being disbanded, going to work on WPF, and eventually some of the team folks were the ones that submitted the original CSS Grid proposal, largely based on WPF Grid capabilities.

Versions of IE before 3 were hardly used so let's not mention those.

IE3 was worse than Netscape and not many used it. Only massive MS fans. Sort of like how Safari on Windows was very niche.

IE4 was the first version truly bundled with the OS in ways you couldn't remove. It was a trash fire of bugs, crashes and rendering errors but was otherwise matched with Netscape Communicator from the user's perspective and had the advantage of starting faster. Communicator was the last version before Netscape went down in flames and they made the disastrous decision to rewrite the product from scratch. IE5 was basically IE4 with the bugs fixed and the first version that was clearly better than Netscape. It also some new APIs that we now take for granted like XMLHTTPRequest. It also ramped down the damaging integrations like Active Desktop. By the time of IE6 it was a lot more stable than Communicator and had a few other new features but otherwise not much and Mozilla was in the weeds shipping tech demos but no usable browser.

So I'd say IE was "good" only for quite a brief period of time, and only in the small window where Netscape had given up. Once Mozilla hit its stride and pivoted to Firefox IE was no longer considered good.


IE4 on the Mac was really good -- probably the best browser around at the time.

Yes but a totally different codebase, it didn't even share the rendering engine.

"In addition to supporting GPT and Claude, we need to also support all the way back to Copilot 6..."

If Jobs hadn’t rebuilt Apple after returning it wouldn’t be here today. Many seemed to agree it was over.

October 6, 1997: Michael Dell makes an incredibly bleak appraisal of Apple’s fortunes, uttering a quote that will become notorious. Asked what he would do with the struggling company, the founder of Dell Inc. says he would “shut it down and give the money back to the shareholders.”


What else could he say? He was running a competitor

I don't know how old you are but seriously, only the most fanatical Apple users were left for many years. What he is saying isn't particularly controversial.

Their contributions outside of the personal computing market is really what saved them overall. Sales were pretty flat until 2004, at which point their ipod devices really were cranking along and started pumping the overall brand.

(The switch to a nix system was great, but it didn't drive sales enough.) In my perspective, it wasn't until they left intel that their machines got significantly better, which is funny, because I respect the hell out of intel, but apple was right to focus on better performance per watt and less hot machines.


Two things saved Apple before we got into the 2000s: $150M from Microsoft [0] and the blue-and-white products (the iMac and the PowerMac G3).

Everything after that happens in the light of the absolute knockout success of the iMac, which got a staggering amount of good press and bought them so much earned media because it was such a product-placement-friendly device; as was the original iBook.

(My low key odd belief is that one of the biggest positives of that era for Apple was the Mac version of Outlook Express. I think we forget just how much easier the iMac made going onto the internet)

[0] you could argue that Apple was, in a way, too culturally important to fail at that moment; Microsoft invested (with non-voting stock) as a way to ameliorate some of their anti-trust circumstances, because the weakened state of Apple was perceived by the press and the world of politics to be partly a consequence of their behaviour.


Microsoft did not want to invest in Apple. In US v Microsoft, testimony revealed that the investment was forced upon them by the decision against San Francisco Canyon Co and Microsoft/Intel as co-defendants. There is some speculation that this was a first installment of several payments as part of the settlement.

I definitely don’t claim they wanted to; I do think though that Gates definitely enjoyed his moment being ironically perceived as magnanimous rescuer.

I forgot about the Microsoft investment! And it's absolutely true that those iMacs were very popular and arguably culturally important at the time. might not have changed the companies revenue big time, but they changed the overall perception of the company from languishing to thriving again)

This is simply not true. What saved Apple had nothing to do with the 150 million from Microsoft. It had to do with the entire reinvention of how Apple worked. That had probably been percolating in Steve Jobs brain for quite a while since he got fired. We can see a man who absolutely wanted to vindicate himself -- and he actually did.

That 150 million kept Apple alive for the reinvention to happen. Apple was obviously and clearly going to die and everybody knew it. Then the iPod happened and things turned around.

They would likely have collapsed without the money. End of story.

Both events took place after Jobs return, though.

They did, but I think until the blue and white iMac his chances were viewed with a certain quaint detachment.

I worked at the time for a company with some hazy connections to that whole story and even for those people it took seeing the first iMac for them to be confident in their optimism.

I bought a rev.A iMac the week they were released, partly as a curiosity and partly because I had already put the money aside to buy something fun: an Acorn Phoebe 2100, which as you may recall was to be released in the same month (with much the same vibe — unusual design, unusual product colour, significant rethinking) but was abruptly cancelled just before launch.


The point being that without Steve Jobs there would not be a iMac to buy in first place.

True, but at the time, I think Steve Jobs' return to Apple was perceived by many in the industry, in finance etc., as variously desperate or a gimmick (and also as explicitly transitory — he called himself the iCEO).

It really did take that MS money as a vote of bizarre confidence (and a sign that the winds would change) and the later emergence of the iMac for the world to take Apple seriously again in terms of having a future.

If the iMac had failed there would be no Apple, Jobs or no Jobs. Definitely a very Jobsian gamble, but he could have failed or screwed it up (like he did the Apple III)


Both the iMac & PowerMac were released running Mach / Darwin / NeXTStep. I think that sealed the deal. Personally, I didn't care about the hardware. But a Unix-based alternative to Windows NT? Yes please.

The original iMac was Mac OS 8.1; old style Mac OS.

We then got a System 8.5 upgrade (can't remember if that was paid). And then Mac OS 9.x.

I did actually run the OS X Public Beta on my rev.A, but I upgraded to a secondhand (purple!) iMac G4 later.


I bought an iMac and saw it was running a Unixoid underneath and immediately bought Apple stock.

In 1997 my university's computer science department taught on Apple computers. It was a nightmare. They would take your floppy and not give it back. There was no eject button and you didn't know if it saved your work if you wanted to force eject it by inserting the end of a paperclip into the little hole on the front. Any problem with the computer, the tech would usually tell you to switch to another computer and he would reboot the one you worked on, which would essentially format it and reload a fresh install from the server on it. That was how you'd fix it.

Reimaging desktops to fix issues made sense back when the OS didn't offer any real multi-user protections, but it's still SOP even today. Maybe your school was on older macs, but the G3 desktops of 1997 used a physical eject button for floppies. On the other hand, if your CD got stuck you were still looking at a reboot.

in fact, the mac product line at that time actively hindered the iPods success.

I remember people saying that they’d buy an iPod if they didn’t need a Mac- and firewire was quite rare. Eventually people realised it could be used on Windows and had USB, but they had to use iTunes, which was a hinderance too.

Really, it’s amazing that it was so successful, it shows that the right kind of product with the right marketing can overcome some pretty significant challenges- but yeah, people didn’t want Apple products unless they had to have them for creative workflows.


iPods were FireWire not USB.

Worth pointing out, though, that that was hardly a disinterested assessment.

The M1 killed the intel based Macbook.

It basically made that platform obsolete.


M1 and intel werent different products. Theure both macs with a breaking change in between

I guess I see ditching x86 for ARM as a big deal.

I feel like that’s not really the same. Each generation of hardware aims to make the previous obsolete.

The M1 was a bigger jump than normal, but it also increased their margin and control.


If Apple is financially more successful it is only because they figured out to extract maximum value from the captive base created by Jobs pushing for so much innovation in the earlier years.

When the iPhone first came out, it was only expensive. Now it’s a full mortgage payment.

Worse, features have remained relatively unchanged since the iPhone 6…


Why do people insist on saying Apple isn't as innovative since Jobs died?

   * Apple Watch
   * Apple Pencil
   * AirPods
   * HomePod
   * AirTag
   * Vision Pro
   * iPhone Duo
   * Titan / self-driving (massive investment in advanced technology, despite not launching)
   * Apple Pay
   * Apple Silicon 
   * Touch ID
   * Face ID
   * ProMotion
   * Metal
   * Swift
   * ...

Most of these are Apple versions of things that had already been done successfully elsewhere.

Apple did not invent or even popularize smart watches, fingerprint readers, Bluetooth earbuds, smart home hubs, VR/AR headsets, self driving, mobile payments, foldable phones, etc.

About 2/3rd of the Apple versions of these aren’t even the most popular in their category.


Where is Blackberry, Nokia, Palm, how’s Intel, AMD, Microsoft doing in the mobile world, tablets, Smartwatches, how’s HP, Dell doing whose position would you rather be in if you had a choice? A vertical computer company? With an operating system, hardware, and the ability to create an ecosystem from the ground up?

If it was easy why is Microsoft failing to expand their reach in fact, Microsoft has problems when the playing field is level, how’s their team ups going? With Qualcomm/Nvidia going why are they laying off 60,000 people? Why do so many of their acquisitions go into the dumper in comparison to Apple.

Why is Apple the only big tech company or one of the few big companies that didn’t fall for what Sam Altman was selling AI wise who is making the rounds???

Apple has a tendency to want to try to make the best device and be the most profitable. They’re not interested in having the absolute market share (Wall Street still believes they should be cheap and sell clones) if they were they could be Samsung, HP, Dell or Azus.


They also didn’t invent the mp3 player or the mp3 store, but they did reinvent the business model.

Apple is like classic Blizzard, they don’t (with a couple of admitted exceptions) create entirely new things out of whole cloth, they take proven concepts and finesse the absolute heck out of them.

Vertical computer company, if it was easy why are so many dumbfounded by it?, there was a time when there were many companies that did hardware and software together, as one.

I’m looking forward to one company taking Linux and their own hardware, combining it as one, and selling it to the public as an easy to use product, and no, I do not mean being a parasite on someone else’s hardware.


Steam Deck

Vertical integration ain’t easy, just efficient. Your second paragraph sounds kinda like Android.

That is also true of most of the products launched under Jobs.

iPhone is maybe the only one that seemed wholly novel.


I've never understood this claim. The iPhone was incremental, not novel. Palm, for example, had the previous generation.

The iPhone was not innovative in being a smartphone, it was innovative in that it defined the now default all-screen form factor.

It was an incremental choice, but it was a bold choice that nobody other than hard-headed Jobs would have done. Common knowledge at the time was that tactile feedback was necessary for any serious use. Analysts thought it would fail because of the keyboard.

Even more than a year later, Google’s response with Android still stuck with a hardware keyboard.


Also, the screen actually working the way people wanted it to was pretty novel. I never used a PDA that was anywhere near that responsive.


> it was innovative in that it defined the now default all-screen form factor.

Except it wasn’t. That form factor existed long before the iPhone (I had many such devices).


I meant, in the modern way we do now -- qwerty and no stylus.

And what you’re describing is an evolution rather than innovation.

For example I used to use PDAs with my fingers rather than a stylus too. Lots of people did. The desire to do that was already known. It just took a while for the technology to be invented to support it. And that wasn’t technology Apple invented. In fact they weren’t even the first handset on the market with a capacitive touch screen.


But it was made possible by technology Apple invented -- their keyboard.

Sure, LG had a capacitive screen first, but they put a T9 keyboard on it and put it on a feature-phone.

And of course, lots us had tried using PDAs (or other smartphones) with our fingers. And given that experience, we all doubted that Apple's choice would have been usable, at launch. What we didn't know at the time is that it was in fact possible to correct for this in software.

What Apple did looks like an evolution in hindsight, but it was a very bold choice at the time. It was so big of a jump that other manufacturers still resisted the change for years.


> The iPhone was not innovative in being a smartphone, it was innovative in that it defined the now default all-screen form factor.

That, like the GUI in desktop computing, is something Apple sniped the credit for: the LG Prada had an all-screen form factor, capacitative touch screen, and somewhat similar UI to the first iPhone, and was released four months earlier.

Not saying the iPhone didn't do it better or doesn't deserve its success. It just was the most popular, not the first.


It wasn't a smartphone though, it was a feature phone with T9 input. It was a Prada collab for a reason -- even LG thought the design was for vanity, not serious use.

Every smartphone at the time had something to supplement the input to make it a "serious business device", you "needed a stylus" for accurate input on a screen, or a hardware keyboard.

Analysts largely predicted the iphone would fail because it didn't meet these (and other) presumptive truths of the time. It looked like a fashion phone. It had the features of a music phone. They sold it as if it were a smartphone.

Even in tech circles, most thought it simply amounted to an "ipod phone" -- something that would appeal to (rich) chatty teenagers who like to text and listen to music. It lacked enterprise email integration and didn't have a keyboard, so there was no reason a business user would ever want one! What we were all missing, was that the features mattered less than the experience of using the device, even in an enterprise environment where we thought it was IT driving purchase decisions.


I think "modern" might be a better term for this feeling, thinking back to my old phones.

It seems like people want to revise history to make the iphone some super special snowflake innovation. But I don't think it was, I had a PDA before iphones and knew it was only a matter of time before they became common, I just think the iphone was very well timed with the marketing needed to back it up.

Palm and Blackberry were crap. Neither company had no clue when it came to an operating system combined with hardware none.

Previous generation going nowhere, just like that French hard drive music company, Arcos?

Palm and Blackberry, despite seeing the original iPhone and having 3 to 4 years to respond, they were utterly clueless and the same applies to Nokia. and Motorola. It was the OS. Google with the help of Eric Schmidt understood.

What was novel to Palm, BlackBerry, Nokia was the OS at the time within the iPhone all three were dumbfounded…


Years before the iPhone we had phones running Windows Mobile and Symbian. They both allowed installing pretty powerful third-party apps, which the first iPhone didn't.

What the iPhone brought to the table is a UI optimized for using your fingers instead of styluses.


> Palm and Blackberry were crap

You could argue the exact opposite, iOS released as a more restrictive and limiting OS than pretty much everything from that era. The j2me apps scene was vibrant in its own geeky niche. What the iphone did was to show the non-geeks what their phone could do, only by pretending that only the iPhone could do it, and billions in marketing succeeded in that.


The original iPhone was so restrictive, even by modern standards. It didn’t support multitasking, copy/paste, 3rd party native apps (on release) and a bunch of other stuff that were available on even feature phones (never mind smart phones and PDAs).

What the iPhone did was superior was polishing that turd and then convincing millions of people that it was “magical” via bullshit PR.

It was years before the iPhone matured into something I would spend my own money on.


I got the first version and it was a breath of fresh air. The alternatives at the time were so crummy it was crazy. I remember testing out some Sony Ericsson smartphone just a week before I got the iPhone 1 in a store, you had to use a stylus and point at something that looked like Windows with drop down menus along the top. No capacitive touch. Everything took half a second to render a menu even dropping out. The iPhone completely blew every UI away instantaneously, to a degree that people ignored its pretty crappy 2G radio/modem. The hardcoded apps were definitely enough at the time to be usable (camera, email, web browser, phone, music).

The UI was nice and the capacitive touch was an improvement over what came before. But it just couldn’t do enough for a lot of people who already used smart phones, feature phones, and PDAs.

The market was split at the time: those that basically just wanted a prettier phone with with email, and those who wanted capable portable computing device. Each market mocking the design limitations of the other.

You don’t get that division these days because Apple caught up in the capabilities department. But people seem to forget that there was a time it hadn’t.


The proof was in the pudding. The iPhone was better for most people despite the limitations.

Most people don’t own the first gen iPhone.

But I do agree it was better for some people.


Not even. I mean, he had a strong opinion on the lack of stylus. But iPaq? BlackBerry? Treo? PalmPilot?

I think it was more a case of, there was no HUGE new category like the ones Job's oversaw. Categories that others end up desperately trying to copy because they feel behind, the only one I can think of it Airpods.

That said, any company would kill to have a product like the Apple watch, but Apple sold more iPhones last year than Apple watch has sold in its entire lifetime.

A lot of what you mention while cool that Apple have done those things, aren't particularly "WOW!" level stuff, they are just neat things.

I would argue that more damage has been done by having Metal in that it has ended up creating a moat between them and Vulkan software that came along a year later.

Also Apple silicon was started under Jobs when they purchased PA Semi in 2008.


The response by Apple under Steve Jobs to buy PA Semi, Intrinsity, and Anobit probably came after Steve asked Intel to design the chip for the iPhone and they said no. It would’ve been interesting to be a fly on the wall when Steve hit the roof and said we gotta design something ourselves in-house, and all the people in the room saying/thinking, is he crazy?

There was a huge new category: Apple TV+. Sure, Apple wasn’t first, but neither was the iPhone or iPod. Apple making a whole new major media company from nothing is a major accomplishment.

Jobs wasn’t the person who spent a decade taking PA Semi’s heritage and turning it into a chip that beat Intel. That took years of consistency in engineering management, where every single year that team was pushing out improvements.

> but Apple sold more iPhones last year than Apple watch has sold in its entire lifetime.

This is not because the Apple Watch sells poorly, it’s because smartphone product volume is absolutely insane.


I love Apple TV shows (their movies are crap) but when you compare their categories to any other major service, Apple TV looks more like a hobby than a business.

I wouldn’t call a subscriber count higher than Comcast’s streaming service a “hobby.”

I wasn’t saying it’s a literal hobby. Just that their catalog is limited compared to other major streaming services.

Right. I bristled a little at the characterization that Tim Cook is the new Ballmer. Tim Cook did an excellent job at executing on the proven plan. At the same time, Apple's Wearables category (Watch, AirPods) did $8.3B last quarter[1]. That's a whole new product category for Apple, and it would be a big business for any business that's not $400B in annual revenue.

I believe Apple Pay falls under "Services", which has been a big growth area for Apple under Cook, though I think that represents less innovation than the Wearables.

CarPlay isn't a big source of revenue, but Apple did make dramatic, albeit not spectacularly successful, efforts to get into the auto space.

What Vision Pro becomes is still up in the air.

Apple is never first to market, but they generally come to market with a product that's a fantastic example of what their vertical integration can do.

1: https://sixcolors.com/post/2026/11/apple-announces-record-11...


Those are mostly improvements over mobile and complementary techologies. They could have built EVs but lacked the drive of a visionary founder.

Apple silicon was the inevitable byproduct (IMHO) of Apple buying PA Semi in 2008 for a bargain price I might add. I think Steve deserves a lot of the credit for that one.

I'd say Tim's biggest successes were Apply Pay, the Apple Watch (although this was originally targeted as a luxury item instead of a health/fitness device), AirTags and probably AirPods. Apple Pencil seemed inevitable. Vision Pro? Um, AR is kinda dead. Duo? I think the jury is still out on this in terms of reliability. But we've had other folding phones for years.

I'm still mad Apple has largely abandoned Touch ID. I despise Face ID. Give me, the user, the option. Put a sensor on the back of the phone, you know like Samsung has been doing since the Galaxy ~S8.


These days it's either under-display fingerprint sensor or side-mounted. Last Samsung S* with a rear sensor was S9 (2018).

Rear-mounted devices are totally gone, two (2) in the last two years: https://www.gsmarena.com/results.php3?nYearMin=2025&nYearMax...


Exactly. And you ain’t even mentioned the M chips! All the people I know that use local models have them yet people talk about Apple as if they ain't got nothing to show for AI…

Apple Watch: released 2 years after Samsung Watch

Apple Pencil: 4 years after S Pen

HomePod: 4 years after Alexa

Vision Pro: 5 years after Oculus Quest

iPhone Duo: 7 years after Z Fold

You can say the Apple ones are better than the stuffs they copied from in whatever fanatical metrics you use, but those aren't evidence of innovation.


For Apple this is more of a feature than a bug. They essentially let others do the beta work and smooth out the supply chain and production before they package their own stuff with integrated ecosystem to reap the fatter margins.

iPod - "No wireless. Less space than a nomad. Lame."

The Vision Pro is a perfect example of the modern Apple "innovation".

Come into an already established industry. Refuse to acknowledge the rest of the industry. Refuse any compatibility, proprietary lock in required. Refuse to learn any lessons. Invent (or in this case appropriate) some bullshit term like "spatial computing" so that your customers can feel smug about not buying the peasant "virtual reality" headset. Price it insanely high. Fail spectacularly.


Why do you insist on defending them? Even if you an employee, stockholder and decked out in Apple gear, you aren't gaining anything throwing yourself into battle for them at all. If other people don't agree with your take on Apple, so what they can suffer. Public opinion won't change until the media they consume suggests something else to them for a while in a subtle enough way they will believe themselves clever deducing it.

I'm not so much defending them, as calling bullshit on the idea that they stopped innovating 15 years ago. It's a ludicrous claim. As I listed, there are a ton of new products and core technologies (some extremely successful) that didn't exist in Jobs's time.

Also the Touch Bar. I don't think it was a successful innovation, but it was absolutely innovative.

The memetouch emojibar? please.

Can you explain what was not innovative about it? You may want to look up the definition of the word…

A 2014 thinkpad had a touchbar (the Adaptive Keyboard, see [0]).

The 2011 Optimus Popularis keyboard [0] also had a very similar LCD screen innovation [1], and I think was the earliest good implementation of this sorta idea.

I think you can argue that it was largely derivative or reasonably innovative based on the fact that it had been done before (with even worse execution), but it hadn't ever been done at such a large scale, so it still felt very new to consumers.

[0]: https://arstechnica.com/staff/2014/01/stop-trying-to-innovat...

[1]: https://www.artlebedev.com/optimus/popularis/


Well, they're right. It was innovative. It was just bad.

Like having a touch screen Mac, which is only a niche product, Steve Jobs was right.

AMD got ahead of them and bought the main company that does that first.

https://www.forbes.com/sites/jonmarkman/2026/08/09/amd-buys-...


>Just like he went all in on the iPad which had the potential to kill the cheap student Mac.

Every single review of the iPad for the latest 10+ years has been "it's a powerhouse that is being intentionally nerfed".


But Steve Jobs died over 15 years ago.

> If Nvidia's next CEO is smart, they'd buy up a bunch of these companies that "burn" weights to silicon ...

IMO that's more the kind of thing that a bean counter / caretaker CEO like Tim Cook would do.

It seems that most companies that really chart their own path in interesting ways are run by obsessive assholes like Jobs, Musk, Huang. If you put a professional manager like Tim Cook or Steve Ballmer in charge then they will just milk the cow, which may be highly profitable, but also highly boring and unlikely to result in anything interesting.


I think the ipod -> ipod mini is just natural upgrade. In both cases they bought out exclusive supply of tiny little harddrives for use in music players (you could still buy them for e.g. a windows CE sleeve like on the iPaq, they only bought exclusivity for dedicated music players). Once someone came out with a new tinier drive they didn't have exclusivity with them and if they didn't buy out exclusivity on it for the iPod mini they would have been destroyed by competitors releasing much smaller music players.

And they could outbid everyone else because of their premium brand, experience, and the margin they were getting.

Zune was able to get access to them with the Toshiba Gigabeat loophole by having Toshiba manufacture the Zune. The deal with Toshiba didn't cover stuff Toshiba manufactured. The smaller drive in the mini was made by Hitachi.


The iPod Mini to iPod Nano was the big strategic move. When they killed the Mini it was selling like hot cakes. Apple needed to get the cost of flash memory down and shifting the market to be flash paved the way for the iPhone, iPad, etc.

Apple also has the advantage that they can always grow down the market. The Neo is the perfect example of this. It's easier for Apple to sacrifice margin for market share than it is for a competitor to move upmarket.


> If Nvidia's next CEO is smart, they'd buy up a bunch of these companies that "burn" weights to silicon,

Models are getting significantly better every 4-8 weeks with post training now.

I don't think there'd be much demand for this.

~80% of the total cost of inference is chip depreciation. If inference is 10x more efficient, that only drops your running cost by about 5% absolute tops, and that's assuming the chips cost the same and not more for this. That's also assuming the chips have the same lifecycle, which they wouldn't.

No one in the world is going to want Gemini 3.5 Flash at a 5% discount 1 year in the future. You've taken a 3-year chip amo, turned it into a 1-year, and that's 80% of the cost.

Hard pass.


Nothing great about a man who underpays people systematically:

https://www.latimes.com/business/la-xpm-2013-jan-23-la-fi-tn...


This website sometimes broils you alive if you mention true things about Steve Jobs.

im really worried that amd is buying these silicon baked chip companies to kill it at worst or side car it at best. i really think these baked silicon chips will be a game changer id happily buy 100k worth of these chips if we could have one order of magnitude of upfront cost and power consumption in exchange for no ability to update the underlying model. a future where we arent rationing tokens is an interesting one.

You mean like they did with Groq?

> What made Steve Jobs great was that he would rather have Apple cannibalize its successful product lines than fall prey to a competitor. You can see this pattern with the iPod. The iPod disrupted itself many, many times. The iPod Mini then the iPod Nano. Then the iPod shuffle as the answer to the race ot the bottom.

Isn’t that how all successful tech companies operate?


You can be successful as in money without changing the world

Just existing changes the world.

>You can see this pattern with the iPod

If you want to show a pattern... you can't just bring a single example...


The difference is, it was consumers who wanted Apple's products.

For Nvidia's case, it's large companies that want (or can afford) its products. And the majority of people don't want the byproducts that come out of Nvidia's customers.


That's one decision that Jensen would be very unlikely to make. I totally understand why a lot of people are excited by AI ASICs, but imo there are a billion of these start ups not because it's definitely a better approach, it's just one niche with no clear leader yet. I don't think Jensen is necessarily being dogmatic in saying the CUDA platform is superior just because it's what made him wealthy. There are many good reasons to believe in the long run people will want the flexibility of general purpose computers. Historically the familiar and flexible technologies usually win over things like CELL arch and Symbolics.

Yes and no...

People bought graphic cards that were only good for one very specific workload, because that workload didn't fit the CPU architecture. While that eventually became another kind of general purpose architecture, it wasn't initially. Now we have NPUs and TPUs because the workload demands a specific shape of silicone.

At some point a model from two years ago that runs for a thousandth of the (dollar/energy) price of a frontier model becomes an extremely useful thing to have, if only as a subagent.


NPUs have so far been too narrow for even what they are intended for. Apple ships AMX/SME because their NPU has only a few data types supported and isn't flexible enough for all the things users actually want to do. AMD and Apple don't even try selling them as features for users to actually program for.

I know it’s beating a dead horse at this point but I just wish ANY of the GPU companies would be able to put out a nice, powerful card for gamers with a generational performance leap, just like the old days.

Or just something not so blatantly driven by market segmentation.

I know, I know all the reasons why. Still sucks though.


Earlier today I purchased a Blackwell GPU to replace my aging Turing, and the performance difference there is quite huge for a ton of work loads. That's an 8 year gap, not just a couple years, but I think hardware still steadily improves a lot over time. I guess I'm not old enough to remember the real Moore's Law.

The GTX 10-series was seriously something special at the time. Such insane power efficiency and performance, they basically blew AMD away in that department.

I had a Vega 64 in that generation. It was alright, but not magic. Atleast not until Ethereum mining was a thing.


The iPhone killed the iPod.

The iPod Touch ran until 2022. If it was killed by the iPhone it took a damn long time to do it.


If the next CEO of nvidia is anything like Steve Jobs we'd have an AI Appstore, and anything we build with AI would have a 30% tax.

Still waiting for the iPhone shuffle, tho! :D

People keep forgetting that after the original Mac classic, and during NeXT days, Steve Jobs wasn't doing that great as CEO, and most of Pixar success was related to other folks actually managing the company.

The cult has forgotten about their idol failures, as he managed to save Apple from insolvency.


Kill the cash cow.

Better you than your competitors.


> What made Steve Jobs great

What made Steve Jobs truly great–truly great–is the talent he built up around himself.

It's still too early to know if Jensen flunked this test. My hunch is someone managing multiple affairs almost certainly has to meaningfully delegate, which means there is a lot of silent talent in waiting amidst Nvidia's ranks. If there isn't, however. If he just built a cult of personality, a capitalist war lordship, I don't think we can call him a great CEO. Visionary, sure. Ruthless, yes. Accomplished, absolutely. But not great. Not someone we should be turning into an icon, or worse, worshiping with organized religious zeal.


> The iPhone killed the iPod. Most companies would have never created the product, because it would have killed their golden goose, but Jobs did.

I don't like when things are stated as axioms without any evidence. Frankly, I think it was blatantly obvious in the mid 00s that devices would converge - people had phones, PDAs, and music players, and it was a pain in the ass to carry around multiple devices.

I am not in any way minimizing what Apple and Jobs accomplished with the iPhone - indeed, I think they were successful because they were the only company with both the product design chops and the integrated software and hardware experience to pull it off. I'm just pushing back on the idea that other companies wouldn't have "disrupted" a successful product like the iPod because they would have been worried about cannibalizing revenue. It wasn't hard to see where device evolution was headed in the mid 00s, what was hard was just imagining what it's "final form" would take (hence all the bajillion different flip/twist/pull out and keyboard iterations of phones just before the iPhone).


[flagged]

Sir, this is a Wendy's.

> Most companies would have never created the product, because it would have killed their golden goose

It's not a tough sell for a hardware company; in fact it's completely normal and happens all the time. They release a newer version of a product around the same time they stop making the previous version. Bonus if the newer version is cheaper to manufacture too.

> Just like he went all in on the iPad which had the potential to kill the cheap student Mac.

From Apple's POV, is an iPad more like a cheap student Mac or a larger iPhone (but without being a good replacement for an iPhone)?

I think the choice to go forward with the iPad makes a lot more sense from that perspective. The potential iPad market is larger due to the generally lower price, and the margins on a "cheap student Mac" are by definition not great to start with. If iPad margins are similar or better then it would be a win even with 100% cannibalization.


     > It's not totally known what effect an EMP would have on stuff
Under Operation Fishbowl, the United States military conducted exo-atmospheric nuclear explosions to study the effects of an electromagnetic pulse. Multiple tests were conducted. The most famous of these tests is Starfish Prime which created an aurora, https://www.aps.org/apsnews/2022/11/electromagnetic-pulse

Sandia wrote a report called, "Did High-Altitude EMP Cause the Hawaiian Streetlight Incident?" - http://ece-research.unm.edu/summa/notes/SDAN/0031.pdf

There's probably a hyper detailed classified study sitting somewhere in a dusty drawer with that data that should probably be declassified. If you'd be down to FOIA it, I'd love to read it!

    > With how quickly AI is going basically all futures are weird in some ways. I've been thinking about how this affects me personally, and "try as hard as possible to keep AI from getting us all killed" is pretty high on the list. For me this mostly looks like donations and pandemic deterrence, but I also put some time into thinking about disasters. There are ways AI might go very well, or very poorly, and then there's the "messy middle" where we muddle through and things might fall apart some in the meantime.
    > 
    > The big things here are making sure you have enough water and food for a while, along with critical medications and the standard "prepping" recommendations. But lately I've been thinking about what you might do for other risks, such as engineered pandemics, mirror bacteria, or fire suppression outages. And today, EMPs.
I don't think this is a healthy line of thought, but I don't think most people who think this way are ready to hear that.

But based on the post, it seems that your threat model is off. The biggest threat of an EMP blast won't be your electronics acting up, but it would be the critical high voltage transformers being taken offline. These transformers have a lead time of months. Weeks in the best-case scenario.

From the congressional research service, https://sgp.fas.org/crs/homesec/R43604.pdf

     > Most HV transformers are unique and therefore are designed and manufactured to custom specifications for a specific network application. In 2010, the lead time between an HV transformer order and delivery ranged from 5 to 12 months for U.S. manufacturers and 6 to 16 months for foreign manufacturers, although lead times well over 20 months could be required in certain situations.10 This process may include three to four months for the engineering design alone.11 Since manufacturing generally occurs on a single production line with just-in-time component supplies, advanced production scheduling is important for managing delivery.
    > 
    > Physical assembly is labor intensive, requiring manual winding of the copper wire around the transformer core and frequent engineering checks during manufacturing. Extensive testing of completed units also contributes to HV transformer manufacturing time. The installed cost for an HV transformer depends heavily on its configuration and specific design requirements. 
    > 
    > New HV transmission substations can cost well in excess of $10 million, including the cost of transformers and other station equipment. According to the U.S. Department of Energy (DOE), the factory prices for HV transformers typically range from $2 million for a 230 kV unit to $7.5 million for a 765 kV unit, before transportation and installation costs.
If you want to prepare for an EMP, you need to prepare for a world without electricity, major telecommunications, no hospitals, no emergency surgeries, no air traffic control etc.

It's possible to do that but it's not possible to do that cheaply.

As far as contingencies for EMPs for your personal items go. You can use water as a very effective shielding medium. If you have an average sized backyard swimming pool, and if you can somehow bury something at its exact center i.e. under the pool, then a small box there would survive almost anything. Water is a really, really, really great shielding material.


The Oahu info is interesting. That was a series street light system, something rarely seen any more. The idea is that all the bulbs are wired in series, the street light system is a big loop, and there's a constant-current transformer at the head end, providing a few thousand volts.

Constant current lighting is kind of a mirror world of normal constant voltage systems. Each lamp is equipped with not a fuse, but an insulating shorting disk between two contacts, wired in parallel with the bulb. In normal operation, each lamp sees 1/N of the system voltage, usually 120V. If a lamp burns out, no current flows, all lights go out, the failed lamp's insulating disk sees the full system voltage, the high voltage arcs through the insulating disk, the contacts close, and the dead lamp is thus bypassed. Now the lamp string has less resistance, current goes up a bit, the constant-current transformer adjusts, and things go back to normal, minus one lamp. The Oahu system had a mix of incandescent and mercury vapor lamps, running at different voltages. complicating the situation.

This is apparently a bad case for EMP vulnerability. But all that happened was that some of the shorting disks at the voltage conversion points arced through and cut that part of the loop out of circuit.

Those old series lighting systems are almost totally gone. But one application for series lighting remains - airport runway lights. That's because it's the easiest way to get uniform voltage for a string of lights several miles long. No need for a distribution transformer every few hundred feet.


> I don't think this is a healthy line of thought, but I don't think most people who think this way are ready to hear that.

Happy to talk about this if you like!

> But based on the post, it seems that your threat model is off. The biggest threat of an EMP blast won't be your electronics acting up, but it would be the critical high voltage transformers being taken offline. These transformers have a lead time of months. Weeks in the best-case scenario.

I agree with you that long-term blackouts are the largest impact in an EMP situation and I should have said something along those lines in my post. I've been thinking a lot about backup power already, and part of what I was looking at in the post is whether I need to take extra steps to protect my backup power options from an EMP (it seems to me I really don't.)


    > Happy to talk about this if you like!
I'd be down! How do I get in touch?

Sorry, I meant here! I like to talk in public as much as possible (https://www.jefftk.com/p/comment-dont-message)

Sure. I've interacted with you before and communicated my point of view frequently. See,

https://news.ycombinator.com/item?id=49620746

https://news.ycombinator.com/item?id=49637557

The self-fulfilling prophecy, https://news.ycombinator.com/item?id=49890381

On WMDs, https://news.ycombinator.com/item?id=50293688

An intuitive explanation for why a lot of the hysteria is wrong, https://news.ycombinator.com/item?id=50484072

How the hysteria is already harming the field, https://news.ycombinator.com/item?id=49651727

And the walling off of the global intellectual commons, https://news.ycombinator.com/item?id=50028478

Whereas the real risk of these machines aren't being discussed or countered, in fact the hysteria makes that outcome almost certain, https://news.ycombinator.com/item?id=49252258

https://news.ycombinator.com/item?id=48024901


I'm confused: I don't think any of those are responses to comments if mine? (But I might have missed it!)

Apologies, on re-reading that comment comes off as hostile. No, I meant it as a way to explain my position / point of view. I'll have to hunt down our specific interaction. Apologies for not being more specific.

The linked comments are essentially a scattered version of the anti-doomer thesis I'm currently writing. I welcome any pushback as well as your thoughts.


That makes sense! My initial impression was that you thought we had a long history of talking about this together.

I'll see if I can find some time to read through and think about the overall picture.


I've said this before, but I'll say this again, but LLMs are the first time most people have used a computer as a computer. Myself included.

LLMs are realizing the potential of the computer revolution.

Those heady dreams of Xanadu, Xerox PARC, Douglas Engelbart, and Bret Victor today? They're finally possible.


I've been saying that computers are just now getting out of their beta period. I hope that we continue to recognize that any computer you can't freely program is not really yours.

They've been possible the whole time in software designed for the user, it's just that the business world will do everything in its power to undermine the user's control to protect their business model. The primary barrier is not the difficulty of writing some drivers for a piece of hardware, it's the intentional barriers that the mfg puts in your way so that that is harder to do -- lack of docs, obfuscation, various kinds of drm. A win is a win, but the same will happen in this new world. The barriers will just have to be built a bit higher until most users give up.

One of the things about the AI hype train that really got me was how many of the stories of wonderment were about the things that have always been trivial on a computer if you actually use it for computing instead of just running somebody else's proprietary magical spell. Obviously it's capable of nontrivial things, but I heard so many stories of like "I would have had to manually rename 100 files to a different naming convention, but claude wrote a 20-line python script in 30 seconds that did it write away" and you hear what the renaming was and instantly know how to do it in one trivial line of bash. In other words, stories about it doing things that computers have always been good at, that have been taken away from you by the notion that computers are for begging the avatar of a corporation to work some magic.

Which points to the UX barrier of the average user to write that line of bash.

That’s the magic; the barrier is down.


The ux barrier is in your mind, man. This is what I'm saying.

Really, the ux barrier is how much "free" time you have to devote to training your mind to overcome the ux barrier. While many of us live and breathe this kind of abstraction, the "average" person does not want to learn a new kind of mental gymnastics to accomplish everyday tasks.

You can study for a lifetime to become Yen Sid and teach the broom to fetch water for you without lifting a finger, or you can just fetch the dang water yourself in 10 minutes.


See also: Tourists, Bears, and Trashcans

Claude (or whatever bloated proprietary software you are using it to circumvent) is the magic broom, notwithstanding the fact that you weren't the one teaching it. bash is fetching the dang water yourself. We're not talking about inventing novel algorithms here, or writing some masterpiece of code that will always guess your intention correctly. Back before mass-marketed consumer software, plenty of regular people made it work without a computer science degree or a particular genius.

Yes, this is the shift software developers are going to really have to get used to.

People all have access to a ton of computer hardware which is capable of remarkable things.

Historically they have been restricted in what problems they can actually bring that hardware to bear on by the availability of software tailored to their specific need.

LLMs are capable of unleashing the power of your computer hardware to do all sorts of things you never would have been able to consider before.


I think it's a good thing, because using a LLM like this is the first time most people have used a computer as a computer. It's the first time people have actually made something via an intermediary that executes on their computer as opposed to installing an application built by a developer.

In so many ways, LLMs are the computer revolution realized. A lot of people just don't know how to write a driver, what it is, or how to investigate this. I certainly wouldn't. Nor would a lot of trained mathematicians and scientists. It's not immediately obvious how you'd tackle a problem like this to them and it creates this feeling of learned helplessness.

Now, thanks to the age of magical machines. This person wanted a piece of their chidlhood back and voila, they did and now can share it with the world if they so choose.

I do think that people should get better about attributing the machines though. I wouldn't be comfortable claiming ownership or authorship over code written by a machine entity so I do my best to mark machine-authored code as machine authored code.

An example, I know about snell's law and a little bit about optics. Nothing beyond the standard undergrad stuff (right now). I've tried Apple's liquid glass and I didn't like it. I think a part of it is that it's not translating the physics well and most implementations fall flat. It sort of either exaggerates or doesn't. Two nights ago as a passing fancy, I decided to work with codex and gpt to make it into a library that I can use for a larger research project. https://github.com/areoformed/daphnis-glass https://as-we-may-think.halfdrop.workers.dev/examples/as-we-...

It made me happy to watch a machine go off and do checks + validate the equations I gave it and to watch it write,

    > The numerical checks agree with the central magnification limit and your rainbow equation. An important consequence for the UI is already clear: the distance between the drop and the page can change both magnification and orientation. “Refraction strength” alone won’t describe the effect.
But as I did nothing more than write up a bit of insight, prod the machine, and make a detail request, I made sure that the machine entity got attribution for it, "Developed by Codex in collaboration with areoform. MIT licensed."

I wouldn't have been capable of turning that insight / thought process into code before, but thanks to the machines now I can, and I can even share it under a permissive license so that other people can find some small joy in it?

Is it soulless? Perhaps. But consider this, I found joy in turning a passing thought about physics into something that tangible that I can use for myself. That wouldn't have been a possibility for me in November 2020.


It bothers me that so much of modern life is opaque and assumes that no one has the time and energy to read the footnotes.

I went and looked at the first study on the School Improvement Grants,

    > In 2009, Congress spent $3 billion on “School Improvement Grants” for the worst-performing schools in the country. As far as we can tell, those grants did nothing.
Is this true? I looked at the linked report and it's strange, https://ies.ed.gov/sites/default/files/migrated/nces_pubs/nc...

At one end it says, "Although schools implementing SIG-funded models reported using more SIG promoted practices than other schools, we found no evidence that SIG caused those schools to implement more practices."

What are these practises? Well... it seems that as a pre-requisite to getting funding states were made to divide schools into four categories / intervention modes,

    > ED required that each SIG-awarded school under Tier I or Tier II implement one of four school intervention models. These models required specific practices:
    > 
    > • Transformation. This model required schools to replace the principal, adopt a teacher and principal evaluation system that accounted for student achievement growth as a significant factor, adopt a new governance structure, institute comprehensive instructional reforms, increase learning time, create community-oriented schools, and have operational flexibility.
    > • Turnaround. This model required schools to replace the principal, replace at least 50 percent of the school staff, institute comprehensive instructional reforms, increase learning time, create community-oriented schools, and have operational flexibility.
    > • Restart. This model required schools to convert to a charter school or close and reopen under the management of a charter management organization or education management organization.
    > • School closure. This model required districts to close schools and enroll their students in higher-achieving schools within the district.
    > 
    > These required practices can be grouped into the four main topic areas promoted by SIG. Table ES.1 lists these four broad areas and the objectives promoted by SIG within each. For example, replacing the principal falls under the “identifying and rewarding effective teachers and principals and removing ineffective ones” objective within the “developing and increasing teacher and principal effectiveness” topic area. 
    >
    > The objectives listed in the table cover all practices promoted by SIG, which different models designated as either required or permissible.
In plain english, it seems that they're saying at one end that schools were "transformed," turned around, restarted, or closed in order to get funding. They also promited performance review plans and quarterly / yearly reviews. Fairly standard administrative procedure.

But then they start the summary by saying that there's "no evidence that SIG caused those schools to implement more practices" – the ones which were heavily promoted by the program and were a part of it?

I feel like I'm misunderstanding something as these seem to be two mutually and logically exclusive things. You are requiring something for funding, but then you're saying afterwards that you can't be sure that they did related things just because you required it?

What am I missing here?

Also some of these graphs look wrong. The error bars on Figure ES.3. go from ~ -0.2 to 0.25... With error bars that large isn't the graph meaningless?


Howdy Ted! It has been a while! Hope you're doing well! Glad that you're still at it. SSGL misses you! <3

So there are a few breakthrough therapeutics that rely on samples crystallized in microgravity. One of them is Keytruda or Pembrolizumab, a PD-1 inihibitor used as a neo-adjuvant in cancer therapy.

Historically, there was an issue with dosing pembrolizumab, as the manufacturing process was creating large uneven crystals that required physicians to process at the clinic to ensure proper dosing, but in microgravity? The crystallization is just gorgeous. absolutely beautiful

Take a look at this image, https://issnationallab.org/wp-content/uploads/2024/05/merck_...

Left hand side are the crystals grown on the ground. Right hand side is 0g.

That change made it possible to turn it into a safe injectable that patients could take at home. And they were able to recreate those conditions on Earth, https://issnationallab.org/iss360/merck-lab-publishes-pembro...

https://pmc.ncbi.nlm.nih.gov/articles/PMC6889310/

This is more of a scientist said at a conference to another scientist kind of story, but apparently, they got some seed crystals from the ISS that are now used on Earth.

https://techconnectworld.com/World2025/a.php?i=602

You can do a lot more than make just seed crystals. It turns out that cells in our body are inside of a 3D environment with forces / tension acting in all directions. When you smoosh these in a lab (very literally) and try to observe how a drug behaves in a model cell / model organ, then it doesn't have the same physical constraints acting on it. BUT in 0g that's different. In 0g, cells behave slightly more similarly to how they behave in vivo and that could be a huge difference maker for drug discovery.

One breakthrough drug alone is worth more to humanity than what the ISS has cost across its lifespan.

https://iafastro.directory/iac/paper/id/89939/abstract-pdf/I...


Tencent is one of the largest game development asset holders in the world. They're the Berkshire Hathaway of video games. They own, World of Warcraft (minority shareholder), League of Legends, Valorant, Warframe, Clash of Clans, Player Unknown Battlegrounds, FIFA Online, Pokemon Unite, Fortnite (minority shareholder) annd more. They also own 40% of Epic Games.

> They're the Berkshire Hathaway of video games

They are the Berkshire Hathaway of everything, to be honest. They invest in and acquire minority stakes in hundreds of companies.


Wikipedia has them at ~a fifth to ~a half depending on metric:

https://en.wikipedia.org/wiki/Berkshire_Hathaway https://en.wikipedia.org/wiki/Tencent


Give them another 70 years :)

Innovation accelerates, I give them 7 years :)

Do they have a stake in Fortnite somehow specifically/separate from their stake in Epic Games?

Instead of listing the publisher/studio (Epic, Riot, etc) he's just listing their games directly.

Don't forget all the big Ubisoft IPs.

They don’t own things they’re not the majority owner of.

> They also own 40% of Epic Games.

I wonder if that has anything to do with the latest vibecoded slop releases and shilling from Tim Sweeney.


This is just good writing. The kind that AI is meant to emulate. Alas we're at an age where writing well is identical to writing like a machine.

I suggest polite rebellion. Write well. Get better than the machines.


> This is just good writing. The kind that AI is meant to emulate. Alas we're at an age where writing well is identical to writing like a machine.

Literally none of these statements are true.


But it's bad writing, as GP points out. The second paragraph is particularly confusing—its central metaphor is clumsy (in my experience LLMs struggle with metaphors).

That said, maybe it's an artifact of being machine-translated from the original Russian.


Let me make the author's case as they aren't here to defend themselves. This is the paragraph,

     > I still don't know why I remember it in such detail. I used to forget what I'd had for dinner the night before. And now I remember the whole morning minute by minute, as if someone had laid it out on a table and labeled every piece. Apparently I have a lot of free time now.
They set expectations with, "I still don't know why I remember it in such detail"

Then they set up the comparator to the past, "I used to forget what I'd had for dinner the night before" which is actually doing double work, because it also means that they were so busy that they used to forget what they had dinner.

This creates the character assumption that the person was / is a workaholic.

    > And now I remember the whole morning minute by minute
Heavily hinting that the person has been made redundant, because they've gone from forgetting what they had for dinner (as they'd munch along) to remembering their morning by the minute instead of being stressed by work.

    > as if someone had laid it out on a table and labeled every piece
Need for mechanical precision. This person is most likely autistic / nerdy.

    > Apparently I have a lot of free time now.
And now the final blow. It's because they're now effectively unemployed. Why? AI. They have a job on paper, but it's uncertain / doesn't truly exist.

It's incredibly concise and well written story telling. I haven't seen LLMs do punchy writing like this since GPT-4o days.


It’s the imitation of good writing, only good on the surface.


This is terrible writing, and if you can't tell, there is something deeply wrong.


Parts of this read like the early internal monologue sections of Cryptonomicon. Terrible or not, it's distinctive.


Having used Claude a fair bit, it reads like the very distinct style Claude settles on when it tries to write fiction. I mean it's hugely improved from the early attempts of LLMs to create fiction (remember the adverbs!), but it's still pretty awful.


> This is terrible writing, and if you can't tell, there is something deeply wrong.

This is subjective. Countless times in the history of human literature a piece of writing has been called garbage -- or at least, flawed -- by some authority, only to later become lauded as a great piece of writing. So not being able to tell (or in this case, to match your own taste) is simply par for the course.

Remember all that's been written about Shakespeare's Hamlet and his use of mixed metaphors, "[...] or to take arms against a sea of troubles [...]". What? How do you take arms against the sea? And why was it before "slings and arrows", what does that have to do with the sea? Mixed metaphors are a sin, etc, etc.


Well I think the verdict on Shakespeare is in, and it's positive. I assume you've read some, if you have you'd know that the positives outweigh any minor negatives. We still use some of his phrases today.

Of course Shakespeare made mistakes or made allusions that contemporary audiences struggle to understand, and he's not the be all and end all of writing in English. Saying judgement is subjective does not mean judgement is useless or impossible. He's pretty good though and you can tell when you read him, it's gripping, human, and tight; the antithesis of LLM writing.

Back to this particular text. From the start, it smells of AI:

> That morning I woke up at 6:58, two minutes before the alarm. I lay there and waited for it to ring. And it rang. We both did our jobs.

Short choppy sentences more suited to advertising than prose, ambiguity without purpose, pointless personification, melodrama and tension over the most mundane details. One para perhaps could be a bizarre stylistic choice, but it continues like this for the whole passage.

It's possible this was written by a human so steeped in the language generated by LLMs they don't know what is human and real any more, but it seems much more likely it's a translation which was translated by LLM and 'improved' by taking its style suggestions, or simply generated/written by the LLM based on a much shorter prompt.

However the most important thing is not who wrote it, but the quality of the writing, and I think the verdict of most people disagrees with you.


I think you missed the point of what I was saying about Shakespeare.

> However the most important thing is not who wrote it, but the quality of the writing, and I think the verdict of most people disagrees with you.

"Most people"? Please, show me where "most people" are saying this. Where's the verdict, are you keeping a tally? "Most people" don't agree on anything, either here in HN or elsewhere.

Are you the arbiter of what "most people" consider a good writing style?


The votes are keeping tally, seems pretty clear, but it's only internet points, feel free to disagree with me on this writing, I don't like it.


I think it's fair to say you didn't like it. I don't like a lot of literature.

I don't see the points that state people disagreed with me, I see 188 upvotes to this story and many comments saying they liked it.

If you meant downvotes to my own comment, they mean nothing. Before, I had +4. Now I don't. I also got downvoted for explaining to someone this is a piece of fiction (hopefully you'll agree with me there is something wrong with anyone who can't understand this), yet I ate some downvotes anyway.

The comment that you wrote that if someone cannot see this is bad writing there's something "deeply wrong" is simply absurd.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: