HNHacker News
TopNewBestAskShowJobs

aspenmartin

364 karma · joined December 29, 2025

submissionscomments
aspenmartin··on AI is slowing down
Yes since frontier perf is too expensive. Consider though frontier perf costs roughly 40x less per year. So you see super adoption for mega costs by the companies with the scale and budget for it, and then folks who can’t afford it that have to wait until maybe fall until they have the game changing perf we saw for opus 4.5 that spurred the enormous adoption by the major players.

I agree that there are incentives to waste tokens but this is strategic: they spend now for people to build and explore, then they keep the things that work and drop the things that don’t.

aspenmartin··on AI is slowing down
https://epoch.ai/data-insights/consumer-gpu-model-gap

I think over the years local models have fairly consistently been ~7 months behind frontier performance. Local models are hugely important but I don’t see the calculus changing. I can imagine it’s certainly the case for many tasks that there will be diminishing gains for performance improvements or reliability pass some threshold, in which case you don’t need frontier performance and you can certainly use local models or at least cheaper tiers of proprietary models if local is too much of a hassle. Plus of course use cases where local is necessary or the pros of having local models or on device models outweighs that of frontier.

Maybe things will change though, I would assume through basically government subsidies from China etc, to undercut existing frontier labs, but you can always spend more (better data more compute etc) for better performance and that I can imagine will always have a selling point.

aspenmartin··on AI is slowing down
It’s 3 different weak but complimentary proxies. We form beliefs from imperfect evidence and I find these fairly convincing when it’s hard to find any hard evidence of no productivity and exactly the scenario you would expect under the hypothesis that we do see productivity gains. None of this is supposed to be unassailable. I would challenge then if you disagree what the evidence you have for this is?

Adoption implying at least some significant productivity gains doesn’t contradict there being other factors. You’re seeing entire companies reshaped. The argument is this is all for show or CEOs are in some sort of idiot class?

“It is possible for practices to be picked up or continued in spite of causing productivity drops” well of course. I just find that incredibly far away from Occam’s razor.

My point is: we have lots of evidence that’s highly consistent with real productivity gains, and I don’t see many pieces of evidence to the contrary.

aspenmartin··on AI is slowing down
All of these points are great. The first one motivates world models which lots of labs work on. Not many people tend to understand the strategic value of those “open world” or interactive generation models: its robotics and planning. But also like you say you’re right, there are complicated problems to solve and it’s not totally clear the timeline. But where there’s data and compute, there’s a way.

For conversational AI these labs do have lots of things to do lol but you’re right; it likely also requires some architectural improvements but you see the infancy: look at the llama4 speech duplex model. Very unimpressive yet all of the components are there. Just a matter of pushing on them, licensing and commissioning better data, etc. takes time and compute is stretched thin.

aspenmartin··on AI is slowing down
I don’t get this. We already have an insane demand. And yes exactly, this is primarily just with coding agents, but are you aware of what’s coming down the pipeline? It’s not hard to be you just have to find a decent way to keep up with literature.

* robotics (need to close data gap and release first viable product to get a data flywheel)

* conversational ai (no one is ready for this and we’re getting closer and closer to natural speech. The quality still isn’t good enough but it’ll be soon).

* other agentic use cases, openclaw adoption was crazy and that had a ton of barriers to entry

* ai products, like the one OpenAI is working on with Johnny Ive

Anyone thinking it’s unreasonable to hit whatever revenue requirements is just not that aware of what’s happening. Not to mention were capacity constrained already!! This is barely speculation at this point.

aspenmartin··on AI is slowing down
Cite a real source for that last bit, I don’t think that is true. Also the budgets should be cut the spend at some places goes beyond any reasonable amount. The strategy there is to hook everything in and find the right processes, then cut the rest. Things then get better and better with each model release.

The way you make a viable service that eats 300bn annually is to have enough demand to service that. Anthropic underbought compute. That tells you something.

aspenmartin··on AI is slowing down
Because it can perform much faster? Monitoring allows you to multitask more effectively. I would also disagree that you can’t one shot anything…claims like this are weak and I have enough counter examples in my own life that it’s trivially false. The question is more: can it one shot the right things with a low enough failure rate for it to be a good replacement. It’s hard to figure that out a priori.
aspenmartin··on AI is slowing down
It’s a very hard experiment to run. You have a population that’s already “treated”. You can’t blind them to the fact that they’re using AI tools. It’s hard to imagine a study that wouldn’t have serious flaws that people would then use to dismiss and form their own conclusions. Sure you have METR but that was very low n with a very old model.

I think the surest sign of productivity gains is the sheer volume of adoption. If you look beyond headlines, adoption is just incredible. Of course adoption does not necessarily point to productivity gains, but if this was some sort of FOMO or smoke and mirrors you would not see this much retention and this feverish a pace of adoption. You would not see a large segment of the profession using coding agents exclusively. All of these companies track productivity, again with imperfect proxies, yet everything points to a pretty consistent picture. Same with benchmarks, again a lot of crappy benchmarks but a lot of high quality ones too and a very diverse collection of tasks and capabilities they probe.

aspenmartin··on Ask HN: What was your "oh shit" moment with GenAI?
Curious what your interesting questions were, you should be able to find them in your chat history.
aspenmartin··on They’re made out of weights
And this has some bearing on the debate about whether these systems do or could in the future exhibit something similar?
aspenmartin··on Expanding Project Glasswing
Ah right sorry news articles I forgot these are the basis that form the entirety of our understanding of economics.

This is not the first technological revolution humanity has experienced. When they pass, aggregate unemployment doesn’t change, work gets shifted around. I mean there is plenty of data and historical precedent for what’s happening.

Again, YOU may disagree but to say that an argument to the contrary, that doesn’t rely on a game of telephone between a journalist with a requirement to be sensational and a tech CEO saying something more nuanced, is an argument in bad faith illustrates you really haven’t thought deeply or consulted any data or historical precedents for similar phenomena that have happened in our past.

aspenmartin··on They’re made out of weights
> With regard to AI, there's a lot of people extrapolating "There is no magical animating spirit, the brain is just a pile of stochastic molecules following the laws of physics" into "The brain is an inert pile of matter, computers are an inert pile of matter, ergo AI/LLMs are like the brain!"

Why do you need a specific organization of molecules for a phenomenon similar to consciousness to arise? Does anyone seriously consider a brain to be something other than “a pile of molecules following the laws of physics”? If so that’s not science or philosophy, that’s religion. You have a virtually complete phenomenological model of the universe for all intents and purposes and yet somehow the onus is on the person being like “hey no laws of physics are being broken ==> the brain is simply following the laws of physics”

How is it possible that people think of subjective experience and get rabbit holed into some mystical world where subjective experience is this special exception to everything else that is simply an emergent property of complex physical systems? “AI/LLMs are just like the brain” is a strawman, why does this claim need to be true for LLMs or any artificial system to be considered to have something akin to the thing you think of as consciousness? It’s more: consciousness is not some mystical or religious thing outside of the realm of physics, it’s an emergent property of a complex system. AI is a relatively complex system. We don’t really know or understand the relationship between the raw physics and again what we consider consciousness, so it’s simply a statement of “we can’t refute that these systems exhibit something similar” because we don’t know enough to refute that

aspenmartin··on Expanding Project Glasswing
Common and discussed opinions they are, yet they ignore any actual econometrics or understanding of how the economy works
aspenmartin··on Expanding Project Glasswing
Yes absolutely. However the benchmarks they did release numbers for are key ones and the leaps were very large. Absolutely possible that they either lied or that the full picture is much muddier but based on the numbers they do show that’s hard for me to imagine a likely scenario that would produce that.
aspenmartin··on Americans don't know how to fight AI so they're fighting data centers
It's a good point I have to say
aspenmartin··on Expanding Project Glasswing
What's your basis for this?
aspenmartin··on Expanding Project Glasswing
Why do you think the impact to the economy is bad? Also, if youre talking about the environment from the lens of data centers, I agree they CAN be incredibly problematic and that should be regulated and pushed back against when they engage in problematic behaviors (stressing water supply in a drought area e.g.). But not blindly -- data centers can be a clear win-win in a lot of ways, especially if done right. "data centers = bad" is way way way too simplistic a picture
aspenmartin··on Expanding Project Glasswing
> Im afraid that the usual mantra that "we just need more scale" that worked well for attracting investments, is not working anymore - bigger models provide marginal improvements while naturally get much more expensive to run.

It's super interesting to hear this refrain on HN, it is alarmingly common. Anthropic released benchmark numbers on Mythos, as they have for all of their models. Once models become public, people evaluate them in a myriad of ways. We have had reliable scaling laws for years and they still hold. Epoch capability index continues to grow exactly as expected. Where does this idea come from?

As for cost, the cost per token at a given level of performance drops up to 40x per year.

aspenmartin··on Americans don't know how to fight AI so they're fighting data centers
Oh sorry I just mean: with respect to AI
aspenmartin··on Americans don't know how to fight AI so they're fighting data centers
Depends, is the net improvement of the internet, electricity, agriculture, steam engine also in the room?
aspenmartin··on Americans don't know how to fight AI so they're fighting data centers
Of course and a lot of people disagree that vaccines work, why does this negate any hard evidence?
aspenmartin··on Americans don't know how to fight AI so they're fighting data centers
What reigning in would you do if you had the power to do it for the US?
aspenmartin··on Expanding Project Glasswing
Ok but they can easily upsell this to enterprise customers at a market price reflective of their capacity constraints. Big corps would pay it, this is clearly a major update.
aspenmartin··on Expanding Project Glasswing
To me this makes little sense — I can’t imagine the orgs they have limited this rollout to don’t already have Claude subscriptions and integrations. And sure this may play nicely into branding a build a mystique around the model but ultimately they are missing out on a ton of revenue and risking being totally front-run now that model performance parameters are out and people have firsthand experience. Feels more like a fairly genuine attempt to be responsible. They could have easily rolled out an update and done some PR to absolve themselves of responsibility
aspenmartin··on Expanding Project Glasswing
Ok you’re totally right, I read this as a cynical “this is all marketing” post ==> a scammy connotation. Without that read, your points are fairly valid, but are you still implying this is all a pure marketing tactic? If so I would still argue against that as a necessity but surely marketing could be heavily involved. But still: this could easily be a footgun. OpenAI will easily release the same model and now that Anthropic has taken the initiative to do a slower more contained rollout they wouldn’t need to do any of that. So from a business perspective I would still argue this whole glasswing initiative would make their sales and marketing department pretty nervous. I mean in a second-order branding sense sure this plays into the “we are ethical” ethos but it hardly seems worth the risk
aspenmartin··on Expanding Project Glasswing
These companies are surely already onboarded…? They claim like 10k verified and high severity CVEs. Would you have preferred they just rolled it out like another opus update? You wouldn’t be insinuating in that situation that they were careless and reckless? They risk missing a boatload of revenue if openAI front runs them for a public launch. In what world is this some sort of scam??
aspenmartin··on Anthropic confidentially submits draft S-1 to the SEC
If 1B people consume via personally or their employer on average $100/mo of tokens or services that solves the problem right there and that is not inconceivable whatsoever and feels like an underestimate. Repayment doesn’t imply you have to build an end to end replacement of an industry.

This also negates this whole like “you have to completely replace a human” fallacy. Why do you need to replace a human? Why not just increase the value each human brings you?

The model of open weights has been around for awhile. You have frontier labs releasing them. They are powerful and capable and track yet lag frontier models. Without massive government subsidies from probably China who did something similar with manufacturing I don’t get the idea of OSS somehow toppling the entire industry of frontier models. How would this happen? Did demand for frontier models drop after GPT-4? GPT-5? Because the cost of GPT-4 perf is maybe 100x cheaper today. You can always pay a premium for a better model (more data spend with proprietary data sources, more compute for training, more thinking budgets), and in the end, things will either go to a monopoly or prices will stabilize around the marginal costs.

Also safeguards are always important. They are thorny fundamental horrific problems and I can tell you there will be a hellscape of pain as people figure out the trivial ways to do bad things with some of the worst security practices we’ve seen. But I don’t get why this is a dealbreaker, we are using these systems everywhere and in loads bearing environments, today, and reliability and hallucination rates continue to increase / decrease.

aspenmartin··on Anthropic confidentially submits draft S-1 to the SEC
> All rallies do come to an end. The fact that we all don't know exactly what will cause this one to end is exactly part of the problem and 100% doesn't mean it won't happen. Usually some external shock spooks the market and a massive sell off happens.

Yes sure, but that statement contains zero information -- why do you believe it will end in a time short enough for the "market bubble" comments to even make sense?

External shocks -- sure of course. Inflation problems in US -- absolutely, it's a ticking time bomb with a debt crisis looming. Housing market I don't really know anything about but I'll take your word for it.

But all of this has been true for awhile, and could have been stated with equal veracity over the course of the last 5 years at least. Your beliefs shape your actions; so why does this belief shape actions any differently than it has earlier?

> ut honestly, I'll tell you after it happens and it will happen. Having lived through a few of these now when everyone tells you it's a sure thing and prices go up for ever you get an inkling you are near the pop.

Again totally true, I have also lived through them and expect more. But "these companies are IPO'ing because they know the market will pop" is kind of the thing that I was trying to address. For all the signals of market danger, there are plenty of optimistic signals all over the data. Growth is pretty robust across all sectors today.

aspenmartin··on Anthropic confidentially submits draft S-1 to the SEC
> But mainly, they are in "I need to kill whole industries to be worth it" tiers of investment.

Yes agreed. Coding is a pretty big industry though in and of itself. Same with healthcare, legal, etc etc etc. Of course we have zero model today that can seriously kill an industry, but if you look at (1) how good things are today (insanely fast and rapid adoption) and (2) robust performance trends from many complimentary sources, it's kind of inevitable and I haven't really heard a coherent steel man argument for why "killing whole industries" is somehow a far-fetched idea.

> still not profitable enough to make up for the costs of either training or the investments gifted away until it became profitable.

Regardless of the weeds of the economics today, you have a clearly valuable asset that at the very least already a must-have for enterprise and will become even more essential over time. There is token economics that either already do or will make sense. You will have some sort of marginal cost + profit margin that things will stabilize at. You can pay a premium for high quality frontier models. "But it costs more in R&D to fund this!" ok but then token costs will increase. Why is this some sort of death knell?

aspenmartin··on Is AI causing a repeat of frontend’s lost decade?
> Nothing was set in stone. The way many historical things happened the way they did was due to accident, sheer chance.

I agree with you on a technical level and even in a non-cynical "humanity really can rally and change things that seem insurmountable" but you have read way more history than I have. All I know is you have such a frantic geopolitical aspect to this, and such a staggering amount of funding and addressable market, which means unlike blockchain this is both powered by business _and_ government (no government would give up control of the money supply, shocking to me that people believe this), that I see zero path to winding anything down.

> I think the current LLM economy will collapse, leaving behind a few survivors. There will be widespread adoption of cheap OSS LLMs and of more limited, economically viable functionality provided by people with deep pockets like Google. As LLM economics start making more sense, LLMs will be everywhere, once the hardware becomes cheaper and more available.

Cheap OSS LLMs are used everywhere, all the time. They are great, and with subsidies from say China, they could even be competitive with frontier models, but that model of the world requires this mysterious OSS development running at a big big loss. It takes almost a billion dollars to train a frontier model. For many many cases, you do not need frontier model performance. When I do say video captioning, I use small OSS VL models.

Is your theory predicated on OSS models filling some sort of performance gap between the frontier? Or a compromise for less spend at a lower performance? What to you doesn't make sense about LLM economics? Like, LLM's are everywhere. If you think "oh people will just settle for slightly less performance for cheaper" that should have already played out, we've had the same dynamics at every scale: frontier performance is expensive, but then that same performance will cost roughly 10x less in 1 years time. But you don't see people stopping at like GPT-4 and not adopting the frontier models of today.

I think you're right about the value of OSS LLMs but I don't see what would change the calculus to make frontier models somehow less important. It's like in the 90's when we were like "1 GIGABYTE OF RAM? how will that ever be necessary!?!?" and sure, you don't need 1 GB of ram for everything! We have embedded systems. But it's not like there isn't a booming market for >> 1 GB memory modules.

> Regarding lawsuits, do you think Disney & co will take this lying down? The freaking DMCA - an American law - is enforced <<internationally>>. It will take a long time but LLMs will be domesticated.

Not saying lawsuits will be fruitless, I'm sure they will chip something off of the industry, but by now it just won't matter. We're talking about trillions in spend, multiple countries, a government that sees this as a non-negotiable game to win from a geopolitical and military standpoint, and our government knows that they can't execute this themselves despite what I imagine is their distaste for Silicon Valley tech CEOs and their grandstanding. Maybe a lawsuit kneecaps someone, which would be huge, but that doesn't matter for AI generally. Maybe a lawsuit restricts data use, that's fine, these companies have deep pockets for licensing and commissioning datasets; they have opt-in-by-default user flywheels.

← PreviousPage 5 of 15Next →