I am trying to understand something about the economy and AI. If AI disrupts an industry that previously had high wages and now can be automated by commodity AI products, would it not actually lower GDP in some way (initially anyway)? What I mean is, if you had an industry that was worth say $1 billion, but now that same industry could be automated and done cheaply then it is no longe worth $1 billion? In some ways this is beneficial to the economy in that has reduced the cost of this industry and so customers see a benefit of cheaper products, but how would it benefit say AI companies since they are not capturing a $1 billion industry, but instead sort of destroying a $1 billion industry? I am mainly looking at this from the perspective of the value of an AI company.. If they just keep commoditizing industries and there is ample competition between AI companies, then how is the AI company capturing value such that they would be worth as much money as people are saying they are worth (trillions of dollars)? I know that there is more to this, like freeing up this money and labor could lead to new industries etc., but how much of this will actually go to AI companies?
The related economic concept here is the Broken Window Fallacy. Doing 'redundant' work doesn't improve the economy, because there's an opportunity cost doing to anything. And replacing the need to "fix broken windows" or do now-AI-automated work would free up that opportunity cost. Which is how it's worked for nearly all automation. (Though Computers are subject to long running debate about this. See the Solow Productivity Paradox)
The gigawatt-sized asterisk on all of this is that AI companies have to capture that $1 billion. Anthropic has to ten-fold increase it's revenues or it goes bankrupt under it's current commitments. It has to increase it's revenues another 10 times to reach the $10,000,000,000,000/yr revenue required to support it's current valuation.
Anthropic put their Total Addressable Market at 35% of Global GDP (~$40 trillion/yr). They're planning to capture all of the money.
The second gigawatt-sized asterisk is that because of the fine handiwork of Mr. Huang, China has made very competitive AI models available for free, so the actual ability for AI companies to capture the value of the labour they displace ends up being quite limited. Thus a problem in the predictions: Something as to break, but it is as of yet unclear what that will be. Maybe Chinese AI will be banned, maybe Anthropic & OpenAI go bankrupt.
If the US actually wanted a world-wide ban, it would probably happen.
Obviously we can't effectively ban already released models, but a ban on models with >10T parameters for instance, is quite feasible. It takes a ton of power to train and operate those, and infrared satellites can spot that.
I'm confident China would be very happy to sign on to a world wide ban.
1. AI is one of the only high tech areas where America is ahead of China. If AI is knee-capped, China would indisputably become the world leader in industrial tech and high tech.
2. China is releasing models as open-weight to commoditize the complements. A ban on 10T models would let them specialize and utilize the lower parameter models to extend their lead in areas like robotics.
And once the US & China signed on I'm sure Europe and the middle powers would quickly agree.
It wouldn't, it would just mire the US in an unproductive and isolationist quagmire - reducing our global competitiveness because every US company would be paying an "anthropic tax" for a service that competing countries access for much lower costs.
The absurdity is the point; The clear-cut ways to resolve the tension of "AI giants need pricing power but don't have it" are absurd (admittedly, the former much more so than the latter)
It's hard to predict an outcome "somewhere in between of two absurdities"
This is such an interesting part of the whole AI equation.
The main benefit of the closed models is the regulatory control to say, insert guard rails against hacking things. That's also the main drawback as you now have handed control over this technology to a handful of companies, who will reap all the rewards of this captured sector.
It would be like the google/apple duopoly on steroids. Everyone forced to buy in to one of two ecosystems, with limited to no ability to undermine the duopoly via control of the device hardware. One can't make a real competitor because it requires access guarded by existing relationships and money. For the AI houses, it means you need approval from these groups to create in a competitive manner. That, IMO is too much power for an unelected organization to carry.
AI will likely only provide real benefits to humanity if it's open. But if it's open, it will also enable extreme harm in the form of cyber attacks, scams, and destruction of IP.
By keeping AI closed, it will bring a new form of imperialism. Instead of the core maintaining IP and the periphery suffering the secondary effects of industry, the core will maintain the capability and the periphery will be locked out of competition due to silly nation-state rivalries. This won't benefit humanity, it will lock us into the same cycle of misery for marginal gains measured in stock values.
So there's two ways AI can reasonably go: Cyber punk corporate overlords or cyber anarchy due to the inability to put a cork on the genie's bottle.
Both seem terrible to me. What good is space travel without freedom? What good is automation if there are incentives to maintain profit over stability? What good is the technology if any usage of it exposes you to fraud, violence, and blackmail at any moment?
Big fans of LLMs never ask "to what end?", they repeat shallow near term "wows" without considering the second and third order effects. "wow think about how good LLMs can spot cancer in scans!", great! but who will afford the scan without work? Will the overlords commit to luxury gay space communism? Or will they simply say "tough luck" and let that cancer fester?
Real AGI is incompatible with our current power structures, yet those current power structures desire the AGI. This cannot end well. Stagnation and stability would be preferable to the level of misery that would follow.
I generally agree, but the main reason I am pro AI (or at least feel it is necessary) is because I do think it will end up helping us solve other problems that I think are existential in themselves like climate change. Maybe it won't happen, but personally I think it will and don't want to go into more detail because some of it will probably seem far fetched but I do think it will happen now. And previously these problems seemed intractable or were only being solved too slowly for us to escape the worst outcomes. So now I think we have this situation where to solve some of our worst problems, we are now creating other problems...
AI could solve the planned economy problem and bring about the endless consumption for consumption's sake that's destroying the planet, but there isn't any incentive in it for anyone with a large enough stake in AI.
That's part of the reason why I really appreciate China just throwing a wrench into the closed model system. As long as there's a profit motive for AI, it will only accelerate our current problems.
Well it could, but planned economies are bad because they are not creative, which neither is AI... In fact the entire reason we are having this conversation is because companies doing unrelated things figured out how to do AI and pivoted to it in an unplanned economy, but yes China obviously would like to prevent the US from dominating in AI and it is in their (and perhaps everyone's) interest to release open models.
As it stands right now, AI is massively accelerating climate change, by consuming lots of energy and keeping fossil fuels alive and thriving. AI solving climate change is believing we'll invent new cows that are capable of reverse farting and do methane capture.
There is only one way AI will solve climate change. If it leads to job losses for 80% of the people, so consumption drops to nothing. Nobody drives anywhere, people stop buying things they don't need with money they don't have to impress people they don't like (because EMIs/BNPL need income). If everyone goes down to subsistence level, back to the feudal system, just 100 - 10K billionaires and everyone else low status barely surviving, TFR will drop to under 0.1, delivering the climate change solution.
Even if you ignore it impact of AI power consumption, just the opportunity cost is enormous in terms of climate change.
AI is what we’re doing instead of solving climate change. With the money the US has spent on AI in the last few years, you’d have approximately enough to fully remove all fossil fuels from current US electricity generation and replace them with renewables.
Climate change will be solved—if we do it soon—as soon as people start treating it as inevitable no matter how many dollars it costs the same way they do with AI. It will not be solved as long as we keep ignoring the existing, known solutions and hoping that some unrelated technological advancement will just happen to make it cost less to solve climate change.
Indeed. The trillions of dollars invested so far and the trillions to be invested in the next few years would've been enough to switch the entire world over to solar (batteries, EVs) as well as fund a a few dozen Manhattan scale problems - develop a printable (like fabric/paper) solar, organic solar (so it can be grown like yeast or bacterial growth). So MANY things could've been accomplished ... instead everyone is doubling down on meme stocks.
we could stop consuming so much and growing at all cost but no, we are being pitched the (idea|dream) that AI will come up with a technological fix to the problem that allows us to never change our behavior
Sorry can you clarify the 10T revenue part? Apple makes about 0.5T revenue and has 5X the valuation. Pricing on revenue multiple you’d say anthropic needs 100B in revenue, which at 16B last quarter isn’t too far fetched.
Apple's P/S above 10 is incredibly high. It reflects high margins, monopoly power, network effects and users with high switching costs. So it's probably not unjustified for Apple.
Anthropic has high margins and is working on the others but hasn't achieved them yet. It's margins are threatened as it and competitors are lowering prices regularly. P/S for commodity producers is well under 1. But lets assume tokens don't completely commoditize so a P/S of 1 is justified.
That's $1T. But Anthropic doesn't have $1T in revenue yet, and it's risky due to spending commitments. So you have to discount for both time and risk. That's where the other order of magnitude comes from.
Did you mean "commoditize your compliments"? The textbook example would be NVDIA giving away free AI models.
What you are probably referencing is "dumping". That's when you sell undercost to disrupt your competitors.
That may be why. Certainly the motivation of the CCP is supporting it isn't short term financial gain. But it's probably more about maintaining independence from a dependence on US AI long term than dominating the AI industry.
But I think it was originally just a side project to keep quants interested.
> Broken Window Fallacy. Doing 'redundant' work doesn't improve the economy.
The problem with AI is that work on an unprecedented scale will be made redundant and economic power will be concentrated into the hands of very few and there will be no way out in perpetuity.
Being able to buy massive intelligence for cheap (or running open weights on your own HW) doesn't give you economic power? Only if you don't use that power to provide something useful to others.
The Broken Window "Fallacy" assumes that distributional problems don't exist, lol.
I know why rich people who own many shares of aggregate economic capacity want me to believe that it is the only important thing and that sacrificing aggregate capacity to fix distributional problems is never worthwhile. But I don't believe it. The industrial revolution was gnarly with unnecessary distribution problems that ensured crippling poverty for the masses amid extraordinary leaps in aggregate capacity. This time around I don't want a century of Cyberpunk Oliver Twist before the peons are allowed a dignified life.
Adjacent, improved argument: taxes are a better fix for distributional problems than throwing stones at windows. I agree, and I will add that broken windows are a better fix for distributional problems than guillotines. I really really hope that we can convince capital to pay taxes because that's the only future where things don't go to monumental shit before getting better.
Exactly. Let's say there was one guy in the world who owned 99% of all the wealth. Is it really a bad idea to break his window so that he may disburse some money on getting it fixed?
Now, let's say the handyman whom the rich guy paid to fix the broken window uses that money to start a company that successfully competes with this 99% person?
If you generalize this, the conclusion is that broken window fallacy assumes that the pie needs to grow in a monotonically increasing fashion, when in reality sometimes the pie needs to shrink a little in order to grow a lot, if that shrinkage results in redistribution of wealth.
I am a big believer in the solow paradox and I don't think AI changes anything about it. Just more money down the tubes.
Go read the papers on it if you want to hear some of the immensely plausible theories as to why micros and the internet and all sorts of what we think is technological innovation doesn't show up in economic productivity data.
“Productivity isn't everything, but, in the long run, it is almost everything."
is one of the better quotes for grounding yourself when people start talking about economic problems to big to know exactly.
to your question.
If an industry is 'worth' 1 billion, meaning it generates 1b in revenue lets say at 800m of investment.
and that drops to 80m of investment and 100m of revenue due to productivity improvements then competition, the economy is now free to invest that extra 820m in the next best productive use for that capital, while still getting exactly the same services.
If the companies investing in ai can manage 80m of investment and 300m in revenue as the end state. They're richer, the consumers getting the same service cheaper, and society is richer due to the extra 700m being used to produce other goods and services. Thats the dream.
I think the simple answer is that it won't. Look to other tech revolutions for precedent. The biggest winners of the Industrial Revolution were not mechanical device peddlers, or even the 'super farms' that utilized those devices in an obvious way to increase efficiency/scale/etc. The biggest winners were instead the people who achieved previously unimaginable things on top of all of this.
And I think this (in our revolution) will also generalize to 'cognitive work' companies as a whole. This is why I think LLMs are going to radically (and positively) reshape the world, but in a way that relatively few seem to be predicting for now.
If labor income is decimated, there's a shrinking demand base for entrepreneurs using this technology to augment themselves. Maybe a select few make good money before the bottom falls out or get support from the remaining consumers of the physical economy (healthcare, etc).
There is a further cascading effect. Say that a specific $100b vertical is gone or severely reduced. In the worst case all the people working in that vertical are now unemployed. These same people now have no money or very little money to purchase or invest in other industries/verticals. Which means that there is a downturn in those verticals, leading to cost cuts and employee reductions. And those people now have less money to spend on other industries/verticals. And so on it goes.
But if the AI makes it cheaper to develop and produce the products, they can lower the cost, so those with less money can still afford them. I don't know if anyone knows how that plays out in the long run. Surely it must reach some sort of equilibrium.
> If AI disrupts an industry that previously had high wages and now can be automated by commodity AI products, would it not actually lower GDP in some way
Yes, in the short term that's true. In the long term, the concept is that that particular activity generates less GDP, but frees up the human capital who can then grow the economy.
A good previous example of this was the computerization of publishing --- the investment costs were balanced by letting employees go (though the compositors' union bargained for sinecures) and it took at least a decade for an increase in GDP to be seen as the folks were retrained in new work, and the publishers adapted to the lower costs by increasing the breadth of book titles which were greenlit and published.
Note that this was further complicated by a change in the tax law which made the backlist a liability rather than an asset.
Don’t forget that all the people that had the big salaries in this industry no longer have those salaries. That is money that they cannot spend that is removed from the economy.
Which is why other economic disruptors like rail roads eventually lead to economic expansions, and were net positive
It doesn’t not look like that is the case with AI. It’s going to cannibalize the economy and not bring the expansion of previous developments. Which is a sad state for humanity
Railroads disrupting a horse and wagon certainly would impact the salaries of wagon drivers.
AI isn’t a magical force that takes your job and prevents you from ever working elsewhere: it’s a technology that will increase productivity and impact jobs via creative destruction.
Could be. Going to be hard to scale current LLM architectures to the level ANT and OAI are promising. Small local models run with better user economics that throwing everything into hyperscaler land that have to pay off entire card stack re-capitalization every 18 months.
There is no real moat anymore unless they force it from the regulation side, hence the heavy lobbying.
Creative destruction assumes there’s a new job to find. That’s not necessarily the case and hasn’t been for decades. Like the security guard who works 80 hours a week and their rent is 55% of their pay being told “just move to management”.
Or like elevator operators, who used to make good money. There used to be 120,000 of them in New York - now there’s about 90 thousand elevator technicians, in the whole USA.
There’s not always a replacement job. Or there is, sweeping the floor.
60% of American employment in 2018 consisted of job specialties that did not exist in 1940, and “the locus of new-work creation has shifted from middle-paid production and clerical occupations over 1940–1980 to high-paid professional occupations and secondarily to low-paid services since 1980”.
There always is a replacement job, it’s not necessarily sweeping a floor, but it’s not 1:1.
People have to understand that your compensation is a function of the value you generate or capture: the security guard might work 80 hours a week, but a security camera works 168 hours a week and costs $100 on Amazon. He’s just not creating enough value over his alternative - the Puritanical thinking of hard work has inherent value isn’t compatible with the real economic world.
the current "real" economic world is the fakest manufactured system on the planet.
100 security guards cant make in one year what one person with 10 million dollars can "capture" from investing in the stock market.
The entire system is manipulated to benefit asset owners and disenfranchise the working class.
These types of lies and delusions are exactly where we are today. The "real" economy has been dying since 2007, everything else like new stock market ATHs are just lies the public has been fed that doesnt understand it just means: companies, banks, and governments are perpetually increasing debt and kicking the can down the road.
> AI isn’t a magical force that takes your job and prevents you from ever working elsewhere
Yeah I'd say if you were being paid six figures to do some office work thing (including software development) and now only 10% as many people are needed to get the same work done, it is very much going to take your high-paying job and make it considerably more difficult to find another one that pays the same. At a minimum, there's going to be a whole lot more candidates to compete with as you attempt to work "elsewhere."
Unless it's a _very_ slow transition to whatever the "next thing" that "can only be done by humans" pops up, this "creative destruction" is going to be devastating.
I remember when I was a young man, meeting a socialist who seemed very incensed about the coal miners.
I just couldn't understand it. If society no longer needed coal mining, was he seriously proposing they should all keep their jobs anyway? I couldn't take in this way of seeing the world.
So I'm pretty sanguine about the same thing happening to my own career, if it does. In the end, you can only expect to get paid for providing a service society needs. I can't think of any other remotely sensible way of running things, myself.
You described an economy, not a society.
Economies exist to support society, not the other way around. No, it can’t all be sunshine and lollipops for everyone - things shift and what was once a solid career disappears. This happens all the time. But a society (or maybe a functioning society) doesn’t throw people away if they aren’t bringing in a paycheque every month. Which was at least part of what the miners strike was about. And every other iteration of the same thing back to the Luddites
And importantly, we've grown so huge (population wise) that not everyone needs to provide a service in the first place, even if you can.
We need to move beyond the "everyone must contribute something" mindset, and instead place value on simply life itself instead of what that living being provides or does not provide because there simply isn't room for everyone to provide some service to society, everything is covered. The things that are not covered are generally high skill and high education (expensive) positions that not everyone has the ability to fulfill anyway.
Basically, and before we have mass unemployment and homelessness from AI, we need some sort of universal basic living floor as a safety net. Make sure no person is able to go without housing, food, utilities, education and healthcare.
Either there's a moral right to earn money through whatever skills one can offer, no matter how productive. Or there's a moral right to free money.
Anything else, in the long term, amounts to murder through abandonment. It's no different than leaving a wounded person in a forest while the group marches ahead, even though the group has plenty of food and a stretcher and a car, and fit, healthy people to lift the stretcher into the car.
please explain what service managers and executives provide to society
No one is saying bring back coal mines or CRUD software jobs. What people are saying is that the working class is being ELIMINATED from the economy. If you are in the working class your options are: service/construction work. Be an electrician, or a waiter, or lawn care professional. And they make jack shit for money. While the managers of AI make millions, so the economy shifts to them, away from people who do labor. But dont forget they all still have to pay taxes which fund all of the massive debt for AI and wars and all other BS.
You cant think of a remotely sensible way of running things because you dont understand what is happening. Society is COLLAPSING. The managerial class is systematically eliminating the working class. So enjoy your pretentious outdated theories about capitalists and socialists. While the world around you decays for the next 5 decades. I assume you're a bit older because (im in my 40s) thats the only people I see with this delusional view of society, they simply dont give a shit about the future.
This is what happened to manufacturing and the advice offered to those folks at the time was: "learn to code". The modern equivalent would be "learn to take care of old people", because that market is going to be booming for the next decade or more.
> because that market is going to be booming for the next decade or more
Well guess what, unless you're willing to be paid less than Haitian immigrants to do that work, it ain't gonna be booming for _you_. Ditto for construction, landscaping, et al. You know, those jobs Americans "won't do."
…because 'old people' clearly all have generational wealth on the order of your typical AI tech billionaire and can hire servants and caretakers at white-collar salaries.
That market is absolutely not going to be booming. It's a societal problem that large numbers of old people are exploited, broke, and brainwashed by generations of social media. That's not a booming market, that's a captive army of peons valuable only as long as human votes matter. They could as easily be disposed of, and so could you.
> …because 'old people' clearly all have generational wealth on the order of your typical AI tech billionaire and can hire servants and caretakers at white-collar salaries.
Yes, all those unemployed software developers who used to make multi-six figure incomes are now 'free' to compete in what will certainly be a ruthlessly competitive job market with all the other white collar professionals who were automated away. Likely into much lower paying service jobs, as those will be one of the few areas of the economy where the 'human touch' is still valued.
This reads like the line 'The rich man and the homeless are free to beg for change'
If they turn a non-AI $1Billion industry in to an AI $50million industry then they have destroyed that industry, made $50 million themselves, and the people who were previously spending $1Billion between them are collectively $950million better off, which they will then get to spend elsewhere.
Ya but they won't necessarily capture the $50 million themselves since there are already several companies that have comparable models that could all likely perform the same tasks, so at best they would get a fraction of the $50m? And if anyone can run an open model on their GPU then the barrier to entry to compete in this industry would be so low as to be worth almost nothing? Seems like a race to the bottom, like ultra efficient competition where the margins would be very low...
>the people who were previously spending $1Billion between them are collectively $950million better off, which they will then get to spend elsewhere.
It's generally worse for one man to have a billion dollars than for a 10,000 to have 100k. Those 10,000 people will spend that money where as the billionaire will almost certainly won't. An economy can only take so much 'investment'. Our economy depends on consumer spending.
If the top .01% get fabulously wealthy automating the bottom 90% out of good jobs, we will all be worse off for it.
> I am trying to understand something about the economy and AI. If AI disrupts an industry that previously had high wages and now can be automated by commodity AI products, would it not actually lower GDP in some way (initially anyway)?
Pre-AI, if you spent $X you got A units of output; post-AI:
* if you need 'only' A units, you will spend less than $X (technically lower GDP)
* but if the limiting factor was not the desired units, but your budget, you will continue to spend $X but get >A units (same GDP)
Additionally, if each unit is less expensive, this allows the end-customers to spend less on it, and reallocate some of their budget to other things, which would keep the overall GDP the same.
This is true for any economy, not just AI and the central contradiction of capitalism.
Automation raises profits for a brief moment until everyone else has caught up.
Once the new profit margin equalizes across an industry, competition sets in and margins shrink.
To keep margins up, labour costs need to fall further. That is why companies like Nike or Adidas produce in low-income countries. The investments in automation is such that profits can only be maintained through mass production, low wages, long working hours and raised demand through brand lock-in, classic marketing.
The current open bet is that equalization of AI technology can't be achieved across industries because of the important entry barrier: capital and compute. Monopolization should hold long enough to make the rest of the world market dependent on US AI infrastructure.
The rest will feel the devaluation of their work, they will get poorer - some straight away, some indirectly because of the shrinking customer base.
It will probably mean the AI companies will get part of that wealth from service companies in the US and pull it away from countries & companies exporting those services. Those service industries serve already "real" industries like: agriculture, construction, factories, governments, health, etc. For some countries with a service oriented industry it could be troubling. It will probably also lead to more software development for a while, but some wealth shift to AI companies will happen.
Historically, driving down the price of a product/service drove up consumption of the product/service. So markets will tend to grow to meet demand up to a saturation point.
In some markets, AI driving down costs will drive up consumption. So people won't necessarily lose their high paying jobs, they'll just do more work. Or, at least, corporations will capture the same revenue.
However, what's really untested is what the commoditization of knowledge is going to do to society. Having an agent at your fingertips with essentially a masters degree in every subject on earth, and the ability to apply that knowledge effectively is a pandora's box.
The closest analog is what industrialization did to craftsmen. How are people going to be qualified for jobs when AI can do a better job than anyone? But I suppose that's a political issue more so than a technical/economic one.
Anthropic (and OpenAI's) value prop is to replace the labor for a cost. Their endgame is diverting the employee wages to them. And eventually the underlying business perhaps.
>If AI disrupts an industry that previously had high wages and now can be automated by commodity AI products, would it not actually lower GDP in some way
If it disrupted that industry, the things produced by that industry would become cheaper (if AI didn't make it cheaper than human labor then it wouldn't have disrupted the industry in the first place), boosting everyone's purchasing power for that thing. GDP calculations try to account for this, so GDP would go up (the same way GDP calculations try to account for inflation; a single product class becoming cheaper is like localized deflation).
goods become cheaper but labor income takes a hit too. you only benefit if you've got savings / investments but even then, stocks would tank if collective purchasing power goes down fast enough.
I've heard this framed in terms of "surplus value," a very old (Marxist) economic theory. The debate is whether AI, which is intrinsically a machine, is capable of generating it - machines classically _transfer_ value. The reason this is important is because surplus value is what a capitalist entrepreneur exploits (whether they realize it or not).
So we don't actually know whether it is good for the economy, I guess we'll find out after the bubble bursts.
I thought they needed to raise money because they had committed to contracts that require money to build out infrastructure to satisfy the requirements.
Pretty much. If you think you're worth trillions and AGI has the market cap of half the world why would you sell, there's enough liquidity from private capital still.
If you think you have a sophisticated LLM which is useful but ultimately a commodity, and your moat is about to dry up, then you're worth much less and want as much exit liquidity as you can get.
> If you think you're worth trillions and AGI has the market cap of half the world why would you sell
You're making the assumption that it's the company's decision to sell rather than the people who shoveled hundreds of billions into the endeavor and want to cash out.
can't believe I'm almost defending billionaires, but I don't think that "X billions now" is a bad choice against "probably 10X billions in some years". Even if the 10X was certain, what are many billions going to allow you to do that you couldn't do with just a few billions.
The mind of a billionaire is truly a perverted one. Whatever you think they think, theirs is certainly a trillion times more perverse.
The only guarantee you can take for certain is that a billionaire hates humanity and anything that is beautiful in this world, and will act accordingly.
Because it will take ages for you to unlock the $XX billions you invested if you wait for it the old fashioned way (read: Anthropic being profitable and repaying you), nevermind making profit on top of that.
If you create a machine that turns a dollar into 5, the last thing you would do is dillute ownership of the machine. You would use your formidable profits to expand operations.
The time now is to dump this into the public markets and pocket it.
Anthropic stock price going down has nothing to do with what you think about the future of AI.
If you think that the open weight models are just 6 months behind (And will eventually catch up as progress eventually plateaus), then even if you think AI will take every job in the world, etc...Anthropic does not end up as one of the 10 most valuable companies in the world in this scenario. Far from it.
Ask Claude itself how far behind it thinks Google and OpenAI's models are and it will tell you "probably a few months at most".
I think the idea that "stock price has nothing to do with what you think about the future of AI" cuts both ways. Beyond a point I really don't think Anthropic's model performance is a strong impactor on their revenue, in much the same way that the quality of Outlook beyond a point does not seem to impact Microsoft's ability to sell M365 subscriptions. If its good enough, the momentum + network effects of the B2B corporate coordination systems they build on top of the model are all that really matter to justify their current valuation.
The problem is that this is only the case because you can either buy Anthropic or nothing, because the only other pureplay AI competition (OAI) decided to postpone the IPO. That is where the current open valuation comes from. The top models are more or less fungible and there is no way to capture the b2b side like MS, if the safeguard bullshit or some other stuff continues to act up the switch is pretty easy.
Not only that we are one model innovation away from them being worthless. You need $1T in compute because transformers are so ungodly inefficient. That compute will still be useful even if we find a more efficient architecture, but the margin on it will be so much less.
>That compute will still be useful even if we find a more efficient architecture
Mathematically there's no way to have lossless attention without O(n^2) cost; some magical massively more efficient architecture is not going to pop up the same way nobody's going to find a way to sort an arbitrary list in log(n) time. There may still be more opportunities for incremental gains (like using the bare-minimum number of full-quadratic-attention layers necessary), but they're not going to be orders of magnitude more efficient without sacrificing intelligence.
Counter proof by existence - humans use significantly less energy and don’t do intelligence through an n^2 process. The human brains seem to operate closer to SSMs like Mamba with a recurrent architecture and attention is implemented using sparse distributed weights. So yes, it’s possible to have very high intelligence without O(n^2). Is it as good as transformers? Hard to tell right now.
Well yesterday the news reported that Claude is basically worth $2T (they later fixed that headline), so surely something worth 2'000'000'000'000 USD is to be taken seriously? /s
I'm merely pointing out the discrepancy between tech insiders not trusting Claude's output and public markets briefly valuing it at $2T (now only $1.14T).
I didn't take the $2T valuation serious, hence the /s marker.
Ask Claude how long it thinks it will take to build anything and the answer is bananas. It will say things like it will take 5 weeks to build a what ever app, and then build that app in an hour. Timelines from Claude are basically a waste of tokens.
Anthropic has been on a nose dive since the start of trading. At this point almost everyone that bought ANT since the start of trading yesterday has lost money.
Thats not exactly helping to shake the reputation that this IPO was an attempt to offload this thing to “dumb money” and GTFO.
IPOs historically heavily underperform hence the saying "It' Probably Overpriced".
The market is selling off today so high beta stocks fall more than the average.
What I want to say is that the move is expected and has nothing to do with dumb or smart money in ANT specifically.
> The market is selling off today so high beta stocks fall more than the average.
Alternatively, ANT dropped because its value dropped, contributing to the market being down, both directly and indirectly. People thinking "if anthropic isn't worth as much as we thought, other stuff probably isn't either".
You really over estimate how much "people" matter in the stock market. You know what matters? Indexes, 401k, wealth funds, large institutions. You know what doesn't matter? A bunch of people buying/selling ANT. Look at every single IPO that was popular every, everyone of them has a price action of GME. They eventually get indexed, just as ANT will and the float will increase and the price will probably still look closer to TSLA, because it's a volatile industry based on "possibilities" rather then "probabilities".
I have also heard of the IPO dip you are mentioning. I would want to see some historical data to determine if this sell off after IPO is normal or abnormal. The decline doesn't seem to fit the pattern I recall being discussed but I don't know about the rebound time. I think their debt commitments are just out of sync with their income. They are looking at exponential increase in capability, exponential increases in compute needs and just tried to eye-ball too many exponentially growing trends and are now in a pickle. I think they assumed they would have created such value by now with some novel discovery, that they would be able to cover their checks. They are pretty close, just a few dozen billion dollars or so, so like if they could figure out how to capture just like 10% of a few markets with their tech, or get some ANT-drug blockbuster on the market, they should be golden. I think they have a hard deadline in front of them and are working hard to beat it.
This is how the rich get richer. I bet Berkshire Hathaway is going to do very well. If I was rich I would definitely start a diversified DCA buy on everything I could.
I really hope this Ponzi/Bubble runs out of money sooner rather than later. The longer it takes the higher the drop will be.
Looking objectively a valuation of 100B would be a long shot.
Not only is Anthropic loosing the lead in programming (I find the latest opus significantly worse in instruction following - it's better programming and reverse engineering capability doesn't offset it), it failed to find anyone willing to pay outside this market niche.
Don't get me wrong. As someone who spends 180eur a month with Anthropic and having subscriptions with all other providers I'm a big fan of AI. But I hate all the lying, the false hype, and most of everything the circular financing and that they chose to use the money from it to attempt to monopolise compute and reverse the "home pc" revolution.
The commoditization of knowledge work means that economies need to reorient themselves to producing physical goods/services (selling propane or wiping old people's asses, for example).
It also means big trouble for service based developed countries, who try to outsource all the physical drudgery.
In the IPO thread yesterday, when the markets were down and ANT up(from IPO price), lots of people were claiming the drop was due to money moving out of other stocks and into ANT. Boy were they wrong.
A year ago most of my friends (outside of tech) have barely heard of Anthropic, now everyone's using Claude. The pace of these changes is both fascinating and scary.
They probably figured they need to show path to profitability first. Anthropic (relatively) fared better and went for it.
As SpaceX, it had a multi variable narrative with space stuff. If it doesn't turn well for them, it isn't even clear OpenAI will get a chance to IPO with any hope. That's my bingo card.
Weren't there rumors that OpenAI's finances are such a mess they were struggling to clean up their accounting enough to file an S-1? I was trying to find a source for that, I can't remember if I read it here or somewhere else.
> In the past, OpenAI’s non-profit status has majorly limited its need to release such financial information. On its latest Form 990 return, used to disclose returns for such companies, Fontanello, Duffield, & Otake, a small San Francisco-based accountancy firm, is listed as the paid preparer — that would appear to be an accountant, rather than an auditor.
> The form says, oddly, that an independent accountant was not used to compile the financial statements, but that an independent accountant did audit the statements
It's a private company all we have is clues. When the CFO corrects the CEO in public, by a large margin, one can imagine how much of a mess it must be.
> It looks like a bad time to be an OpenAI [...] employee...
Apart from being one of the most privileged people on the planet of course. Still incredibly lucky to have ridden the gravy train, regardless of how long.
I am trying to understand something about the economy and AI. If AI disrupts an industry that previously had high wages and now can be automated by commodity AI products, would it not actually lower GDP in some way (initially anyway)? What I mean is, if you had an industry that was worth say $1 billion, but now that same industry could be automated and done cheaply then it is no longe worth $1 billion? In some ways this is beneficial to the economy in that has reduced the cost of this industry and so customers see a benefit of cheaper products, but how would it benefit say AI companies since they are not capturing a $1 billion industry, but instead sort of destroying a $1 billion industry? I am mainly looking at this from the perspective of the value of an AI company.. If they just keep commoditizing industries and there is ample competition between AI companies, then how is the AI company capturing value such that they would be worth as much money as people are saying they are worth (trillions of dollars)? I know that there is more to this, like freeing up this money and labor could lead to new industries etc., but how much of this will actually go to AI companies?
The related economic concept here is the Broken Window Fallacy. Doing 'redundant' work doesn't improve the economy, because there's an opportunity cost doing to anything. And replacing the need to "fix broken windows" or do now-AI-automated work would free up that opportunity cost. Which is how it's worked for nearly all automation. (Though Computers are subject to long running debate about this. See the Solow Productivity Paradox)
The gigawatt-sized asterisk on all of this is that AI companies have to capture that $1 billion. Anthropic has to ten-fold increase it's revenues or it goes bankrupt under it's current commitments. It has to increase it's revenues another 10 times to reach the $10,000,000,000,000/yr revenue required to support it's current valuation.
Anthropic put their Total Addressable Market at 35% of Global GDP (~$40 trillion/yr). They're planning to capture all of the money.
The second gigawatt-sized asterisk is that because of the fine handiwork of Mr. Huang, China has made very competitive AI models available for free, so the actual ability for AI companies to capture the value of the labour they displace ends up being quite limited. Thus a problem in the predictions: Something as to break, but it is as of yet unclear what that will be. Maybe Chinese AI will be banned, maybe Anthropic & OpenAI go bankrupt.
> Maybe Chinese AI will be banned, maybe Anthropic & OpenAI go bankrupt.
Unless the entire world bans it, how would that help?
If the US actually wanted a world-wide ban, it would probably happen.
Obviously we can't effectively ban already released models, but a ban on models with >10T parameters for instance, is quite feasible. It takes a ton of power to train and operate those, and infrared satellites can spot that.
I'm confident China would be very happy to sign on to a world wide ban.
1. AI is one of the only high tech areas where America is ahead of China. If AI is knee-capped, China would indisputably become the world leader in industrial tech and high tech.
2. China is releasing models as open-weight to commoditize the complements. A ban on 10T models would let them specialize and utilize the lower parameter models to extend their lead in areas like robotics.
And once the US & China signed on I'm sure Europe and the middle powers would quickly agree.
It wouldn't, it would just mire the US in an unproductive and isolationist quagmire - reducing our global competitiveness because every US company would be paying an "anthropic tax" for a service that competing countries access for much lower costs.
I agree, but the US would likely coerce the EU to do the same. So they still could outcompete some countries.
The US has burned a lot of goodwill with the EU. Though the threat of Russia might still be worth something.
How do you spell “unproductive and isolationist quagmire”?
U-S-A
The absurdity is the point; The clear-cut ways to resolve the tension of "AI giants need pricing power but don't have it" are absurd (admittedly, the former much more so than the latter)
It's hard to predict an outcome "somewhere in between of two absurdities"
This is such an interesting part of the whole AI equation.
The main benefit of the closed models is the regulatory control to say, insert guard rails against hacking things. That's also the main drawback as you now have handed control over this technology to a handful of companies, who will reap all the rewards of this captured sector.
It would be like the google/apple duopoly on steroids. Everyone forced to buy in to one of two ecosystems, with limited to no ability to undermine the duopoly via control of the device hardware. One can't make a real competitor because it requires access guarded by existing relationships and money. For the AI houses, it means you need approval from these groups to create in a competitive manner. That, IMO is too much power for an unelected organization to carry.
AI will likely only provide real benefits to humanity if it's open. But if it's open, it will also enable extreme harm in the form of cyber attacks, scams, and destruction of IP.
By keeping AI closed, it will bring a new form of imperialism. Instead of the core maintaining IP and the periphery suffering the secondary effects of industry, the core will maintain the capability and the periphery will be locked out of competition due to silly nation-state rivalries. This won't benefit humanity, it will lock us into the same cycle of misery for marginal gains measured in stock values.
So there's two ways AI can reasonably go: Cyber punk corporate overlords or cyber anarchy due to the inability to put a cork on the genie's bottle.
Both seem terrible to me. What good is space travel without freedom? What good is automation if there are incentives to maintain profit over stability? What good is the technology if any usage of it exposes you to fraud, violence, and blackmail at any moment?
Big fans of LLMs never ask "to what end?", they repeat shallow near term "wows" without considering the second and third order effects. "wow think about how good LLMs can spot cancer in scans!", great! but who will afford the scan without work? Will the overlords commit to luxury gay space communism? Or will they simply say "tough luck" and let that cancer fester?
Real AGI is incompatible with our current power structures, yet those current power structures desire the AGI. This cannot end well. Stagnation and stability would be preferable to the level of misery that would follow.
I generally agree, but the main reason I am pro AI (or at least feel it is necessary) is because I do think it will end up helping us solve other problems that I think are existential in themselves like climate change. Maybe it won't happen, but personally I think it will and don't want to go into more detail because some of it will probably seem far fetched but I do think it will happen now. And previously these problems seemed intractable or were only being solved too slowly for us to escape the worst outcomes. So now I think we have this situation where to solve some of our worst problems, we are now creating other problems...
AI could solve the planned economy problem and bring about the endless consumption for consumption's sake that's destroying the planet, but there isn't any incentive in it for anyone with a large enough stake in AI.
That's part of the reason why I really appreciate China just throwing a wrench into the closed model system. As long as there's a profit motive for AI, it will only accelerate our current problems.
Well it could, but planned economies are bad because they are not creative, which neither is AI... In fact the entire reason we are having this conversation is because companies doing unrelated things figured out how to do AI and pivoted to it in an unplanned economy, but yes China obviously would like to prevent the US from dominating in AI and it is in their (and perhaps everyone's) interest to release open models.
As it stands right now, AI is massively accelerating climate change, by consuming lots of energy and keeping fossil fuels alive and thriving. AI solving climate change is believing we'll invent new cows that are capable of reverse farting and do methane capture.
There is only one way AI will solve climate change. If it leads to job losses for 80% of the people, so consumption drops to nothing. Nobody drives anywhere, people stop buying things they don't need with money they don't have to impress people they don't like (because EMIs/BNPL need income). If everyone goes down to subsistence level, back to the feudal system, just 100 - 10K billionaires and everyone else low status barely surviving, TFR will drop to under 0.1, delivering the climate change solution.
Even if you ignore it impact of AI power consumption, just the opportunity cost is enormous in terms of climate change.
AI is what we’re doing instead of solving climate change. With the money the US has spent on AI in the last few years, you’d have approximately enough to fully remove all fossil fuels from current US electricity generation and replace them with renewables.
Climate change will be solved—if we do it soon—as soon as people start treating it as inevitable no matter how many dollars it costs the same way they do with AI. It will not be solved as long as we keep ignoring the existing, known solutions and hoping that some unrelated technological advancement will just happen to make it cost less to solve climate change.
Indeed. The trillions of dollars invested so far and the trillions to be invested in the next few years would've been enough to switch the entire world over to solar (batteries, EVs) as well as fund a a few dozen Manhattan scale problems - develop a printable (like fabric/paper) solar, organic solar (so it can be grown like yeast or bacterial growth). So MANY things could've been accomplished ... instead everyone is doubling down on meme stocks.
we could stop consuming so much and growing at all cost but no, we are being pitched the (idea|dream) that AI will come up with a technological fix to the problem that allows us to never change our behavior
Sorry can you clarify the 10T revenue part? Apple makes about 0.5T revenue and has 5X the valuation. Pricing on revenue multiple you’d say anthropic needs 100B in revenue, which at 16B last quarter isn’t too far fetched.
Apple's P/S above 10 is incredibly high. It reflects high margins, monopoly power, network effects and users with high switching costs. So it's probably not unjustified for Apple.
Anthropic has high margins and is working on the others but hasn't achieved them yet. It's margins are threatened as it and competitors are lowering prices regularly. P/S for commodity producers is well under 1. But lets assume tokens don't completely commoditize so a P/S of 1 is justified.
That's $1T. But Anthropic doesn't have $1T in revenue yet, and it's risky due to spending commitments. So you have to discount for both time and risk. That's where the other order of magnitude comes from.
> China has made very competitive AI models available for free,
Is this a "commoditize your competitors" strategy when your motive isn't profit but disruption?
Did you mean "commoditize your compliments"? The textbook example would be NVDIA giving away free AI models.
What you are probably referencing is "dumping". That's when you sell undercost to disrupt your competitors.
That may be why. Certainly the motivation of the CCP is supporting it isn't short term financial gain. But it's probably more about maintaining independence from a dependence on US AI long term than dominating the AI industry.
But I think it was originally just a side project to keep quants interested.
> Broken Window Fallacy. Doing 'redundant' work doesn't improve the economy.
The problem with AI is that work on an unprecedented scale will be made redundant and economic power will be concentrated into the hands of very few and there will be no way out in perpetuity.
Being able to buy massive intelligence for cheap (or running open weights on your own HW) doesn't give you economic power? Only if you don't use that power to provide something useful to others.
That seems to be what US leaders want, but the other 96% of the world thinks differently.
You don’t know that.
I also don't know that the drunk driver going twice the speed limit and weaving from lane to lane will crash.
The Broken Window "Fallacy" assumes that distributional problems don't exist, lol.
I know why rich people who own many shares of aggregate economic capacity want me to believe that it is the only important thing and that sacrificing aggregate capacity to fix distributional problems is never worthwhile. But I don't believe it. The industrial revolution was gnarly with unnecessary distribution problems that ensured crippling poverty for the masses amid extraordinary leaps in aggregate capacity. This time around I don't want a century of Cyberpunk Oliver Twist before the peons are allowed a dignified life.
Adjacent, improved argument: taxes are a better fix for distributional problems than throwing stones at windows. I agree, and I will add that broken windows are a better fix for distributional problems than guillotines. I really really hope that we can convince capital to pay taxes because that's the only future where things don't go to monumental shit before getting better.
Exactly. Let's say there was one guy in the world who owned 99% of all the wealth. Is it really a bad idea to break his window so that he may disburse some money on getting it fixed?
Now, let's say the handyman whom the rich guy paid to fix the broken window uses that money to start a company that successfully competes with this 99% person?
If you generalize this, the conclusion is that broken window fallacy assumes that the pie needs to grow in a monotonically increasing fashion, when in reality sometimes the pie needs to shrink a little in order to grow a lot, if that shrinkage results in redistribution of wealth.
But the government could just print the money and hand it out as a UBI, without anyone having to waste time fixing a broken window.
> The industrial revolution was gnarly with unnecessary distribution problems that ensured crippling poverty for the masses
Which basically only changed when people fought for their rights
> I really really hope that we can convince capital to pay taxes
I sadly can't see that happening without ww3
> Doing 'redundant' work doesn't improve the economy
Are we talking about the real economy or the stock market casino? I have a feeling one doesn't care about facts and logic
> Anthropic put their Total Addressable Market at 35% of Global GDP (~$40 trillion/yr). They're planning to capture all of the money
Oh, goody. They left room for their friends, Google and OpenAI.
SpaceX already claimed another 25% in its IPO, so they are left with only 40%.
We will inhabit Mars so there is a bigger pie to share /S
I am a big believer in the solow paradox and I don't think AI changes anything about it. Just more money down the tubes.
Go read the papers on it if you want to hear some of the immensely plausible theories as to why micros and the internet and all sorts of what we think is technological innovation doesn't show up in economic productivity data.
“Productivity isn't everything, but, in the long run, it is almost everything."
is one of the better quotes for grounding yourself when people start talking about economic problems to big to know exactly.
to your question.
If an industry is 'worth' 1 billion, meaning it generates 1b in revenue lets say at 800m of investment.
and that drops to 80m of investment and 100m of revenue due to productivity improvements then competition, the economy is now free to invest that extra 820m in the next best productive use for that capital, while still getting exactly the same services.
If the companies investing in ai can manage 80m of investment and 300m in revenue as the end state. They're richer, the consumers getting the same service cheaper, and society is richer due to the extra 700m being used to produce other goods and services. Thats the dream.
I think the simple answer is that it won't. Look to other tech revolutions for precedent. The biggest winners of the Industrial Revolution were not mechanical device peddlers, or even the 'super farms' that utilized those devices in an obvious way to increase efficiency/scale/etc. The biggest winners were instead the people who achieved previously unimaginable things on top of all of this.
And I think this (in our revolution) will also generalize to 'cognitive work' companies as a whole. This is why I think LLMs are going to radically (and positively) reshape the world, but in a way that relatively few seem to be predicting for now.
If labor income is decimated, there's a shrinking demand base for entrepreneurs using this technology to augment themselves. Maybe a select few make good money before the bottom falls out or get support from the remaining consumers of the physical economy (healthcare, etc).
There is a further cascading effect. Say that a specific $100b vertical is gone or severely reduced. In the worst case all the people working in that vertical are now unemployed. These same people now have no money or very little money to purchase or invest in other industries/verticals. Which means that there is a downturn in those verticals, leading to cost cuts and employee reductions. And those people now have less money to spend on other industries/verticals. And so on it goes.
But if the AI makes it cheaper to develop and produce the products, they can lower the cost, so those with less money can still afford them. I don't know if anyone knows how that plays out in the long run. Surely it must reach some sort of equilibrium.
> If AI disrupts an industry that previously had high wages and now can be automated by commodity AI products, would it not actually lower GDP in some way
Yes, in the short term that's true. In the long term, the concept is that that particular activity generates less GDP, but frees up the human capital who can then grow the economy.
A good previous example of this was the computerization of publishing --- the investment costs were balanced by letting employees go (though the compositors' union bargained for sinecures) and it took at least a decade for an increase in GDP to be seen as the folks were retrained in new work, and the publishers adapted to the lower costs by increasing the breadth of book titles which were greenlit and published.
Note that this was further complicated by a change in the tax law which made the backlist a liability rather than an asset.
Don’t forget that all the people that had the big salaries in this industry no longer have those salaries. That is money that they cannot spend that is removed from the economy.
Which is why other economic disruptors like rail roads eventually lead to economic expansions, and were net positive
It doesn’t not look like that is the case with AI. It’s going to cannibalize the economy and not bring the expansion of previous developments. Which is a sad state for humanity
Railroads disrupting a horse and wagon certainly would impact the salaries of wagon drivers.
AI isn’t a magical force that takes your job and prevents you from ever working elsewhere: it’s a technology that will increase productivity and impact jobs via creative destruction.
> AI isn’t a magical force.
It is the final piece in the re-feudalization/fascization(does this world even exist?).
We're back to feudal times where birth given rights to land, energy, resources, AI is all that matters, no more upper mobility.
Could be. Going to be hard to scale current LLM architectures to the level ANT and OAI are promising. Small local models run with better user economics that throwing everything into hyperscaler land that have to pay off entire card stack re-capitalization every 18 months.
There is no real moat anymore unless they force it from the regulation side, hence the heavy lobbying.
Creative destruction assumes there’s a new job to find. That’s not necessarily the case and hasn’t been for decades. Like the security guard who works 80 hours a week and their rent is 55% of their pay being told “just move to management”.
Or like elevator operators, who used to make good money. There used to be 120,000 of them in New York - now there’s about 90 thousand elevator technicians, in the whole USA.
There’s not always a replacement job. Or there is, sweeping the floor.
60% of American employment in 2018 consisted of job specialties that did not exist in 1940, and “the locus of new-work creation has shifted from middle-paid production and clerical occupations over 1940–1980 to high-paid professional occupations and secondarily to low-paid services since 1980”.
There always is a replacement job, it’s not necessarily sweeping a floor, but it’s not 1:1.
People have to understand that your compensation is a function of the value you generate or capture: the security guard might work 80 hours a week, but a security camera works 168 hours a week and costs $100 on Amazon. He’s just not creating enough value over his alternative - the Puritanical thinking of hard work has inherent value isn’t compatible with the real economic world.
the current "real" economic world is the fakest manufactured system on the planet.
100 security guards cant make in one year what one person with 10 million dollars can "capture" from investing in the stock market.
The entire system is manipulated to benefit asset owners and disenfranchise the working class.
These types of lies and delusions are exactly where we are today. The "real" economy has been dying since 2007, everything else like new stock market ATHs are just lies the public has been fed that doesnt understand it just means: companies, banks, and governments are perpetually increasing debt and kicking the can down the road.
> Or there is, sweeping the floor.
I got a robot to cover that one already.
> AI isn’t a magical force that takes your job and prevents you from ever working elsewhere
Yeah I'd say if you were being paid six figures to do some office work thing (including software development) and now only 10% as many people are needed to get the same work done, it is very much going to take your high-paying job and make it considerably more difficult to find another one that pays the same. At a minimum, there's going to be a whole lot more candidates to compete with as you attempt to work "elsewhere."
Unless it's a _very_ slow transition to whatever the "next thing" that "can only be done by humans" pops up, this "creative destruction" is going to be devastating.
I remember when I was a young man, meeting a socialist who seemed very incensed about the coal miners.
I just couldn't understand it. If society no longer needed coal mining, was he seriously proposing they should all keep their jobs anyway? I couldn't take in this way of seeing the world.
So I'm pretty sanguine about the same thing happening to my own career, if it does. In the end, you can only expect to get paid for providing a service society needs. I can't think of any other remotely sensible way of running things, myself.
You described an economy, not a society. Economies exist to support society, not the other way around. No, it can’t all be sunshine and lollipops for everyone - things shift and what was once a solid career disappears. This happens all the time. But a society (or maybe a functioning society) doesn’t throw people away if they aren’t bringing in a paycheque every month. Which was at least part of what the miners strike was about. And every other iteration of the same thing back to the Luddites
> In the end, you can only expect to get paid for providing a service society needs
It was a fair expectation when nearly everyone had a service to provide. But why now? No one asked to be born.
And importantly, we've grown so huge (population wise) that not everyone needs to provide a service in the first place, even if you can.
We need to move beyond the "everyone must contribute something" mindset, and instead place value on simply life itself instead of what that living being provides or does not provide because there simply isn't room for everyone to provide some service to society, everything is covered. The things that are not covered are generally high skill and high education (expensive) positions that not everyone has the ability to fulfill anyway.
Basically, and before we have mass unemployment and homelessness from AI, we need some sort of universal basic living floor as a safety net. Make sure no person is able to go without housing, food, utilities, education and healthcare.
To be clear I'd frikkin love free money so I can chill and do whatever I want. But I don't think I have any moral right to it.
Either there's a moral right to earn money through whatever skills one can offer, no matter how productive. Or there's a moral right to free money.
Anything else, in the long term, amounts to murder through abandonment. It's no different than leaving a wounded person in a forest while the group marches ahead, even though the group has plenty of food and a stretcher and a car, and fit, healthy people to lift the stretcher into the car.
Right but either that person provides for themself, or other people who also did not ask for them to be born have to provide for them.
Personally, the second option seems even less fair than the first one.
> Right but either that person provides for themself
How though? Just saying "do it yourself" doesn't make it so.
> or other people who also did not ask for them to be born have to provide for them
They have to personally provide for them? That's not quite it is it? Society as a whole is more productive. Automation provides for them.
> Personally, the second option seems even less fair than the first one.
Why? Consider the alternative.
please explain what service managers and executives provide to society
No one is saying bring back coal mines or CRUD software jobs. What people are saying is that the working class is being ELIMINATED from the economy. If you are in the working class your options are: service/construction work. Be an electrician, or a waiter, or lawn care professional. And they make jack shit for money. While the managers of AI make millions, so the economy shifts to them, away from people who do labor. But dont forget they all still have to pay taxes which fund all of the massive debt for AI and wars and all other BS.
You cant think of a remotely sensible way of running things because you dont understand what is happening. Society is COLLAPSING. The managerial class is systematically eliminating the working class. So enjoy your pretentious outdated theories about capitalists and socialists. While the world around you decays for the next 5 decades. I assume you're a bit older because (im in my 40s) thats the only people I see with this delusional view of society, they simply dont give a shit about the future.
so are you rich? or are you just planning to be homeless once your entire career field is wiped out?
This is what happened to manufacturing and the advice offered to those folks at the time was: "learn to code". The modern equivalent would be "learn to take care of old people", because that market is going to be booming for the next decade or more.
> because that market is going to be booming for the next decade or more
Well guess what, unless you're willing to be paid less than Haitian immigrants to do that work, it ain't gonna be booming for _you_. Ditto for construction, landscaping, et al. You know, those jobs Americans "won't do."
Except they blew all their money on cruises and AI generated images.
…because 'old people' clearly all have generational wealth on the order of your typical AI tech billionaire and can hire servants and caretakers at white-collar salaries.
That market is absolutely not going to be booming. It's a societal problem that large numbers of old people are exploited, broke, and brainwashed by generations of social media. That's not a booming market, that's a captive army of peons valuable only as long as human votes matter. They could as easily be disposed of, and so could you.
> …because 'old people' clearly all have generational wealth on the order of your typical AI tech billionaire and can hire servants and caretakers at white-collar salaries.
But they can vote for taxpayers to support it.
Yes, all those unemployed software developers who used to make multi-six figure incomes are now 'free' to compete in what will certainly be a ruthlessly competitive job market with all the other white collar professionals who were automated away. Likely into much lower paying service jobs, as those will be one of the few areas of the economy where the 'human touch' is still valued.
This reads like the line 'The rich man and the homeless are free to beg for change'
You’re not entitled to more free stuff because you previously had a high paying job. Divorce court rules don’t apply to the broader economy.
If they turn a non-AI $1Billion industry in to an AI $50million industry then they have destroyed that industry, made $50 million themselves, and the people who were previously spending $1Billion between them are collectively $950million better off, which they will then get to spend elsewhere.
Ya but they won't necessarily capture the $50 million themselves since there are already several companies that have comparable models that could all likely perform the same tasks, so at best they would get a fraction of the $50m? And if anyone can run an open model on their GPU then the barrier to entry to compete in this industry would be so low as to be worth almost nothing? Seems like a race to the bottom, like ultra efficient competition where the margins would be very low...
Oh yes, I don't expect Anthropic (for instance) to ever be able to make back the money they've spent.
>the people who were previously spending $1Billion between them are collectively $950million better off, which they will then get to spend elsewhere.
It's generally worse for one man to have a billion dollars than for a 10,000 to have 100k. Those 10,000 people will spend that money where as the billionaire will almost certainly won't. An economy can only take so much 'investment'. Our economy depends on consumer spending.
If the top .01% get fabulously wealthy automating the bottom 90% out of good jobs, we will all be worse off for it.
> I am trying to understand something about the economy and AI. If AI disrupts an industry that previously had high wages and now can be automated by commodity AI products, would it not actually lower GDP in some way (initially anyway)?
Pre-AI, if you spent $X you got A units of output; post-AI:
* if you need 'only' A units, you will spend less than $X (technically lower GDP)
* but if the limiting factor was not the desired units, but your budget, you will continue to spend $X but get >A units (same GDP)
Additionally, if each unit is less expensive, this allows the end-customers to spend less on it, and reallocate some of their budget to other things, which would keep the overall GDP the same.
except the budget goes down too because your leverage as a worker / avenues to well compensated work diminish
> about the economy and AI.
This is true for any economy, not just AI and the central contradiction of capitalism.
Automation raises profits for a brief moment until everyone else has caught up.
Once the new profit margin equalizes across an industry, competition sets in and margins shrink.
To keep margins up, labour costs need to fall further. That is why companies like Nike or Adidas produce in low-income countries. The investments in automation is such that profits can only be maintained through mass production, low wages, long working hours and raised demand through brand lock-in, classic marketing.
The current open bet is that equalization of AI technology can't be achieved across industries because of the important entry barrier: capital and compute. Monopolization should hold long enough to make the rest of the world market dependent on US AI infrastructure.
The rest will feel the devaluation of their work, they will get poorer - some straight away, some indirectly because of the shrinking customer base.
> Monopolization should hold long enough to make the rest of the world market dependent on US AI infrastructure.
Seems like the only way to avoid that is Chinese ai. Are they the good guys then?
It will probably mean the AI companies will get part of that wealth from service companies in the US and pull it away from countries & companies exporting those services. Those service industries serve already "real" industries like: agriculture, construction, factories, governments, health, etc. For some countries with a service oriented industry it could be troubling. It will probably also lead to more software development for a while, but some wealth shift to AI companies will happen.
Historically, driving down the price of a product/service drove up consumption of the product/service. So markets will tend to grow to meet demand up to a saturation point.
In some markets, AI driving down costs will drive up consumption. So people won't necessarily lose their high paying jobs, they'll just do more work. Or, at least, corporations will capture the same revenue.
However, what's really untested is what the commoditization of knowledge is going to do to society. Having an agent at your fingertips with essentially a masters degree in every subject on earth, and the ability to apply that knowledge effectively is a pandora's box.
The closest analog is what industrialization did to craftsmen. How are people going to be qualified for jobs when AI can do a better job than anyone? But I suppose that's a political issue more so than a technical/economic one.
Anthropic (and OpenAI's) value prop is to replace the labor for a cost. Their endgame is diverting the employee wages to them. And eventually the underlying business perhaps.
It would help to understand if you replaced “AI” with “mechanical loom” in your post.
>If AI disrupts an industry that previously had high wages and now can be automated by commodity AI products, would it not actually lower GDP in some way
If it disrupted that industry, the things produced by that industry would become cheaper (if AI didn't make it cheaper than human labor then it wouldn't have disrupted the industry in the first place), boosting everyone's purchasing power for that thing. GDP calculations try to account for this, so GDP would go up (the same way GDP calculations try to account for inflation; a single product class becoming cheaper is like localized deflation).
goods become cheaper but labor income takes a hit too. you only benefit if you've got savings / investments but even then, stocks would tank if collective purchasing power goes down fast enough.
The $1 billion is not “getting destroyed”, it’ll get spent somewhere else.
Economy is such a pseudo science that I don't think anyone can answer this. It is just vibes all around. Bioweapons btw
"Massive productivity increase that lowers the GDP" just can't happen.
> If AI disrupts an industry..
But it hasn't, and probably won't, so it's not even worth considering.
> In some ways this is beneficial to the economy
I've heard this framed in terms of "surplus value," a very old (Marxist) economic theory. The debate is whether AI, which is intrinsically a machine, is capable of generating it - machines classically _transfer_ value. The reason this is important is because surplus value is what a capitalist entrepreneur exploits (whether they realize it or not).
So we don't actually know whether it is good for the economy, I guess we'll find out after the bubble bursts.
Any of the frontier labs IPOing was a signal that the top has been reached.
I thought they needed to raise money because they had committed to contracts that require money to build out infrastructure to satisfy the requirements.
Pretty much. If you think you're worth trillions and AGI has the market cap of half the world why would you sell, there's enough liquidity from private capital still.
If you think you have a sophisticated LLM which is useful but ultimately a commodity, and your moat is about to dry up, then you're worth much less and want as much exit liquidity as you can get.
> If you think you're worth trillions and AGI has the market cap of half the world why would you sell
You're making the assumption that it's the company's decision to sell rather than the people who shoveled hundreds of billions into the endeavor and want to cash out.
Why would they decide differently than described?
they're myopic billionaires who don't care about anything beyond 6 months.
can't believe I'm almost defending billionaires, but I don't think that "X billions now" is a bad choice against "probably 10X billions in some years". Even if the 10X was certain, what are many billions going to allow you to do that you couldn't do with just a few billions.
The mind of a billionaire is truly a perverted one. Whatever you think they think, theirs is certainly a trillion times more perverse.
The only guarantee you can take for certain is that a billionaire hates humanity and anything that is beautiful in this world, and will act accordingly.
Maybe they want 10X the influence?
Regardless, the 2028 election cycle is shaping up to be one to behold. I fear for San Francisco democracy.
Because it will take ages for you to unlock the $XX billions you invested if you wait for it the old fashioned way (read: Anthropic being profitable and repaying you), nevermind making profit on top of that.
> hundreds of billions into the endeavor and want to cash out
They could have cashed out with one phone call
I don't think you're wrong but all investors keep this maxim in mind; the market can remain irrational longer than you can remain solvent
Obviously.
If you create a machine that turns a dollar into 5, the last thing you would do is dillute ownership of the machine. You would use your formidable profits to expand operations.
The time now is to dump this into the public markets and pocket it.
Anthropic stock price going down has nothing to do with what you think about the future of AI.
If you think that the open weight models are just 6 months behind (And will eventually catch up as progress eventually plateaus), then even if you think AI will take every job in the world, etc...Anthropic does not end up as one of the 10 most valuable companies in the world in this scenario. Far from it.
Ask Claude itself how far behind it thinks Google and OpenAI's models are and it will tell you "probably a few months at most".
I think the idea that "stock price has nothing to do with what you think about the future of AI" cuts both ways. Beyond a point I really don't think Anthropic's model performance is a strong impactor on their revenue, in much the same way that the quality of Outlook beyond a point does not seem to impact Microsoft's ability to sell M365 subscriptions. If its good enough, the momentum + network effects of the B2B corporate coordination systems they build on top of the model are all that really matter to justify their current valuation.
The problem is that this is only the case because you can either buy Anthropic or nothing, because the only other pureplay AI competition (OAI) decided to postpone the IPO. That is where the current open valuation comes from. The top models are more or less fungible and there is no way to capture the b2b side like MS, if the safeguard bullshit or some other stuff continues to act up the switch is pretty easy.
Not only that we are one model innovation away from them being worthless. You need $1T in compute because transformers are so ungodly inefficient. That compute will still be useful even if we find a more efficient architecture, but the margin on it will be so much less.
Deepseek was a sign of this on a smaller scale.
Can you expand on that deepseek part?
>That compute will still be useful even if we find a more efficient architecture
Mathematically there's no way to have lossless attention without O(n^2) cost; some magical massively more efficient architecture is not going to pop up the same way nobody's going to find a way to sort an arbitrary list in log(n) time. There may still be more opportunities for incremental gains (like using the bare-minimum number of full-quadratic-attention layers necessary), but they're not going to be orders of magnitude more efficient without sacrificing intelligence.
Counter proof by existence - humans use significantly less energy and don’t do intelligence through an n^2 process. The human brains seem to operate closer to SSMs like Mamba with a recurrent architecture and attention is implemented using sparse distributed weights. So yes, it’s possible to have very high intelligence without O(n^2). Is it as good as transformers? Hard to tell right now.
Why would anyone take Claude answer to that question seriously?
Because even the most biased, optimistic source proclaims doom of Anthropic, in the grandparent's view at least.
humans in aggregate are largely stupid and myopic.
here's the paradox de juor: https://mathparadoxes.com/ergodicity/
Aggregating beliefs and saying "this is true" works only in vert general ways.
If you can't take Claude's answer to that question seriously, that in itself is an indictment of Anthropic's valuation.
The same reason anyone would take any Claude answer to any question seriously.
There's nothing special about that question.
Well yesterday the news reported that Claude is basically worth $2T (they later fixed that headline), so surely something worth 2'000'000'000'000 USD is to be taken seriously? /s
> yesterday
Today down to $1.2T [0].
So you should change "is" to "was", or apply a factor of 0.6 to however seriously you thought it should be taken.
[0] https://shattered.io/anthropic-ipo-valuation-buyers-1-2-tril...
I'm merely pointing out the discrepancy between tech insiders not trusting Claude's output and public markets briefly valuing it at $2T (now only $1.14T).
I didn't take the $2T valuation serious, hence the /s marker.
Ask Claude how long it thinks it will take to build anything and the answer is bananas. It will say things like it will take 5 weeks to build a what ever app, and then build that app in an hour. Timelines from Claude are basically a waste of tokens.
Anthropic has been on a nose dive since the start of trading. At this point almost everyone that bought ANT since the start of trading yesterday has lost money.
Thats not exactly helping to shake the reputation that this IPO was an attempt to offload this thing to “dumb money” and GTFO.
The money is not lost until the share is sold.
This is the way
This is printed on the bags one's left holding.
The cat's not dead until we open the box. Schrödinger's share.
IPOs historically heavily underperform hence the saying "It' Probably Overpriced". The market is selling off today so high beta stocks fall more than the average.
What I want to say is that the move is expected and has nothing to do with dumb or smart money in ANT specifically.
> The market is selling off today so high beta stocks fall more than the average.
Alternatively, ANT dropped because its value dropped, contributing to the market being down, both directly and indirectly. People thinking "if anthropic isn't worth as much as we thought, other stuff probably isn't either".
You really over estimate how much "people" matter in the stock market. You know what matters? Indexes, 401k, wealth funds, large institutions. You know what doesn't matter? A bunch of people buying/selling ANT. Look at every single IPO that was popular every, everyone of them has a price action of GME. They eventually get indexed, just as ANT will and the float will increase and the price will probably still look closer to TSLA, because it's a volatile industry based on "possibilities" rather then "probabilities".
I have also heard of the IPO dip you are mentioning. I would want to see some historical data to determine if this sell off after IPO is normal or abnormal. The decline doesn't seem to fit the pattern I recall being discussed but I don't know about the rebound time. I think their debt commitments are just out of sync with their income. They are looking at exponential increase in capability, exponential increases in compute needs and just tried to eye-ball too many exponentially growing trends and are now in a pickle. I think they assumed they would have created such value by now with some novel discovery, that they would be able to cover their checks. They are pretty close, just a few dozen billion dollars or so, so like if they could figure out how to capture just like 10% of a few markets with their tech, or get some ANT-drug blockbuster on the market, they should be golden. I think they have a hard deadline in front of them and are working hard to beat it.
I wouldn’t characterize a 22% decline from the start of trading as “expected.”
There’s folks out there sitting on sizeable paper losses on Anthropic now.
Related content:
https://news.ycombinator.com/item?id=50573288 The Market Is Down (anihil.com) 41 comments | 3 hours ago
And of course...
https://news.ycombinator.com/item?id=50562754 Anthropic’s Nasdaq debut (reuters.com) 2216 comments | 21 hours ago
This is how the rich get richer. I bet Berkshire Hathaway is going to do very well. If I was rich I would definitely start a diversified DCA buy on everything I could.
I really hope this Ponzi/Bubble runs out of money sooner rather than later. The longer it takes the higher the drop will be.
Looking objectively a valuation of 100B would be a long shot.
Not only is Anthropic loosing the lead in programming (I find the latest opus significantly worse in instruction following - it's better programming and reverse engineering capability doesn't offset it), it failed to find anyone willing to pay outside this market niche.
Don't get me wrong. As someone who spends 180eur a month with Anthropic and having subscriptions with all other providers I'm a big fan of AI. But I hate all the lying, the false hype, and most of everything the circular financing and that they chose to use the money from it to attempt to monopolise compute and reverse the "home pc" revolution.
The time for the cloud will come as well.
What are you using that you like better than Opus? I'm currently using that with the VS code extension.
The commoditization of knowledge work means that economies need to reorient themselves to producing physical goods/services (selling propane or wiping old people's asses, for example).
It also means big trouble for service based developed countries, who try to outsource all the physical drudgery.
What kinds of goods and services? We have affordable general purpose robotics on the way, wont that commoditize manufacturing?
> wiping old people's asses
That should be worth $100/hr, right?
Arguably no worse than having to put up with corporate BS and Zoom calls all day.
Big buyers sold the "wet paper bags" of ANT stock to all the retails, same as always ...
IPO retail allocation was c. 20 perc. The rest of the company is owned mostly by other tech companies, some by employees.
I love how quickly this disappears off the front page despite its relevance and number of upvotes
In the IPO thread yesterday, when the markets were down and ANT up(from IPO price), lots of people were claiming the drop was due to money moving out of other stocks and into ANT. Boy were they wrong.
Apparently GRH doesn't translate to business value?
https://news.ycombinator.com/item?id=50574968
I just have sparking WWIII left on my bingo card for this regime of clowns in DC.
Not in this economy. Who can afford a war nowadays?!
The ruskies, seemingly.
It looks like a bad time to be an OpenAI late investor or employee...
How did they miss IPOing first? SpaceX being first and Anthropic second wasn't on my bingo card for 2026.
A year ago most of my friends (outside of tech) have barely heard of Anthropic, now everyone's using Claude. The pace of these changes is both fascinating and scary.
And a year from now Claude is unlikely to be dominant.
Yes, very true, with RSI/ASI there will not be a need for harnesses anymore.
Turns out Superbowl advertising works, who knew?
Anthropics moment was it's rebuke of the DoD
FTX imploding also drew some attention to Anthropic
After yesterday, OpenAI looks like they’re toast. When the current cash runs out it’s going to get very ugly there.
They probably figured they need to show path to profitability first. Anthropic (relatively) fared better and went for it.
As SpaceX, it had a multi variable narrative with space stuff. If it doesn't turn well for them, it isn't even clear OpenAI will get a chance to IPO with any hope. That's my bingo card.
Weren't there rumors that OpenAI's finances are such a mess they were struggling to clean up their accounting enough to file an S-1? I was trying to find a source for that, I can't remember if I read it here or somewhere else.
This isn't quite it https://www.ft.com/content/3cff198e-25e5-481a-bd34-e26941e1d...
> In the past, OpenAI’s non-profit status has majorly limited its need to release such financial information. On its latest Form 990 return, used to disclose returns for such companies, Fontanello, Duffield, & Otake, a small San Francisco-based accountancy firm, is listed as the paid preparer — that would appear to be an accountant, rather than an auditor.
> The form says, oddly, that an independent accountant was not used to compile the financial statements, but that an independent accountant did audit the statements
It's a private company all we have is clues. When the CFO corrects the CEO in public, by a large margin, one can imagine how much of a mess it must be.
Can you share more context regarding the correction?
It was one of the many interviews/podcasts, Sam throws a number for revenue, CFO was there and on the spot says, not really.
Can't find it, search show a more recent corrected revenue as they tend to count in unconventional ways that of course favors them.
https://finance.yahoo.com/markets/stocks/articles/openai-cor...
Sam thought that somehow waiting could get him a better price.
When he went to the banks last year and tried to IPO they blew off his price target, so he decided to stay private thinking he could pump it up more.
Generational whiff.
> It looks like a bad time to be an OpenAI [...] employee...
Apart from being one of the most privileged people on the planet of course. Still incredibly lucky to have ridden the gravy train, regardless of how long.
Everything is relative/comparative. Especially in comparison with personal expectations.
I imagine most of them were thinking 12-18 months ago they'd end up on top of the world. There's a chance they still will, but it's much lower now.