I wish I could be this optimistic. The reality is that as we speak the average consumer doesn't have access to cheap electricity or affordable inference hardware or powerful models that work out of the box on existing computers. The only realistic option for the 99% is to buy a $20-$200 monthly AI subscription, and I don't see that changing anytime soon.
I'm old enough to remember when people thought that anyone can build a website, so internet businesses will never have a moat.
"But like people who are good with computers, the models want a terminal, not some candy ass iPad UI."
Back before the iPhone I used to get into arguments with HCI specialists that phones could be like butlers and should know with all the sensors that they have that you put it in your bag and behaved accordingly. I was told that was impossible then but it seems more possible now. Had the world gone at all that way we'd have a freakin' API to make a restaurant reservation and wouldn't have to go through multimodal hell.
Where does this hypothetical API live? Within a walled garden? We almost have that already with delivery apps and some credit cards. They're awful precisely because they are rent seeking in the form of convenience fees and artificial scarcity.
What you call multimodal hell is what others call meaningful choice and market competition.
I'm not sure I buy the "everyone will be AI coding to replace things that cost money with their own apps" idea. I only have so much limited time in my day (and only so many tokens on my claude account per week). It's probably going to make more sense for me to buy a tool that's been given human attention over the span of weeks over something i prompt into existence in a few hours (especially if I need 10 such tools to accomplish something).
I can’t always see the personal appeal, however when I view through the lens of businesses that buy very expensive enterprise software and other SaaS products (maybe blending into consumer market), well I think they’re toast. I think the acceleration of AI tools recently isn’t going to be indicative of how long the full transformation will take, but a lot of companies will start preferring Build over Buy. I have no idea the scope, but this is already happening at some partial scale.
I agree the free money in like one month coded SaaS apps are in big trouble. But like there's no way I'm gonna have the vision to desire to play a game I made myself using AI for instance (just the fact that I prompted it into existence ruins some of the exploration of a game made by someone else). So at the low end of the extreme (easy to make SaaS apps with basic code and a db) AI is a thread, but at the other end of the extreme (requiring vision and where human attention is a bottleneck) there's definitely still tons of opportunity.
The choice isn't between A) the expensive, proven tool and B) the thing I promoted in a few hours - it's b/w A, B and also C) the less expensive, somewhat proven tool that someone else prompted over a couple of days. I can see, over time, a slow drift towards "free".
Factor in how a lot of tools have weaponized their interfaces against their users - then the motivation isn't just cost, but usability.
funny. people are probably more likely to know him as geohot here. and afaik he is ex-comma. a more likely motivation (if you wanted to suppose one) is that he sells local compute. https://tinygrad.org/#tinybox
The work on self-driving cars would not be done if there were not a way to profit from it. If it isn’t expected to earn more than it costs, it wouldn’t be done.
Now maybe it’s all being done because of expectations of a monopoly (this is a free market consequence, right?), but …
Im starting to believe that the biggest moats will be in the application layer, and that people are starting to realize traditional saas moats apply to those companies too
network effects, distribution, proprietary data, systems of record
companies like opencode have none of the above
cursor's distribution has been faltering and they're hard pivoting to training their own models with their proprietary data to try to build their moat back
The network effects and distribution moats are subject to the same erosion. proprietary data less so. Systems of record the least so. The real value-add over the next decades is gonna be around providing "stability". In multiple domains, from service stability to cultural stability. The "disruption" formula will be flipped on it's head, and ppl will be motivated to "move slow and fix things".
> Over the past fifty years, the U.S. economy built a giant rent-extraction layer on top of human limitations: things take time, patience runs out, brand familiarity substitutes for diligence, and most people are willing to accept a bad price to avoid more clicks. Trillions of dollars of enterprise value depended on those constraints persisting. – Citrini Research
Agentic commerce will render Amazon and the rest of the rent seeking marketplaces obsolete given enough time. Because LLMs can literally go straight to the seller and perform checkout, do market research to make sure the seller is legit, and the seller can sell for lower than on the marketplace since they aren’t paying a 15-20% cut.
First, I think there is value in the “rent seeking” Amazon marketplace because how else would the models “go straight to the seller”, another centralized search engine? Why not just use the Amazon one then?
Second, one of LLMs big weaknesses is judgement on what to trust. I would not trust the judgment of an LLM to determine “the seller is legit”… unless we outsource trust verification to a third party marketplace (who will want a cut).
Finally, OpenAI has been aggressively pushing for this so they can take a cut of the transaction. So it’ll just be another middle man.
Exactly this, and not just another middleman, a middleman with an obscene burn rate that isn't close to profitability and is incentivized to ratchet up prices as soon as they can.
And then AI procurement has problems on the buyer side. Do I just blindly trust that the model is going to make the purchase as specified? Do I trust the model's search capabilities and objectivity of returning results? How do I know that OpenAI isn't running its own "marketplace", only showing me options to buy that they want me to see while filtering out less desirable options for them?
It's a fundamentally less transparent experience than Amazon.
Who mentioned blockchain? You just need to allow e-commerce website to expose their storefront as a MCP app or UI and then clients can interact with them directly.
>Third party marketplaces sprung up for reservations, and idk it’s been a while since I went to fancy dinner, but I imagine the restaurants have just started charging. Or at least the first party reservation sites do.
Yes, this is what Tock is for. It's not clear to me that it's a bad thing. It replaces the old $20 in a handshake I used to do with the maitre d at the front of the restaurant. Democratizes opportunity and improves transparency
I think self driving cars are important for a couple of reasons:
1. Demographics - Aging population needs transportation. God knows we certainly don't need really old people driving themselves. We got a taste of the future in West Portal in that regard not long ago.
2. Human Capital - The US has pretty much demonstrated that there is little desire to import low skilled labor. Where do these theoretical Taxi drivers come from? Or welders or plumbers. Labor is going to become increasing expensive no matter how you slice the pie.
3. Younger US citizens are going to gravitate to non-manual labor jobs. It is not just that every one is being steered toward college. Physical labor (trades) take a toll on the body. I know - I have work in them - and you quickly extrapolate what that will be like when you are 50.
Public transportation solved this before cars were commonplace. Implement ubiquitous and free public transportation in every urban center (where 80% of the American population lives [0]) and you'll save billions from not having to manufacture (and store!) cars.
Methinks this post conflates “rent seeking” with “return on investment” just a tad.
Economic rent is the extra money you can charge for owning a scarce resource. ML models are not waterfront real estate, they are IP. Other people can make more models if they can/want to.
Now, whether IP should be legally protected is a totally separate question, and while we in the West tend to assume the answer is obvious geohot would certainly not be the first person to suggest broadly applying private property rights to information makes questionable sense.
This is a highly sensational take that is basically fan fiction. From "the era of purposefully frustrating humans is over", to "the added bonus of the collapse of the US economy. Frankly, it’s well deserved." and "everyone in the world is rooting for the Chinese models"; nothing of that is grounded in reality.
The Chinese models are open source because they are not state of the art. Once they catch-up or lead, they will likely close them down by a government mandate. Just like Meta was fine with Llama being open source but once they started to get close to OpenAI/Google/Anthropic, they shifted their language to "maybe we won't keep doing that."
The idea that AI will end the "rent-seeking class" that has effectively existed for thousands of years is... not going to happen! The business model just adjusts. And if AI is going to be an economy-shaping super disruptor, the cloud-hosted models will continue evolving beyond what you could ever run at home under the desk.
> The Chinese models are open source because they are not state of the art
I think geohot is burying the lead in this text in his post with a lot of speculation.
It's not not that these specific models will become closed it's that the hardware/hosting vendors have an incentive to train models where inference is custom tuned to their chip's dimensions and VRAM.
The Chinese models do a great job of showing what's capable on consumer/prosumer hardware because of export restrictions but anyone entering the hardware space has the same incentives to undercut the frontier labs so they can sell more hardware.
It's also not clear if being at the forefront of inference quality really matters. The open source models appear to be doing a fine enough job of keeping up even if they're a few months behind. So it seems like there's not much of a technology moat for these labs other than the capital costs of training/serving.
> the "rent-seeking class" that has effectively existed for thousands of years
This "rent seeking class" is not a historical universal, regardless of how much college marxists insist that it is. Leaders can be good or bad, and they hold power in different ways. In America today we have bad leaders (across the entirety of the political spectrum) - and AI poses a lot of challenges in how they hold power. This is not to say Chinese leaders are any better - but the way they hold power is not challenged by AI. Business models will indeed adapt - but the condition is excellent, as they say.
> The era of purposefully frustrating humans is over. The Chinese open source model running on the box under my desk can pass the Turing Test. When you call, e-mail, text, or show me an ad, you’ll never know if it’s me or my model seeing it.
But at some point, you're going to want to do something, like, e.g., buy something. Then you're right back to the problem in the opening quote:
> things take time, patience runs out, brand familiarity substitutes for diligence, and most people are willing to accept a bad price to avoid more clicks.
& we're already seeing AI used to do this. E.g., Amazon listings where product photos are AI generated. (… not that many product photos weren't "bad photoshop of product onto hot sexy model who is obviously not using our product" before … but now it's AI!) Whereas before someone would have had to spend a modicum of time badly using Photoshop, now AI can just churn out the same fraudulent result in a fraction of the time.
Now, if I have a problem with a product, instead of just calling a number, browsing a phone tree, getting put on hold, and finally having to struggle to get some human to understand the basic logistics of "I paid for X, I did not get X, I demand X or refund", I get to do all that but with the extra step of "forced engagement with an AI that is incapable of actually solving my problem". (This somehow still manages to apply even when the problem is seemingly trivial enough that I find myself thinking "… this actually should be something an AI can do" but inevitably, no, the AI is "sorry", it cannot do that.)
And besides, calls, emails, etc. are already handled without AI: I (and everyone I really care about) have either allowlisted all inbound comms, or abandoned the medium altogether. Moreover, any communications medium is useful because it is not infested with spam, and will eventually be destroyed by spam. At least until we grow laws for mediums like phone/email, maybe named things like "Do Not Call" or "CAN-SPAM" and those laws are enforced. But the GOP has no interest in enforcing any level of consumer protection, so here we are.
> Agentic commerce will render Amazon and the rest of the rent seeking marketplaces obsolete given enough time. Because LLMs can literally go straight to the seller and perform checkout, do market research to make sure the seller is legit, and the seller can sell for lower than on the marketplace since they aren’t paying a 15-20% cut.
Right.
You'll just end up paying the 15-20% cut to the people who train the model and keep it updated and run the agents that you rent from them.
To me it seems like the rent seeking is everywhere and getting strong in all facets of our lives. Tech companies, utilities, marketplaces, car companies, appliance companies - everyone is pushing subscriptions, gate keeping features and milking every possible dime.
It's undeniable that technology over the past decades has increased democratization in something every step of the way.
YouTube destroyed Hollywood's monopolization of entertainment. Anyone with a smartphone now has a shot at becoming a full-time creator. Prior to this, it was gate kept by Hollywood execs.
Smartphones destroyed Microsoft's monopolization of apps.
Not a leap to believe this will happen to some extent with AI (and it's already happening to some degree).
It seems like the rent seeking class is just moving to selling you access to LLMs in data centers by the token. In the past, the "rent-seeking class" being described here was at least part of the middle class. Now a few billionaires are going to capture all of the value, but the rent-seeking isn't going away.
Since the article is largely about open weights models, I think the argument is that this is the "last gasp" and soon doing inference at home will be common.
The small models that I can run at home are becoming more capable, and I have replaced some API-based tasks with local inference as they improved, but large open weights models are still a lot stronger. The nice thing with larger open weights models is that competing providers serve them at modest margins and prices. I don't have the hardware to run the largest Qwen models, but I can get API access at low cost. Since there are only modest barriers to new commercial inference providers for these models I'm not worried that API access to them will become drastically more expensive at some future time.
The trend over the last decades was towards more centralization and I don't see that changing. Unless we radically change our economic system, the rent seekers will always win. There will be probably less of them but they will be even bigger.
> The best anyone can hope for is a free market, with everything properly priced. But for decades, the American market has not been free. It’s used purposefully added friction to exploit a time asymmetry between the business and you. And due to things like call centers, this has been very profitable for the businesses.
I just think this analysis is wrong from the start. The "proper" pricing structure, the one tracking the actual costs involved, would be that you don't get to talk to a human being at all unless you pay for their time. Human frictions are what allow no-charge customer support to exist.
you do pay for their time, when you (or someone else) buys the company's products. but customer support exists because it makes the company money, so really they should be paying you
ah yes, capitalism is over because the chinese are benevolent people that just give away the goose that lays golden eggs out of the goodness of their hearts
this will continue forever and no rugs, chinese or otherwise, will ever be pulled
we know that because the label on the rug says "open source"
It's going to be the same asymmetry with different tooling. Why does he think people will be the only ones with these tools? Companies will utilize them, as they already are, and will also have more time to fine-tune, manage, and improve them.
It's going to be your bots vs theirs. Theirs will have more resources. Net result? probably fewer jobs and wealthier companies yet again.
> The best anyone can hope for is a free market, with everything properly priced. But for decades, the American market has not been free. It’s used purposefully added friction to exploit a time asymmetry between the business and you.
A free market that is "properly priced" is a not a real state of existence.
Resource and information asymmetry, and the exploitation by the those with resource and information privilege of those without it has been present from the very beginning. A free market is just a tool (among many) to achieve a goal for a society.
For some, that goal is explicitly the concentration of wealth and welfare in very few hands. This is oligarchy.
For others, it's the advantaging the welfare and dignity of their "tribe" at the cost of welfare and dignity to perceived outsiders.
And for yet others, it is the advancement of universal welfare and dignity.
Neither a free market nor socialism gets you any of these. What gets you there are the shared narratives that utilize tools like free markets, regulation, and redistribution.
This argument is a bit scattered. "Rent seeking" is being misused here. It's a relatively new term (~50 years old) but it has a lot of history behind it, specifically with enclosures. The Enclosure Acts [1] were a series of laws that took what was common property open to all and made them private property. This was embryonic capitalism [2].
Anyway, I remember that Google demo of making restaurant reservations. I believe it was scripted and had a human fallback. Little did we know that Google would drop the bag on the whole transformer thing that came out soon after. I wouldn't be surprised if it was some of the same people involved.
What the author is talking about isn't rent-seeking per se but a moat. The entire proposition of OpenAI is that they can build a moat and recoup the billions of investment. I'm not convinced that's true, which is part of the author's point, for some of the same reasons:
1. Cost of hardware and training and tokens keeps going down. We saw the same thing with Bitcoin mining. I wonder if we'll see ASICs enter the fray here too; and
2. China will make sure no one company owns this future. DeepSeek was a shot across the bow of OpenAI, Google and Anthropic. It is a national security issue for China.
Where I disagree is that this will be an end for the rent-seeking class. I think we're bouldering towards a dystopian future of even more wealth concentration where most people get displaced by automation and AI, which suppresses wages and ultimately leads to a situation where a handful of people have all the money and almost everyone else has no money.
> This argument is a bit scattered. "Rent seeking" is being misused here.
It's being used in a more literal-meanings-of-the-words sense ("pursuing monopoly rents") rather than the narrow economic term-of-art sense of "pursuing monopoly rents through influence over public policy by means that do not create, or which inhibit the creation of, additional wealth" (the definition you seem to be complaining about it not adhering to without actually providing.)
But most of the usages would also be correct in the narrower sense, because virtually ever actor referred to as rent-seeking in the broader sense are also rent-seeking in the narrow sense as part of that. (E.g., actively lobbying for "safety" regulation which would disproportionately impair non-incumbent new competitors.)
> What the author is talking about isn't rent-seeking per se
but a moat.
Pursuing a moat is just another term for seeking monopoly rents by any means, including rent-seeking in the narrow sense.
(There's also obviously an ideological angle in creating the term "rent seeking" as a term of criticism to those seeking monopoly rents through means that the creators of the term disapprove of, excluding seeking the same kind of rents by other means from "rent seeking".)
No, a moat is a competitive advantage/ OpenAI in particular is predicated on the belief that they will have a compeitive advantage. ASML is a compeitive advantage with EUV (for now). You can overcharge for if you have a compeititve advantage but that's not the same as rent-seeking.
Rent-seeking is fundamentally intermediation like a health insurer putting themselves between a patient and a healthcare provider or privatizing grazing lands or controlling water supplies from snowmelt (like the Resnicks) or privatizing trains in the UK.
Microsoft has a compeitive advantage with Windows, Google with its search engine and the ad business that funds it, Amazon with AWS or NVidia with GPUs. There are alternatives to all of these things but these companies maintain dominance with a combination of scale, cost, technology and network effects. That's not rent-seeking in a broad or narrow sense.
Rent-seeking would be Microsoft lobbying lawmakers to require schools and governments to purchase Windows, for example.
> The best anyone can hope for is a free market, with everything properly priced. But for decades, the American market has not been free. It’s used purposefully added friction to exploit a time asymmetry between the business and you. And due to things like call centers, this has been very profitable for the businesses. Cable companies and insurance rely on the fact that your time is more valuable than theirs. They can hire people in India at scale to waste your time. They can use procedure and big data to design protocols to drive you just to the point of frustration at little cost to them. How often do you diligently check Uber and Lyft and select the cheaper one?
> Enter AI, the great equalizer of time.
I didn't read any futher: this article is dumb. If a company has the capability to hire literal people to waste your time, they can deploy more AI than you to waste the time of your AI.
Or they just use price to limit access instead of time. Which means you're totally SOL if you have time but no money. Pay to win, that game everyone loves /s!
AI doesn't flatten asymmetries, it exacerbates them.
> If a company has the capability to hire literal people to waste your time, they can deploy more AI than you to waste the time of your AI.
I don't totally buy OP's argument, but I think you're dismissing it unfairly.
His point is that in the pre-LLM world, if a company wants to waste your time, they can hire a call center employee overseas for US$4/hr and make you wait for an hour to talk to them for 30 minutes. If you value your time at $80/hr, then the 90-minute call cost you $120, but it only cost the company about $2, thus the asymmetry.
OP's claim is that now, the asymmetry is gone. If both you and the company try to use AI, the company has less leverage to impose costs on you. They can deploy more AI to waste more of your time, but that means the asymmetry now is in the customer's favor because it costs the company more than the customer to get support.
At some point you were part of that group that you are labeling a mistake. At another point, your children are or will be if you have any. Locking all of these productivity tools in one generation is a recipe for failure. You should think outside the box that you are currently trapped in.
I wish I could be this optimistic. The reality is that as we speak the average consumer doesn't have access to cheap electricity or affordable inference hardware or powerful models that work out of the box on existing computers. The only realistic option for the 99% is to buy a $20-$200 monthly AI subscription, and I don't see that changing anytime soon.
I'm old enough to remember when people thought that anyone can build a website, so internet businesses will never have a moat.
"But like people who are good with computers, the models want a terminal, not some candy ass iPad UI."
Back before the iPhone I used to get into arguments with HCI specialists that phones could be like butlers and should know with all the sensors that they have that you put it in your bag and behaved accordingly. I was told that was impossible then but it seems more possible now. Had the world gone at all that way we'd have a freakin' API to make a restaurant reservation and wouldn't have to go through multimodal hell.
What's that adage about premature optimization?
Where does this hypothetical API live? Within a walled garden? We almost have that already with delivery apps and some credit cards. They're awful precisely because they are rent seeking in the form of convenience fees and artificial scarcity.
What you call multimodal hell is what others call meaningful choice and market competition.
I'm not sure I buy the "everyone will be AI coding to replace things that cost money with their own apps" idea. I only have so much limited time in my day (and only so many tokens on my claude account per week). It's probably going to make more sense for me to buy a tool that's been given human attention over the span of weeks over something i prompt into existence in a few hours (especially if I need 10 such tools to accomplish something).
"The economics of opportunity cost are unchanged" a friend told me recently, and I think that's exactly what is driving your intuition here as well.
I can’t always see the personal appeal, however when I view through the lens of businesses that buy very expensive enterprise software and other SaaS products (maybe blending into consumer market), well I think they’re toast. I think the acceleration of AI tools recently isn’t going to be indicative of how long the full transformation will take, but a lot of companies will start preferring Build over Buy. I have no idea the scope, but this is already happening at some partial scale.
I agree the free money in like one month coded SaaS apps are in big trouble. But like there's no way I'm gonna have the vision to desire to play a game I made myself using AI for instance (just the fact that I prompted it into existence ruins some of the exploration of a game made by someone else). So at the low end of the extreme (easy to make SaaS apps with basic code and a db) AI is a thread, but at the other end of the extreme (requiring vision and where human attention is a bottleneck) there's definitely still tons of opportunity.
The choice isn't between A) the expensive, proven tool and B) the thing I promoted in a few hours - it's b/w A, B and also C) the less expensive, somewhat proven tool that someone else prompted over a couple of days. I can see, over time, a slow drift towards "free".
Factor in how a lot of tools have weaponized their interfaces against their users - then the motivation isn't just cost, but usability.
Like self-driving taxis where the business model is to stop paying drivers so we can pay more to big tech companies. Viva la revolution!
I wonder how many people noticed that geohot is George Hotz, who authored the article on his GitHub page.
George Hotz runs Comma AI, a self driving car company.
https://comma.ai
funny. people are probably more likely to know him as geohot here. and afaik he is ex-comma. a more likely motivation (if you wanted to suppose one) is that he sells local compute. https://tinygrad.org/#tinybox
In a free market it would be stop paying drivers so everyone can pay less for taxis.
But the market is so unbelievably messed up that this is not what happens in practise.
Disagree.
The work on self-driving cars would not be done if there were not a way to profit from it. If it isn’t expected to earn more than it costs, it wouldn’t be done.
Now maybe it’s all being done because of expectations of a monopoly (this is a free market consequence, right?), but …
I think you made the assumption that gp is disagreeing with, that reducing costs necessarily reduces prices. (hence the bit about messed up markets)
The safety improvements seem pretty substantial, though.
Im starting to believe that the biggest moats will be in the application layer, and that people are starting to realize traditional saas moats apply to those companies too
network effects, distribution, proprietary data, systems of record
companies like opencode have none of the above
cursor's distribution has been faltering and they're hard pivoting to training their own models with their proprietary data to try to build their moat back
The network effects and distribution moats are subject to the same erosion. proprietary data less so. Systems of record the least so. The real value-add over the next decades is gonna be around providing "stability". In multiple domains, from service stability to cultural stability. The "disruption" formula will be flipped on it's head, and ppl will be motivated to "move slow and fix things".
Exactly this. Also global sales and marketing and support.
> Over the past fifty years, the U.S. economy built a giant rent-extraction layer on top of human limitations: things take time, patience runs out, brand familiarity substitutes for diligence, and most people are willing to accept a bad price to avoid more clicks. Trillions of dollars of enterprise value depended on those constraints persisting. – Citrini Research
Agentic commerce will render Amazon and the rest of the rent seeking marketplaces obsolete given enough time. Because LLMs can literally go straight to the seller and perform checkout, do market research to make sure the seller is legit, and the seller can sell for lower than on the marketplace since they aren’t paying a 15-20% cut.
I don’t buy it.
First, I think there is value in the “rent seeking” Amazon marketplace because how else would the models “go straight to the seller”, another centralized search engine? Why not just use the Amazon one then?
Second, one of LLMs big weaknesses is judgement on what to trust. I would not trust the judgment of an LLM to determine “the seller is legit”… unless we outsource trust verification to a third party marketplace (who will want a cut).
Finally, OpenAI has been aggressively pushing for this so they can take a cut of the transaction. So it’ll just be another middle man.
> So it’ll just be another middle man.
Exactly this, and not just another middleman, a middleman with an obscene burn rate that isn't close to profitability and is incentivized to ratchet up prices as soon as they can.
And then AI procurement has problems on the buyer side. Do I just blindly trust that the model is going to make the purchase as specified? Do I trust the model's search capabilities and objectivity of returning results? How do I know that OpenAI isn't running its own "marketplace", only showing me options to buy that they want me to see while filtering out less desirable options for them?
It's a fundamentally less transparent experience than Amazon.
How long until the AI provider takes a 15%-20% cut? "Affiliate fees"...
It's why I'm building an open-source agentic commerce chat
You disregard the fact that the sellers will have opposing LLMs constantly adding more signals that they're legitimate.
The sellers and the marketplaces can spend more time on their LLMs because it's their livelihood. It's the same asymmetry with different tooling.
"We are going to put online shopping on the blockchain and eliminate gatekeepers and fraud"
Who mentioned blockchain? You just need to allow e-commerce website to expose their storefront as a MCP app or UI and then clients can interact with them directly.
>Third party marketplaces sprung up for reservations, and idk it’s been a while since I went to fancy dinner, but I imagine the restaurants have just started charging. Or at least the first party reservation sites do.
Yes, this is what Tock is for. It's not clear to me that it's a bad thing. It replaces the old $20 in a handshake I used to do with the maitre d at the front of the restaurant. Democratizes opportunity and improves transparency
I think self driving cars are important for a couple of reasons:
1. Demographics - Aging population needs transportation. God knows we certainly don't need really old people driving themselves. We got a taste of the future in West Portal in that regard not long ago.
2. Human Capital - The US has pretty much demonstrated that there is little desire to import low skilled labor. Where do these theoretical Taxi drivers come from? Or welders or plumbers. Labor is going to become increasing expensive no matter how you slice the pie.
3. Younger US citizens are going to gravitate to non-manual labor jobs. It is not just that every one is being steered toward college. Physical labor (trades) take a toll on the body. I know - I have work in them - and you quickly extrapolate what that will be like when you are 50.
Public transportation solved this before cars were commonplace. Implement ubiquitous and free public transportation in every urban center (where 80% of the American population lives [0]) and you'll save billions from not having to manufacture (and store!) cars.
0: https://www.census.gov/programs-surveys/geography/guidance/g...
Methinks this post conflates “rent seeking” with “return on investment” just a tad.
Economic rent is the extra money you can charge for owning a scarce resource. ML models are not waterfront real estate, they are IP. Other people can make more models if they can/want to.
Now, whether IP should be legally protected is a totally separate question, and while we in the West tend to assume the answer is obvious geohot would certainly not be the first person to suggest broadly applying private property rights to information makes questionable sense.
This is a highly sensational take that is basically fan fiction. From "the era of purposefully frustrating humans is over", to "the added bonus of the collapse of the US economy. Frankly, it’s well deserved." and "everyone in the world is rooting for the Chinese models"; nothing of that is grounded in reality.
The Chinese models are open source because they are not state of the art. Once they catch-up or lead, they will likely close them down by a government mandate. Just like Meta was fine with Llama being open source but once they started to get close to OpenAI/Google/Anthropic, they shifted their language to "maybe we won't keep doing that."
The idea that AI will end the "rent-seeking class" that has effectively existed for thousands of years is... not going to happen! The business model just adjusts. And if AI is going to be an economy-shaping super disruptor, the cloud-hosted models will continue evolving beyond what you could ever run at home under the desk.
> The Chinese models are open source because they are not state of the art
I think geohot is burying the lead in this text in his post with a lot of speculation.
It's not not that these specific models will become closed it's that the hardware/hosting vendors have an incentive to train models where inference is custom tuned to their chip's dimensions and VRAM.
The Chinese models do a great job of showing what's capable on consumer/prosumer hardware because of export restrictions but anyone entering the hardware space has the same incentives to undercut the frontier labs so they can sell more hardware.
It's also not clear if being at the forefront of inference quality really matters. The open source models appear to be doing a fine enough job of keeping up even if they're a few months behind. So it seems like there's not much of a technology moat for these labs other than the capital costs of training/serving.
> the "rent-seeking class" that has effectively existed for thousands of years
This "rent seeking class" is not a historical universal, regardless of how much college marxists insist that it is. Leaders can be good or bad, and they hold power in different ways. In America today we have bad leaders (across the entirety of the political spectrum) - and AI poses a lot of challenges in how they hold power. This is not to say Chinese leaders are any better - but the way they hold power is not challenged by AI. Business models will indeed adapt - but the condition is excellent, as they say.
Let's not gloss over the electrical supply. These chips won't work for free.
LLM inference uses on the order of 1 Wh per query. That's under 10 meters of driving on an EV or running air conditioning for under 5 seconds.
https://hannahritchie.substack.com/p/ai-footprint-august-202...
Or a human riding a stationary bike for 36 seconds.
Betteridge's Law of Headlines hard here.
> The era of purposefully frustrating humans is over. The Chinese open source model running on the box under my desk can pass the Turing Test. When you call, e-mail, text, or show me an ad, you’ll never know if it’s me or my model seeing it.
But at some point, you're going to want to do something, like, e.g., buy something. Then you're right back to the problem in the opening quote:
> things take time, patience runs out, brand familiarity substitutes for diligence, and most people are willing to accept a bad price to avoid more clicks.
& we're already seeing AI used to do this. E.g., Amazon listings where product photos are AI generated. (… not that many product photos weren't "bad photoshop of product onto hot sexy model who is obviously not using our product" before … but now it's AI!) Whereas before someone would have had to spend a modicum of time badly using Photoshop, now AI can just churn out the same fraudulent result in a fraction of the time.
Now, if I have a problem with a product, instead of just calling a number, browsing a phone tree, getting put on hold, and finally having to struggle to get some human to understand the basic logistics of "I paid for X, I did not get X, I demand X or refund", I get to do all that but with the extra step of "forced engagement with an AI that is incapable of actually solving my problem". (This somehow still manages to apply even when the problem is seemingly trivial enough that I find myself thinking "… this actually should be something an AI can do" but inevitably, no, the AI is "sorry", it cannot do that.)
And besides, calls, emails, etc. are already handled without AI: I (and everyone I really care about) have either allowlisted all inbound comms, or abandoned the medium altogether. Moreover, any communications medium is useful because it is not infested with spam, and will eventually be destroyed by spam. At least until we grow laws for mediums like phone/email, maybe named things like "Do Not Call" or "CAN-SPAM" and those laws are enforced. But the GOP has no interest in enforcing any level of consumer protection, so here we are.
> How often do you diligently check Uber and Lyft and select the cheaper one?
Almost always nowadays lol. Shit I’ve gotten poorer over the past few years.
> Agentic commerce will render Amazon and the rest of the rent seeking marketplaces obsolete given enough time. Because LLMs can literally go straight to the seller and perform checkout, do market research to make sure the seller is legit, and the seller can sell for lower than on the marketplace since they aren’t paying a 15-20% cut.
Right.
You'll just end up paying the 15-20% cut to the people who train the model and keep it updated and run the agents that you rent from them.
To me it seems like the rent seeking is everywhere and getting strong in all facets of our lives. Tech companies, utilities, marketplaces, car companies, appliance companies - everyone is pushing subscriptions, gate keeping features and milking every possible dime.
It's undeniable that technology over the past decades has increased democratization in something every step of the way.
YouTube destroyed Hollywood's monopolization of entertainment. Anyone with a smartphone now has a shot at becoming a full-time creator. Prior to this, it was gate kept by Hollywood execs.
Smartphones destroyed Microsoft's monopolization of apps.
Not a leap to believe this will happen to some extent with AI (and it's already happening to some degree).
It seems like the rent seeking class is just moving to selling you access to LLMs in data centers by the token. In the past, the "rent-seeking class" being described here was at least part of the middle class. Now a few billionaires are going to capture all of the value, but the rent-seeking isn't going away.
Since the article is largely about open weights models, I think the argument is that this is the "last gasp" and soon doing inference at home will be common.
The small models that I can run at home are becoming more capable, and I have replaced some API-based tasks with local inference as they improved, but large open weights models are still a lot stronger. The nice thing with larger open weights models is that competing providers serve them at modest margins and prices. I don't have the hardware to run the largest Qwen models, but I can get API access at low cost. Since there are only modest barriers to new commercial inference providers for these models I'm not worried that API access to them will become drastically more expensive at some future time.
And since there are only modest barriers to new commercial inference providers for these models...
Congress: "Hold my beer and watch this"
The trend over the last decades was towards more centralization and I don't see that changing. Unless we radically change our economic system, the rent seekers will always win. There will be probably less of them but they will be even bigger.
Running on what devices (and additionally, purchased with what money)?
I find it amusing and, IDK, charming how “rent-seeking” has become a general purpose pejorative, like “bourgeoisie” was at one time.
> The best anyone can hope for is a free market, with everything properly priced. But for decades, the American market has not been free. It’s used purposefully added friction to exploit a time asymmetry between the business and you. And due to things like call centers, this has been very profitable for the businesses.
I just think this analysis is wrong from the start. The "proper" pricing structure, the one tracking the actual costs involved, would be that you don't get to talk to a human being at all unless you pay for their time. Human frictions are what allow no-charge customer support to exist.
you do pay for their time, when you (or someone else) buys the company's products. but customer support exists because it makes the company money, so really they should be paying you
ah yes, capitalism is over because the chinese are benevolent people that just give away the goose that lays golden eggs out of the goodness of their hearts
this will continue forever and no rugs, chinese or otherwise, will ever be pulled
we know that because the label on the rug says "open source"
It's going to be the same asymmetry with different tooling. Why does he think people will be the only ones with these tools? Companies will utilize them, as they already are, and will also have more time to fine-tune, manage, and improve them.
It's going to be your bots vs theirs. Theirs will have more resources. Net result? probably fewer jobs and wealthier companies yet again.
> The best anyone can hope for is a free market, with everything properly priced. But for decades, the American market has not been free. It’s used purposefully added friction to exploit a time asymmetry between the business and you.
A free market that is "properly priced" is a not a real state of existence.
Resource and information asymmetry, and the exploitation by the those with resource and information privilege of those without it has been present from the very beginning. A free market is just a tool (among many) to achieve a goal for a society.
For some, that goal is explicitly the concentration of wealth and welfare in very few hands. This is oligarchy.
For others, it's the advantaging the welfare and dignity of their "tribe" at the cost of welfare and dignity to perceived outsiders.
And for yet others, it is the advancement of universal welfare and dignity.
Neither a free market nor socialism gets you any of these. What gets you there are the shared narratives that utilize tools like free markets, regulation, and redistribution.
This argument is a bit scattered. "Rent seeking" is being misused here. It's a relatively new term (~50 years old) but it has a lot of history behind it, specifically with enclosures. The Enclosure Acts [1] were a series of laws that took what was common property open to all and made them private property. This was embryonic capitalism [2].
Anyway, I remember that Google demo of making restaurant reservations. I believe it was scripted and had a human fallback. Little did we know that Google would drop the bag on the whole transformer thing that came out soon after. I wouldn't be surprised if it was some of the same people involved.
What the author is talking about isn't rent-seeking per se but a moat. The entire proposition of OpenAI is that they can build a moat and recoup the billions of investment. I'm not convinced that's true, which is part of the author's point, for some of the same reasons:
1. Cost of hardware and training and tokens keeps going down. We saw the same thing with Bitcoin mining. I wonder if we'll see ASICs enter the fray here too; and
2. China will make sure no one company owns this future. DeepSeek was a shot across the bow of OpenAI, Google and Anthropic. It is a national security issue for China.
Where I disagree is that this will be an end for the rent-seeking class. I think we're bouldering towards a dystopian future of even more wealth concentration where most people get displaced by automation and AI, which suppresses wages and ultimately leads to a situation where a handful of people have all the money and almost everyone else has no money.
[1]: https://en.wikipedia.org/wiki/Inclosure_act
[2]: https://medium.com/@jrcoleman97/the-hidden-origins-of-capita...
> This argument is a bit scattered. "Rent seeking" is being misused here.
It's being used in a more literal-meanings-of-the-words sense ("pursuing monopoly rents") rather than the narrow economic term-of-art sense of "pursuing monopoly rents through influence over public policy by means that do not create, or which inhibit the creation of, additional wealth" (the definition you seem to be complaining about it not adhering to without actually providing.)
But most of the usages would also be correct in the narrower sense, because virtually ever actor referred to as rent-seeking in the broader sense are also rent-seeking in the narrow sense as part of that. (E.g., actively lobbying for "safety" regulation which would disproportionately impair non-incumbent new competitors.)
> What the author is talking about isn't rent-seeking per se but a moat.
Pursuing a moat is just another term for seeking monopoly rents by any means, including rent-seeking in the narrow sense.
(There's also obviously an ideological angle in creating the term "rent seeking" as a term of criticism to those seeking monopoly rents through means that the creators of the term disapprove of, excluding seeking the same kind of rents by other means from "rent seeking".)
No, a moat is a competitive advantage/ OpenAI in particular is predicated on the belief that they will have a compeitive advantage. ASML is a compeitive advantage with EUV (for now). You can overcharge for if you have a compeititve advantage but that's not the same as rent-seeking.
Rent-seeking is fundamentally intermediation like a health insurer putting themselves between a patient and a healthcare provider or privatizing grazing lands or controlling water supplies from snowmelt (like the Resnicks) or privatizing trains in the UK.
Microsoft has a compeitive advantage with Windows, Google with its search engine and the ad business that funds it, Amazon with AWS or NVidia with GPUs. There are alternatives to all of these things but these companies maintain dominance with a combination of scale, cost, technology and network effects. That's not rent-seeking in a broad or narrow sense.
Rent-seeking would be Microsoft lobbying lawmakers to require schools and governments to purchase Windows, for example.
> The best anyone can hope for is a free market, with everything properly priced. But for decades, the American market has not been free. It’s used purposefully added friction to exploit a time asymmetry between the business and you. And due to things like call centers, this has been very profitable for the businesses. Cable companies and insurance rely on the fact that your time is more valuable than theirs. They can hire people in India at scale to waste your time. They can use procedure and big data to design protocols to drive you just to the point of frustration at little cost to them. How often do you diligently check Uber and Lyft and select the cheaper one?
> Enter AI, the great equalizer of time.
I didn't read any futher: this article is dumb. If a company has the capability to hire literal people to waste your time, they can deploy more AI than you to waste the time of your AI.
Or they just use price to limit access instead of time. Which means you're totally SOL if you have time but no money. Pay to win, that game everyone loves /s!
AI doesn't flatten asymmetries, it exacerbates them.
> If a company has the capability to hire literal people to waste your time, they can deploy more AI than you to waste the time of your AI.
I don't totally buy OP's argument, but I think you're dismissing it unfairly.
His point is that in the pre-LLM world, if a company wants to waste your time, they can hire a call center employee overseas for US$4/hr and make you wait for an hour to talk to them for 30 minutes. If you value your time at $80/hr, then the 90-minute call cost you $120, but it only cost the company about $2, thus the asymmetry.
OP's claim is that now, the asymmetry is gone. If both you and the company try to use AI, the company has less leverage to impose costs on you. They can deploy more AI to waste more of your time, but that means the asymmetry now is in the customer's favor because it costs the company more than the customer to get support.
>The best anyone can hope for is a free market, with everything properly priced
This is an oxymoron.
1. This gentleman appears to write with an optimism that befits a sliver of society.
2. Anthropic does not care about what models and hardware he is running under his desk.
3. When you look behind the cupboard—Anthropic is "rent seeking" on a level well above consumers.
4. I've got "AI safety" + "Capitalism" + "Military-industrial complex" bound together on my mental corkboard.
"But like people who are good with computers, the models want a terminal, not some candy ass iPad UI."
Giving functionally illiterate people computers with GUIs should be regarded as a mistake.
A society/country that produces functionally illiterate adults should be regarded as a mistake.
At some point you were part of that group that you are labeling a mistake. At another point, your children are or will be if you have any. Locking all of these productivity tools in one generation is a recipe for failure. You should think outside the box that you are currently trapped in.
nope, I learned how to compute on the DOS command line
Same here. It's been a while.