Most people are running a whole lot less than $20's worth of tokens per day on cloud platforms. (Is that assuming a frontier model? 1M output tokens per day?) Local hardware could easily take up that workload, at least the part of it that's non-time-critical.
Phones and laptops are terrible devices for local AI, way too constrained by bad thermals and small batteries. MiniPC's (many of them using mobile hardware) don't have that particular issue, and can easily run on a 24/7 basis.
That level of local AI is also more or less what you need for competent autonomous robots, too. If your household robots are orchestrated from your phone, the local security and cloud convenience converge on a single device. No extra servers, etc, reduced cost, all that - local AI is a massive market amplifier.
Let me speculate - we are going in the weird direction of no private property unless you're an overlord that rents his property to peasants. I like to call it the revenge of communism. See how the market behaves in the llm space - it's more viable to share infrastructure than to own it. Imagine the private car revolution in the US was a bus revolution.
These are more like HPC supercomputers than garden variety datacenters. That's why there's so much concern re: water use for the electricity being supplied. (That's easy to address in principle, of course: wind and solar power use up negligible amounts of water compared to other sources.)
Most of the water concern is evaporative cooling of the datacenter itself. But IMO not too much of a concern. The energy use and the resource use to make the chips, etc, is bigger.
OpenAI and Anthropic have the know-how for building much larger models that will be a lot smarter and run on datacenter-scale compute. This is a natural 'moat' that will be inherently hard to replicate for on-prem compute or small neoclouds running open-weight/local AI. They can easily coexist with a robust local AI scene.
All modern Western-like societies involve some amount of indirect redistribution already. Outside of extremely peculiar places like Singapore or the Gulf states, it's just not seen as desirable or even sensible to have extreme wealth alongside people living in extreme poverty on the equivalent of less than a dollar a day. This actually used to be relatively common in the 19th century, it was the actual kind of widespread pathology that early social reformers railed against.
It is still extremely common today, if you look at the demographics along the Atlantic Coast of the US. The richest zip codes always have poor ones nearby.
By the standards of underdeveloped countries today or historical poverty in general, these "poor" people are nonetheless living in outright opulence: their genuine plight is mostly one of social marginalization, that can't really be solved by purely economic means. That's partly the effect of new technology (developed by capitalism) but partly redistribution in action.
Historically, more efficient agriculture meant a population boom. That's kinda the opposite of people starving to death. A lot of agriculture historically and in poor countries like India today is subsistence agriculture, yeoman farmers living off what they grow directly. More efficiency allows them to sell their surplus and to invest the proceedings, kicking off economic growth.
Yes, but the AI that is metaphor is comparing to does not create more food. More to the point, it may not create more jobs.
After a few decades of turmoil the industrial and agricultural revolutions netted out far more jobs. The verdict is still out on AI, but I wouldn't bet on it.
It also doesn't destroy food. Right now, we have enough food. After the AI revolution we have more food and more free labour and fundamentally more effective administrators to run a welfare system. I don't want my society to be the first one to try it, but if we can move the average administrator from an ordinary human to something that is a little better at math than Gauss with infinite clones to get into the details ... there is a chance that we can run an effective centrally planned welfare system.
It is really hard to see how the AI revolution would lead to any issues with food shortages. It looks more like previously unthinkable upside than anything else.
Your analysis is greatly under estimating the risk that the capitalists that control the system use it to build cheap, automated weapons to guard their cheap robots and lock everyone of us out, just because they can. They're far more likely to be narcissists and sociopaths than the average population, empathy isn't their strong suit.
> Yes, but the AI that is metaphor is comparing to does not create more food.
Mostly because food is incredibly cheap, so it's not the main focus of present-day economies. AI does however help provide many basic services that improve quality of life. The most natural and most cost-effective use of AI is arguably in helping answer simple questions, not really in cranking out tokens to somehow help write complex software. And other service work is perhaps in the middle of this range.
> But the answers it gives are not reliable. They sound plausible if you don't know anything about the subject, but they're not reliable.
Do not underestimate the utility of having a starting point overview on a topic you know absolutely nothing about. It may be immensely valuable even if some details are off. That's what made the XVIII's Encyclopedia such a valuable tool for civil society.
By the time you get to the point where those wrong details become relevant, you have gotten a basic understanding of what the overall topic is about, so you're prepared to get a second opinion from a different source - and this time you may know enough to start asking relevant questions, rather than starting from full ignorance.
> Do not underestimate the utility of having a starting point overview on a topic you know absolutely nothing about.
Perhaps, but we already had that in the form of search-engines and primers and how-to guides and Wikipedia. The actionable questions already had answers.
Adding an obsequious device that dynamically hallucinate half of a conversation with not-necessarily-true dialog is (if not a detriment) only a marginal improvement.
Hallucinations are not a matter of some "details" being off. They are a matter of plausible, confident-sounding claims that are just plain wrong. They don't help anyone to get a "basic understanding". All they "help" with is getting a wrong understanding, that the poor person who's asking can't tell is wrong, because it sounds plausible and is stated with such confidence.
When humans do this, we call them "bullshit artists", and we don't view them favorably. Why should AIs get a pass?
> Hallucinations are not a matter of some "details" being off. They are a matter of plausible, confident-sounding claims that are just plain wrong.
This is no worse than Wikipedia, or the original encyclopedia for that matter. Those contain dubious claims that you'll need to verify on your own too.
LLMs help because they have a gigantic amount of compressed knowledge, and they are able to find relevant information and present it incredibly fast. You wouldn't trust the ten first results of a Google search either, but you wouldn't say that having a search engine is totally useless and in no way an improvement over your local library, would you?
> the poor person who's asking can't tell is wrong, because it sounds plausible and is stated with such confidence.
True, but having to learn how to use a tool properly doesn't make the tool useless, even if it can hurt those who use it carelessly.
> you wouldn't say that having a search engine is totally useless and in no way an improvement over your local library, would you?
A search engine that just indexes the web and gives me results is of course a great improvement over my local library.
But LLMs are not the same as a search engine. LLMs don't give me links. (Well, sometimes they do, and sometimes the links don't even exist, or if they exist, they don't actually say what the LLM said they say. At least with a search engine it's just the link, with no claims about what I'll find if I click on it.) They give me authoritative-sounding text. That's what they're for. And no, that text is not coming from "compressed knowledge". Text is not knowledge. Knowledge requires connections with the real world. LLMs don't have that. All they have is text.
A comparison I've used before is between LLMs and Wolfram Alpha. If I ask an LLM what's the distance from New York to Tokyo, the LLM has no idea that New York and Tokyo are places on the Earth, that the distance is a physical distance that can be measured, and that there is a right answer to the question. LLMs just generate text based on what their algorithm spits out as the most likely text to follow my prompt. The LLM doesn't even have any concept of what's happening if it gives me a wrong answer and I tell it the answer is wrong. It will spit out text saying, oh, yes, you're right, that's a wrong answer...and then spit out more text that might contain a different wrong answer, or even the same wrong answer. It literally has no concept that I am trying to extract meaning from its text.
Wolfram Alpha, on the other hand, if I ask it what's the distance from New York to Tokyo, figures out that I'm asking for a geographic distance, looks it up in its database of geographic distances (which has been curated by humans using actual geographic data from actual measurements on the actual Earth), and formats the answer in readable text. That's still a very simple connection to the real world outside of text, but at least it's some connection. LLMs have none. And that is what makes them useless as tools for trying to learn actual knowledge.
> AI does however help provide many basic services that improve quality of life.
Not yet. Like, not at all and there is a constantly expressed threat we will all become poorer and unemployable because of it. I dont believe it, but AI did not made life better ... and its creators claim it will make life worst for most of us. That is their literal sales pitch.
Comparable being the key word there. AI marketing is threatening to eventually eliminate most white collar work. The exact high paying jobs (at least in the US) that enable upward class mobility and fuel the consumption based economy.
Take those away and tell everyone "sorry, go do physical labor now for half or worse of the salary" and that's a big problem.
Automation is a boon when it automates physical labor, not when it automates away knowledge work.
The emphasis there should be on "marketing". The actual state of things is that white-collar work is alive and well, and if anything is being helped by AI.
Interesting there is a possible implication here. If salaries drop from more people doing physical labor instead of white collar work then the automation of physical work may be delayed even longer. It may be cheaper in the short term to pay humans than machines due to an oversupply in physical labor.
> Automation is a boon when it automates physical labor, not when it automates away knowledge work.
Says the knowledge workers, who have collectively spent the last 50 years talking down to the physical laborers with a smug "should have gone to college!" attitude.
You'll be fine. Automation of any kind is a boon for everyone. We massively over-allocated human talents to office jobs over the past few decades and stopped building anything in the physical world (like houses, infrastructure, etc), this is only the pendulum swinging back to reality. Graeber wrote about this astutely in his original 2013 Bullshit Jobs essay, long before AI was a thing.
How many people do we actually need sitting in meetings about meetings about powerpoint presentations for future meetings....or implementing react components into a dashboard UI in a slightly different way for the 3,000,000th time? Even without AI, this was bound to happen.
In the early 1900s there were literally hundreds of different automobile manufacturers globally. We didn't need that many, just as we don't need 1,000,000 people working on 100 slightly different versions of the same CRUD project management software. Humans will human. We'll find new stuff to do, as we have done since the dawn of humanity.
Funny you make this statement without giving examples of timeframes, what this looked like in real terms for real peoples lives living back then, nothing. Just 'between 1880 and 1950 we found new stuff to do'. It's all selling magic and hopium based on nothing.
Real example: 75% of the global workforce were farmers in 1880, most on a subsistence basis. The people who left the farm for the factories during that time period weren't forced to. They chose to, because working in a factory was better than staying on the farm. Just like a generation of rural Chinese people made the same choice more recently.
In fact, there's nothing stopping you from buying a farm and living like its 1880 today.
You can quite literally go out and start living a subsistence farmer lifestyle tomorrow. The average person in 1880 did not have the tools needed to cultivate a large parcel of land, so you'd approximate their lifestyle quite easily with a tiny parcel of arable rural land which is extremely cheap to acquire in most countries.
It's not magic and hopium, its simply automation and increased productivity via leverage. AI is the assembly line of the digital revolution.
You need 0.5 to 1.5 acres per person for non-mechanized industrial argriculture. Nowhere with land that is truly arable enough for that is going to _sell_ you 1 acre at a time. In the U.S., you buy at least 40 acres at a time. In the U.S. Midwest, that's going to set you back (on average) $379,000. That's before you buy the equipment you need to be able to farm the land in the first place. Unless you industrialize and grow crops to sell to other people, you will not be able to afford the property taxes on the land to be able to keep it, either.
So, no, you cannot just go out and buy an acre and garden.
What? Yes, you absolutely can buy half an acre. You think in 1880 people went on Zillow to buy land?
You're just going to have to do this the 1880 way.
Knock on the door of a land owner and offer to buy/rent half an acre so you can farm. You'll find takers. My extended family literally has this arrangement with many people who farm different crops during different seasons.
Too hard to to do it that way? Welcome to 1880! Most people weren't land owners on the land they farmed back then and didn't have 'Perfectly arable' plots, and this was pre-fertilizer.
Oh and you'll have to use horse and buggy to get around to find land owners (no evil automobiles from those evil factories full of automation!) who will allow you to farm their land, just like 1880. So good luck.
I don't know how many times I need to explain this to tech doomers: nobody forced people out of subsistence farming. They chose to leave it. It was not a utopia.
Why is you're 'real example' 100% hypothetical? Give me real examples. Or at least real information from places like Manchester at the time. Not this hypothetical stuff the implies much ignoring things got worse for generations.
If you read up on the industrial revolution, those people that moved became less healthy, less happy, forced into dorm style housing. Give me real examples of what 'working out' looked like in the past. Because from my research, 'things worked out' meant worse outcomes for quite a long time (like generational timeframe). Give me examples please of what this successful transition in the past looked like for real individuals.
The other big worry is, what if it just doesn't do what is promised and these trillions of dollars that were spent assuming magic would happen were all for nothing? I mean, other than to make a handful of extremely wealthy individuals even more wealthy at the expense of everyone's retirement funds.
Fiber buried in the ground in 1996 is still useful. Servers from 1996, not so much outside of the retrocomputing community. The bulk of those trillions of dollars on AI is not going into useful long term infrastructure. It's going into equipment that will only be useful to scrappers after its initial life is over in three to five years as the sorts of places that can handle the heat load of 25 clothes dryers on high stuffed into 3.5 cu ft of space aren't going to run second hand machines. They aren't useful as in-office developer machines unless your office has 1000A of power to dedicate to that one single machine and the air conditioner need to keep the room the server is in from bursting into flames.
> The other big worry is, what if it just doesn't do what is promised and these trillions of dollars that were spent assuming magic would happen were all for nothing?
Honestly? That's the best case scenario for humanity.
When AI can replace knowledge-based work, capital has no need for humans anymore. That's almost an ELE.
There is a basic problem with framing though. Why does the labour need to find somewhere to go, but capital doesn't? Why can't the increase in productivity be captured by labour and denied to capital?
Do you mean to imply a political/social revolution? In any other scenario I can think of when my boss gets a new machine, he captures the value from my increased productivity or the machine eliminates my job entirely.
Changing the tax system to tax capital rather than labour would probably get you 90% of the way there without great societal upheaval (capital would fight back though).
There are precedents for a lower workforce. It was not so long ago that women did not participate much in formal labor, but rather spent their creative energies improving their families and homes. That might not be an empowering choice today, and I'm not advocating for it, but it shows that the economy has in the past and probably still can get along perfectly well with a lot of sidelined labor capacity. The important thing is that the sidelined labor find some useful purpose outside the workplace rather than simply consuming welfare: in the past, domestic work like childrearing and social/emotional work like building community soaked up excess labor capacity and still had pro-social effects.
Just because they weren't paid doesn't mean women were not doing economically valued labour. The washing machine is probably the greatest productivity unlock since the steam engine.
If you focus on writing ricketty software or overblown emails then no it isn't real. But if you think of e-gates instead of border officers then it is.
> Historically, more efficient agriculture meant a population boom. That's kinda the opposite of people starving to death.
not necessarily. you're inadvertently conflating things. just more people alive doesn't mean they aren't starving. a population boom can be had in the starving population too.
While you are not wrong, it is still historically correct to say that "more efficient agriculture meant a population boom". We don't know what they were doing for birth control back then (because this was a woman's job and they didn't write history), but there is plenty of evidence they must have been doing something that was effective (rhythm is more than good enough to explain this, and so likely what they were doing). People had a good idea of how much the farm could support and they tried to get just enough kids to ensure it would pass on - with enough spares for war, infant mortality and the like.
> more efficient agriculture meant a population boom
More efficient agriculture meant a more efficient population. In cases where environmentally possible this obviously encourages a population boom but they're not necessarily synonymous.
The most productive places in the world are also those with the highest incomes and wealth generation. "Excess productivity" is either temporary or a sector-specific phenomenon, it doesn't apply to society as a whole.
This is true; and in time it's entirely possible that AI makes us an overall wealthier and more productive society. However, from the article:
> “Most economists will acknowledge that technological progress can cause some adjustment problems in the short run. What is rarely noted is that the short run can be a lifetime.”
The redundant professionals will need to find other ways of generating wealth from their productivity, and that may not be possible in a reasonable amount of time. Not in the scope of their remaining lifetime.
We will most likely redefine what we mean by human "productivity". A plumber might be considered a highly productive worker, whereas many intellectual professions will partially refocus on effectively prompting AI and assessing/revising its output.
Temporary excess productivity can linger a very long time and sector-specific excess productivity can still be broadly damaging. Detroit and southeastern Michigan were devastated by the collapse of American automotive industry in the '00s, taking something like 10-15 years before starting to recover.
Michigan was doing okay until around the late 1990s. The real downslide started out after that; Detroit became the largest municipal bankruptcy in U.S. history in 2013.
Someone will always have to prompt the AI, it can't just do that on its own. Or rather, maybe it can (you can just prompt it to "kindly do the needful" in a completely unspecific way) but the results won't be any good.
Sure, but the question is at what layer of abstraction do you have to prompt the AI?
You used to have to prompt the AI by starting to write the actual line of code you want, which it could autocomplete. Then you had to prompt it to write simple scripts or functions. The amount of scope you can prompt keeps getting bigger and bigger. Eventually, you have a PM or a CEO just telling it what features you need. Maybe it's a PM and a designer and a CEO and a CTO, but it will eventually get to the point where the number of people you need to do the prompting shrinks orders of magnitude from company sizes today. Maybe you just give the AI some money, prompt it to start a money-making business, and it goes out and does the same research and analysis that a seasoned entrepreneur would do to find an opportunity then builds out the business from there.
> the results won't be any good
Maybe, but I wouldn't bet on that. The trend over time has been that the results from prompting AI to do things have gotten better. I used to prompt it to build me dashboards and it would fail spectacularly. Now it one-shots them. Maybe the code is terrible (though doesn't matter for me, I'm the only one using it and I can verify the dashboard content is correct), but if the trend continues, it'll get better. Maybe the trend won't continue, but I've yet to come across a good explanation of why AI capabilities will just top out and cease improving forever.
There's no evidence that AI is replacing human jobs to any real extent. We're just seeing AI being blamed for ordinary layoffs that have more to do with broad economic instability.
Yes they have. They cost more. Why keep the senior curmudgeon employee around when the Jr who costs half as much is deemed sufficiently competent? And the Jr isn't going to quit in solidarity either, they're just happy to have not gotten cut.
reply