ChatGPT (and all high end players in the current AI iteration) do not follow the normal laws of software economics.
Normally software has a near 0 marginal cost. This allows sellers to offer steap discounts because they cost the seller almost nothing. In many cases, sellers are better off having you use their software without paying instead of not using it; because they are out almost no money, but have increased their odds of a future sale
High end LLMs are different. Sellers are not setting their price to maximize revenue. They are setting their price to cover the marginal cost of providing the service plus margin.
Lowering their price is not a matter of price discrimination. It is a matter of engineering a cheeper product.
>I find this very troubling for two reasons. First, while Claude and OpenAI are new, Google has been around for a long time. It should have thought better about pricing so that the tool is within the reach of the developing world.
Why? I don't see a practical argument for why Google would want to offer this service at a massive loss.
The phrasing is making it sound like a business argument though. "[Google] should have thought better ..." sounds like "well it would be better for Google the company to do ..."
Why? We are not talking about access to safe drinking water, AI is the latest tech vanity / bubble. Some people have more direct life concerns like unstable governments or access to education than making a ghiblified version of their photo.
I feel like this is a huge problem with the progressive movement in the US. Morally sound arguments that rarely make practical or economic sense - and no, "tax the rich" doesn't get us there: https://freakonomics.com/podcast/ten-myths-about-the-u-s-tax...
It is now US policy to push US AI onto the rest of the world. They can either lower prices to make it affordable in those regions, or, get ignored and use Chinese models which are an order of magnitude cheaper, like Deepseek, or just immigrate.
That's not an unfair assessment, but I would pose additional questions:
Why does doing the morally sound thing always have to be easier, better, and cheaper than ruining the world and creating human misery? Why must the morally superior argument be financially unassailable?
I'm getting older and angrier and I've grown tired of meeting greedy, disingenuous adversaries halfway just to have my hand slapped away after they get what they want. Since we've established that conservatives will absolutely put up with pain to harm outgroup bogeymen, I don't listen to their appeals for things that I hope will reduce human misery in the world.
With that in mind: Tax the rich like it's the 1950s, healthcare for all, stop burning fossil fuels, etc. Do it all and see what works, darn it.
Once you accept that doing the morally sound thing costs money, the question becomes what is the most good you could do with that money.
We could pay to run LLMS at highly subsidized rates for the global poor. Or we could take the money we would have spent running those LLMs and just give it to the poor. I'm not saying direct cash infusions is the best way of spending that money. But doubling their income seems a lot more effective that a ChatGPT subscription.
> Once you accept that doing the morally sound thing costs money, the question becomes what is the most good you could do with that money.
Relative to the topic of this thread, providing access to LLMs at a loss would not be at the top of my list of ways to right moral wrongs either. But more broadly, taxing the rich costs nothing, unless one believes that Reagan economic theory is backed by actual empirical evidence. Some actions in civic life are done for symbolic reasons. People doff their hats and apply their right hand to their precordium for symbolic reasons. We can progressively tax all to symbolize something about economic fairness and opposition to the winner-take-all ethos.
It does though. As one of "the rich" that progressives target continuously, I still can't afford a house in CA. I'm moving to Washington and CA will lose many median Californians' worth of income.
> We can progressively tax all to symbolize something about economic fairness
Anyways, "taxing the rich" simply doesn't get us there:
- You could liquidate every billionaire, including all their assets, and you wouldn't fund the government for 9 months.
- The rest of us that progressives consider "rich" already pay a fuckton in taxes. If you look at the sources above, this is explained clearly.
Most people who are desperate to immigrate and risking their lives to cross the US border aren't doing it because they can't afford ChatGPT Pro but because of living in extreme poverty or danger.
They don't get work visas, they start businesses even illegally. Or they could just use a Chinese model. Either way this is a losing proposiion for the US.
US doesn't want gov debt even more. (not under dems ruling like last 4 year)
Money don't grow on the trees. Someone has to work or take debt. The question is should rich countries sponsor poor? Probably yes. How much? Trump decided enough is enough. EU want US to keep sponsoring the war. You know, they want it to continue but don't want to pay for it. So they try to hijack US government again by attacking Trump personally.
Immigration can be solved by border control. That's what dems fought so hard against trying to get more voters in. Another reason: they control food distribution.
for now. Wait until you have to compete with those who do have it. It's like competing in a race with a ferrari without a race car. The same thing used to be said of the internet.
I can think of a few job related reasons you’d use it, you know, to make money. If those few specific ones you mentioned are all you can think of then…
Not everything is a high-stakes competition. Not every car is a Ferrari, precisely because not every car needs to be. Same goes for your other example, the local sellers at my local market don’t require an internet presence to sustain their businesses.
That can be said about any infrastructure element. AI becomes one of them along with roads, telephones, electricity, etc. Should rich sponsor all of it to make poor around the world competitive? There are charities for this. Actually it's more complicated. The idea of world wide taxes was around for years.
A CS degree costs something like 124 yrs salary for someone in a low income nation—and it’s a much longer harder road. I’m not AI’s biggest fan, but arguably, this type of tech actually narrows the gap. Even if it’s expensive.
That’s a false equivalence¹. You don’t need a CS degree² to become a good programmer, you can do it with time, perseverance, and access to an old machine. Additionally, even if you needed a CS degree, you don’t keep paying for that indefinitely³ and get skills which last for life. An LLM subscription is something you have to keep paying for, and are screwed if you can no longer afford it, it goes down, or you’re in a place without internet connectivity.
³ In the sense that it’s not a subscription. I get that in the US you may be paying for student loans for an unreasonably long time, but that’s not normal for the rest of the world.
It sounds like you are talking about the cost of doing a CS degree in a developed country. The $736 (number used in article, source World bank) number multiplied by 124 gives you $93k. That is enough to manage a degree at one of the cheaper (but perfectly OK - regulated to ensure minimum standards) universities in the UK such as Chester. It cover one year of fees at Oxford but not leave you much to live on. I am pretty sure there are cheaper options in Europe.
Of course, someone from a low income nation is most likely to go to university in their own country which is a whole lot cheaper (and a lot of low and middle income countries have free or subsidised university education - which is why British hospitals were historically had lots of South Asian doctors, and now Africans). If their own country does not offer the right degree or demand for limited places is very high they can study in another low or middle income country (I know Sri Lankans who have studied in India).
This is utter bullshit. An American CS degree might cost that much, but why would someone from a developing country (who isn't rich or getting a scholarship) want one of those?
The actual cost of a CS degree varies a lot depending on the country, but here in Vietnam I think it's about $1000 per term at public universities. That's not cheap, it's about a year at minimum wage here. But it's a long, long way from your claim of 124 years.
And to forestall the obvious next claim: Vietnamese education is quite good actually. Maybe you won't be going to Harvard but there's plenty of universities in the top 1000 worldwide with a few in the top 200 (no idea for the ranking for CS specifically though).
That's for international students going to the UK to study CS. There is not much point for anyone to go to the UK to study CS for that amount of money (unless they already live there, but they get their degree for a third or even less of that money).
It's normal for international student tuition fees to be inflated by many universities, they try to collect some extra revenue based on a perceived extra prestige, especially the US and UK.
Similar to charging different prices for tourists than for locals.
> There is not much point for anyone to go to the UK to study CS for that amount of money (unless they already live there, but they get their degree for a third or even less of that money)
A lot of people do though. There are lots of international students, many from low and middle income countries. Obviously from high income families.
> It's normal for international student tuition fees to be inflated by many universities, they try to collect some extra revenue based on a perceived extra prestige,
In the UK the government heavily subsidises the fees of British students (basically defined as having lived in the UK for the previous three years, other than on a student visa - there are some other complexities but that is the simple version) whereas overseas fees are the full market rate.
> Similar to charging different prices for tourists than for locals.
You are factually wrong, a Yemeni does not need to pay 100+ years of salary to get a CS degree, end of story.
Also, I've been a researcher and have few scientific papers published (you can search for my name on scholar: Enrico Polanski) and I've seen ZERO evidence that a student from Harvard or Imperial to be more knowledgeable than one in unnamed universities from the third world you've never heard about. None.
It's way too personal and student dependent. Plenty of people in ivy league famous colleges study to ace exams and don't remember shit few weeks later. Plenty of people in unnamed universities are genuinely curious.
Your college makes very little difference in how prepared you will be. Single teachers/courses may have an impact, but the location very little.
> You are factually wrong, a Yemeni does not need to pay 100+ years of salary to get a CS degree, end of story.
So far as I can tell, the like-for-like comparison is as per the other commenter you responded to: here's a fancy thing rich people in rich countries use, therefore the comparison is to a rich country's degree.
This is because you also don't need to pay 38.6 months of income to get access to an AI. Not even to access OpenAI's best. And even the downgrade after usage limits is not terrible.
Of course, if you don't like this comparison, then sure, I'd accept what you say. I'm disagreeing about the assumptions of what's comparable here.
> I've seen ZERO evidence that a student from Harvard or Imperial to be more knowledgeable than one in unnamed universities from the third world you've never heard about.
Mm. Tempted to agree even without looking you up: I'm British, so my reference point for "fancy university isn't automatically great" is half the British politicians.
OTOH, after I graduated, I did live in Cambridge (UK) for nearly a decade, and I do miss how incredibly densely packed it was with nerds, it's not something I found in other places.
The difference in quality is not the same. I have worked with graduates of universities from a middle income country (Sri Lanka) and they are pretty good. Plenty of them get jobs in western countries as developers. What they miss out on is not technical skills so much as international exposure to a more developed economy and culture.
I don't think it's got anything to do with ChatGPT, especially given that it has a generous free tier too, let alone a about much cheaper (than the pro) plus tier.
It's an international economy problem, not an AI problem.
This has the to be the dumbest blog and argumentation i've read the whole year... Corporations aren't charities. A ChatGPT Pro subscription isn't necessary for human survival in those countries. A ChatGPT Pro subscription isn't necessary for basic human needs.
Regrettable, but did it take o3 mega pro to find out about real and nominal value? Even something a trivial as an iPhone is a far bigger purchase if you're not on a Bay Area salary.
Home computers were ridiculously expensive in their initial years. Some could cost more than a car. ChatGPT Pro gives you access to cutting edge technology so it isn't surprising that it's expensive.
Just remember that a Wal Mart $50 phone is faster than a supercomputer from the 70s/80s. Prices will go down.
So a handful of things here. One is that you can actually at the moment at least use a lot of it for free. Secondarily, I think when it comes to access to ChatGPT and other services, in a lot of low-income countries there's a much bigger hurdle than the money, which is a combination of language and device that is capable of connecting to ChatGPT. There are a lot of countries still where you're limited to feature phones or not even having a phone at all.
How is it fixed for Africa? Africa has a ton of phones, but it does not have a lot of smart phones. There is a reason that SMS and USSD are corner stones of the african digital economy.
There's 5 billion people with a smartphone worldwide. The device hurdle is gone. ChatGPT is worse in languages other than English, but it's still so much better than searching on Google. The alternative before ChatGPT was watching time consuming videos on Youtube.
One issue I have with the widely used metric of 'Productivity per hours worked' - if I have a second apt I rent out to Bob for $1000, I have a 'productivity' of $6.25 per hour 'worked', despite nobody producing anything and nobody working for any length of time.
But that's the same with GDP: if you and me are living on an uninhabited island and grow/catch all of our food for consumption, our GDP is $0. If however I rent out my bungalow to you and rent yours instead, and I sell you all my food and buy your food instead, somehow our GDP skyrockets.
That's false equivalence. GDP being gameable does not mean the metric has anything to do with how valuable the output of each individual worker is, yet the metric suggests the opposite - if you made a metric like 'Number of Nobel laureates per large bodies of water', that would imply the presence of large lakes somehow aids or hinders top-tier scientific research, which is obviously a false premise.
Same as cooking for youtself or family/friends, taking care of family/friends, exercising, anything that doesn't involve accounting transactions basically.
If 200 USD / month would replace a developer, then it would be a bargain.
If 2000 USD / month would replace a developer, then it would be a fair price.
Problem is that none of existing models are capable to replace a developer. They all need babysitting. And I am saying that as a probable customer who would actually pay something between 200-2000USD for AI junior developer which can actually work and understand what it is doing. Either those models will start doing what has been promised by tech CEOs, or AI winter is upon us.
The morons who want to replace developers with these so called tools already feel like developers need babysitting. Partly because they suck at giving requirements or sticking to them or both. It will be fun to see them finding out the devs do more that bashing code.
This argument would make more sense if machine learning inference had zero (or close to it) marginal cost, which is the case for intellectual property among other things.
However, inference infrastructure is anything but free, and I'd estimate that cost to currently dominate the fraction of dollar per token.
someone has to pay for the servers at the end. are you asking for openai to subsidize ChatGPT Pro for low-income countries? Since OpenAI is for-profit entity focussed on profits, I don't think it might be a wise idea financially for OpenAI to do so.
It might be better for OpenAI, but this disparity only increases immigration. If the US is serious about keeping immigrants out, they should subsidize access to AI.
I take one look at our smartphone/social media addicted society and the damage that does and think of the parallels of forcing opium on a population in the 19th century
Not that the rural third world don’t already have phones. Whether they are engineering to be as addictive as crack like in the west I’m not sure.
Oh, they are already as addictive as crack because Zuckerberg has successfully pushed his social media onto them. And it's very hard to compete with Zuck. The network effect is a very wide moat.
A warning about ChatGPT Pro. The 128k tokens context claim is deceptive advertising.
Messages above ~65k tokens are rejected. Messages between about 50k-65k are accepted, but the right-side of the text is pruned before the LLM call is made. Messages just below ~50k are accepted, but are then partly "forgot" on any follow up questions (either the entire first prompt is excluded, or the left-side of the text is chopped off).
The internet used to be a luxury too. No, the internet doesn't cost 100 dollars a month everywhere in the world for 300 mbps. In developing countries it's as cheap as 20 dollars a month for the same speed.
Albeit I'm no economist, I'm quite sure you should compare salaries to costs, not gdp/capita. Whether unemployed/retirees and children can afford a ChatGPT pro subscription seems irrelevant.
Let's take Madagascar, GDP per capita is $ 538. But the average salary is above $ 150/ month.
What interests us really though is not really the average salary in the country, rather the white collar (the end user's) worker's one.
In Madagascar software engineering salaries seems to range from an average $ 850/month for junior roles to well beyond $ 2000 per senior/specialized roles.
And this further ignores that such expenses are generally paid by employees, often with bulk discounts compared to B2C customers.
Which leads us to conclude that if ChatGPT Pro is such a performance multiplier, it is worth the price even in the poorest of the poorest countries in the world.
I believe it's more about witnessing a deepening AI-divide between rich and poor countries. Or the maturing of the AI market.
But in the end, this will only matter if we see these more expensive tools/subscriptions actually having a market advantage that cannot be mitigated by low income countries.
(Certainly no new startup will be able to train new models in low income countries, hire world class AI professionals etc.)
So what? My salary pays for 2-3 developers in the UK, even more in India, and I am not necessarily any better than those developers. Should I feel bad for that as well?
So, sell the same thing at the same price everywhere? Things are cheaper in the developing world so US AI can either follow their pricing or get ignored.
Have a look at industrial accident data globally, considering underreporting in developing nations.
- Same, environmental accidents.
- Same, WMD proliferation, including chem, bio and nuclear.
- Same, malicious cyber.
Now, ask yourself if we have enough problems aligning & regulating AI at the moment?
Are we sure that in the name of laudable egalitarian ideals that we are prepared for the second and third order effects of broad global accessibility to AI, including frontier models?
This is a really stupid blog post, I’m not sure why someone decided to post it here other than to ridicule it.
A person in a low-income country would also need to work for 38.6 months ($2400?) to afford to hire one of my electricians for two days of labor. Things in high-income countries are expensive, who would’ve guessed??
If I lived somewhere that the average income is $200/month, there’s a lot of things on Maslow’s Hierarchy that come before ‘ChatGPT Pro’… like um a stable electrical grid, clean water, sewer system, etc.
ChatGPT (and all high end players in the current AI iteration) do not follow the normal laws of software economics.
Normally software has a near 0 marginal cost. This allows sellers to offer steap discounts because they cost the seller almost nothing. In many cases, sellers are better off having you use their software without paying instead of not using it; because they are out almost no money, but have increased their odds of a future sale
High end LLMs are different. Sellers are not setting their price to maximize revenue. They are setting their price to cover the marginal cost of providing the service plus margin.
Lowering their price is not a matter of price discrimination. It is a matter of engineering a cheeper product.
>I find this very troubling for two reasons. First, while Claude and OpenAI are new, Google has been around for a long time. It should have thought better about pricing so that the tool is within the reach of the developing world.
Why? I don't see a practical argument for why Google would want to offer this service at a massive loss.
The author is making a moral argument, not a practical (from Google’s perspective) one.
The phrasing is making it sound like a business argument though. "[Google] should have thought better ..." sounds like "well it would be better for Google the company to do ..."
Not a great company to make a moral argument given that Google search has given the world free access to searching information for decades.
Not everything can be reasonably available to everyone if it is cost prohibitive.
Why? We are not talking about access to safe drinking water, AI is the latest tech vanity / bubble. Some people have more direct life concerns like unstable governments or access to education than making a ghiblified version of their photo.
I feel like this is a huge problem with the progressive movement in the US. Morally sound arguments that rarely make practical or economic sense - and no, "tax the rich" doesn't get us there: https://freakonomics.com/podcast/ten-myths-about-the-u-s-tax...
It is now US policy to push US AI onto the rest of the world. They can either lower prices to make it affordable in those regions, or, get ignored and use Chinese models which are an order of magnitude cheaper, like Deepseek, or just immigrate.
That's not an unfair assessment, but I would pose additional questions:
Why does doing the morally sound thing always have to be easier, better, and cheaper than ruining the world and creating human misery? Why must the morally superior argument be financially unassailable?
I'm getting older and angrier and I've grown tired of meeting greedy, disingenuous adversaries halfway just to have my hand slapped away after they get what they want. Since we've established that conservatives will absolutely put up with pain to harm outgroup bogeymen, I don't listen to their appeals for things that I hope will reduce human misery in the world.
With that in mind: Tax the rich like it's the 1950s, healthcare for all, stop burning fossil fuels, etc. Do it all and see what works, darn it.
Once you accept that doing the morally sound thing costs money, the question becomes what is the most good you could do with that money.
We could pay to run LLMS at highly subsidized rates for the global poor. Or we could take the money we would have spent running those LLMs and just give it to the poor. I'm not saying direct cash infusions is the best way of spending that money. But doubling their income seems a lot more effective that a ChatGPT subscription.
> Once you accept that doing the morally sound thing costs money, the question becomes what is the most good you could do with that money.
Relative to the topic of this thread, providing access to LLMs at a loss would not be at the top of my list of ways to right moral wrongs either. But more broadly, taxing the rich costs nothing, unless one believes that Reagan economic theory is backed by actual empirical evidence. Some actions in civic life are done for symbolic reasons. People doff their hats and apply their right hand to their precordium for symbolic reasons. We can progressively tax all to symbolize something about economic fairness and opposition to the winner-take-all ethos.
> But more broadly, taxing the rich costs nothing
It does though. As one of "the rich" that progressives target continuously, I still can't afford a house in CA. I'm moving to Washington and CA will lose many median Californians' worth of income.
> We can progressively tax all to symbolize something about economic fairness
Despite the common memes the US has one of the most progressive tax systems in the world. https://manhattan.institute/article/correcting-the-top-10-ta...
40% of Americans pay no federal income tax.
Anyways, "taxing the rich" simply doesn't get us there: - You could liquidate every billionaire, including all their assets, and you wouldn't fund the government for 9 months. - The rest of us that progressives consider "rich" already pay a fuckton in taxes. If you look at the sources above, this is explained clearly.
I would absolutely agree that ChatGPT access loses out to physiological and safety needs by more than a little bit.
Now that I think about it, I'm not quite sure where to put cost-effective LLM access in Maslow's Hierarchy of Needs.
Well, this will only increase immigration. Isn't this what the US doesn't want?
Most people who are desperate to immigrate and risking their lives to cross the US border aren't doing it because they can't afford ChatGPT Pro but because of living in extreme poverty or danger.
not literally, but this will widen income disparity and make the US so much more attractive to immigrate to.
Not when US work visas are so hard to get. And probably you'd wan to import smart people who can think beyond ChatGPT.
They don't get work visas, they start businesses even illegally. Or they could just use a Chinese model. Either way this is a losing proposiion for the US.
well, maybe we can spend some of the money we save not subsidizing the third world on better border security
US doesn't want gov debt even more. (not under dems ruling like last 4 year)
Money don't grow on the trees. Someone has to work or take debt. The question is should rich countries sponsor poor? Probably yes. How much? Trump decided enough is enough. EU want US to keep sponsoring the war. You know, they want it to continue but don't want to pay for it. So they try to hijack US government again by attacking Trump personally.
Immigration can be solved by border control. That's what dems fought so hard against trying to get more voters in. Another reason: they control food distribution.
Luckily, a ChatGPT Pro subscription is not mandatory. It's not like one can't live without it.
for now. Wait until you have to compete with those who do have it. It's like competing in a race with a ferrari without a race car. The same thing used to be said of the internet.
Why? Will you have access to it during an exam? During an interview? Can't you pass a calculus exam without it?
I can think of a few job related reasons you’d use it, you know, to make money. If those few specific ones you mentioned are all you can think of then…
Such as?
Not everything is a high-stakes competition. Not every car is a Ferrari, precisely because not every car needs to be. Same goes for your other example, the local sellers at my local market don’t require an internet presence to sustain their businesses.
That can be said about any infrastructure element. AI becomes one of them along with roads, telephones, electricity, etc. Should rich sponsor all of it to make poor around the world competitive? There are charities for this. Actually it's more complicated. The idea of world wide taxes was around for years.
I mean, they just released their major version and it's not a Ferrari, it's marginally better than what they had on the market 3 years ago.
A CS degree costs something like 124 yrs salary for someone in a low income nation—and it’s a much longer harder road. I’m not AI’s biggest fan, but arguably, this type of tech actually narrows the gap. Even if it’s expensive.
That’s a false equivalence¹. You don’t need a CS degree² to become a good programmer, you can do it with time, perseverance, and access to an old machine. Additionally, even if you needed a CS degree, you don’t keep paying for that indefinitely³ and get skills which last for life. An LLM subscription is something you have to keep paying for, and are screwed if you can no longer afford it, it goes down, or you’re in a place without internet connectivity.
¹ https://en.wikipedia.org/wiki/False_equivalence
² Nor do they cost the same everywhere.
³ In the sense that it’s not a subscription. I get that in the US you may be paying for student loans for an unreasonably long time, but that’s not normal for the rest of the world.
It sounds like you are talking about the cost of doing a CS degree in a developed country. The $736 (number used in article, source World bank) number multiplied by 124 gives you $93k. That is enough to manage a degree at one of the cheaper (but perfectly OK - regulated to ensure minimum standards) universities in the UK such as Chester. It cover one year of fees at Oxford but not leave you much to live on. I am pretty sure there are cheaper options in Europe.
Of course, someone from a low income nation is most likely to go to university in their own country which is a whole lot cheaper (and a lot of low and middle income countries have free or subsidised university education - which is why British hospitals were historically had lots of South Asian doctors, and now Africans). If their own country does not offer the right degree or demand for limited places is very high they can study in another low or middle income country (I know Sri Lankans who have studied in India).
This is utter bullshit. An American CS degree might cost that much, but why would someone from a developing country (who isn't rich or getting a scholarship) want one of those?
The actual cost of a CS degree varies a lot depending on the country, but here in Vietnam I think it's about $1000 per term at public universities. That's not cheap, it's about a year at minimum wage here. But it's a long, long way from your claim of 124 years.
And to forestall the obvious next claim: Vietnamese education is quite good actually. Maybe you won't be going to Harvard but there's plenty of universities in the top 1000 worldwide with a few in the top 200 (no idea for the ranking for CS specifically though).
It doesn't
You can't compare the cost of a degree in the US with how much that person would pay in their country (even for a top uni there)
And even if you literally compare US costs, that person would probably be eligible to scholarships etc (if they manage to be selected, of course)
Do you have a source for this claim?
UK tuition fees £9,535/year - https://www.gov.uk/government/publications/tuition-fees-and-...
£9,535/year * 3 year degree / 124 years ~= £231/year ~= 310 USD/year
UN estimates GDP/capita of Yemen and Burundi were less than this, that Tajikistan has lower gross average monthly wages. Those are nominal, not PPP.
https://en.wikipedia.org/wiki/List_of_countries_by_GDP_(nomi...
https://en.wikipedia.org/wiki/List_of_countries_by_average_w...
The World Bank numbers here are adjusted for cost of living, say that 1.31% of the world population are living on a dollar a day: https://ourworldindata.org/explorers/poverty-explorer?tab=li...
Why is the basis a high-tuition fee high-cost country? Many countries have much lower fees and costs, including Western ones.
https://educationdata.org/average-cost-of-college-by-country
And many AI are cheaper than OpenAI Pro. For example, OpenAI free.
which has stricter usage limits.
That's for international students going to the UK to study CS. There is not much point for anyone to go to the UK to study CS for that amount of money (unless they already live there, but they get their degree for a third or even less of that money). It's normal for international student tuition fees to be inflated by many universities, they try to collect some extra revenue based on a perceived extra prestige, especially the US and UK. Similar to charging different prices for tourists than for locals.
> There is not much point for anyone to go to the UK to study CS for that amount of money (unless they already live there, but they get their degree for a third or even less of that money)
A lot of people do though. There are lots of international students, many from low and middle income countries. Obviously from high income families.
> It's normal for international student tuition fees to be inflated by many universities, they try to collect some extra revenue based on a perceived extra prestige,
In the UK the government heavily subsidises the fees of British students (basically defined as having lived in the UK for the previous three years, other than on a student visa - there are some other complexities but that is the simple version) whereas overseas fees are the full market rate.
> Similar to charging different prices for tourists than for locals.
Not a thing in the UK.
> That's for international students going to the UK to study CS
Sadly, no, £9k/year is the price for UK students. International students studying in the UK have much higher costs: https://www.uniadmissions.co.uk/international-students/inter...
For Yemen, you should look at the cost of a CS degree in Yemen, which is $ 700/year, not in a foreign country.
To the earlier version of your comment:
> Why would someone in Yemen have to go to UK to get a CS degree when they have multiple universities offering the same course.
Same reason they'd be using an American AI company instead of a cheaper one that e.g. runs on their phone.
That's not an answer.
You are factually wrong, a Yemeni does not need to pay 100+ years of salary to get a CS degree, end of story.
Also, I've been a researcher and have few scientific papers published (you can search for my name on scholar: Enrico Polanski) and I've seen ZERO evidence that a student from Harvard or Imperial to be more knowledgeable than one in unnamed universities from the third world you've never heard about. None.
It's way too personal and student dependent. Plenty of people in ivy league famous colleges study to ace exams and don't remember shit few weeks later. Plenty of people in unnamed universities are genuinely curious.
Your college makes very little difference in how prepared you will be. Single teachers/courses may have an impact, but the location very little.
> That's not an answer.
???
> You are factually wrong, a Yemeni does not need to pay 100+ years of salary to get a CS degree, end of story.
So far as I can tell, the like-for-like comparison is as per the other commenter you responded to: here's a fancy thing rich people in rich countries use, therefore the comparison is to a rich country's degree.
This is because you also don't need to pay 38.6 months of income to get access to an AI. Not even to access OpenAI's best. And even the downgrade after usage limits is not terrible.
Of course, if you don't like this comparison, then sure, I'd accept what you say. I'm disagreeing about the assumptions of what's comparable here.
> I've seen ZERO evidence that a student from Harvard or Imperial to be more knowledgeable than one in unnamed universities from the third world you've never heard about.
Mm. Tempted to agree even without looking you up: I'm British, so my reference point for "fancy university isn't automatically great" is half the British politicians.
OTOH, after I graduated, I did live in Cambridge (UK) for nearly a decade, and I do miss how incredibly densely packed it was with nerds, it's not something I found in other places.
The difference in quality is not the same. I have worked with graduates of universities from a middle income country (Sri Lanka) and they are pretty good. Plenty of them get jobs in western countries as developers. What they miss out on is not technical skills so much as international exposure to a more developed economy and culture.
In Greece university education is free. You just have to finish highschool and take an exam.
I don't think it's got anything to do with ChatGPT, especially given that it has a generous free tier too, let alone a about much cheaper (than the pro) plus tier.
It's an international economy problem, not an AI problem.
This has the to be the dumbest blog and argumentation i've read the whole year... Corporations aren't charities. A ChatGPT Pro subscription isn't necessary for human survival in those countries. A ChatGPT Pro subscription isn't necessary for basic human needs.
I thought it was about comparing the equivalent labor you could hire instead of buying a subscription perhaps
>Corporations aren't charities
So? Perhaps they should be.
> Epitome of AI Access Gap
There is an access gap to almost everything including water when we extend to low income countries.
Also, don't forget that Google for decades have provided free Google search that has allowed the world to search for information for free.
Regrettable, but did it take o3 mega pro to find out about real and nominal value? Even something a trivial as an iPhone is a far bigger purchase if you're not on a Bay Area salary.
Home computers were ridiculously expensive in their initial years. Some could cost more than a car. ChatGPT Pro gives you access to cutting edge technology so it isn't surprising that it's expensive.
Just remember that a Wal Mart $50 phone is faster than a supercomputer from the 70s/80s. Prices will go down.
For all the shorcomings of AI and everyone involved this must be one of the silliest takes.
- The latest concept sports car costs a few lifetimes of income in low-income countries
- Aircraft carrier costs 200 GDP of Tuvalu
Except AI boosts worker productivity and with it, GDP increases. They will use a Chinese model that is an order of magnitude cheaper.
So a handful of things here. One is that you can actually at the moment at least use a lot of it for free. Secondarily, I think when it comes to access to ChatGPT and other services, in a lot of low-income countries there's a much bigger hurdle than the money, which is a combination of language and device that is capable of connecting to ChatGPT. There are a lot of countries still where you're limited to feature phones or not even having a phone at all.
> There are a lot of countries still where you're limited to feature phones
I thought China fixed this for most of the world, at least for Africa it's fixed. It's the Internet access being the bottleneck now.
> at least for Africa it's fixed
How is it fixed for Africa? Africa has a ton of phones, but it does not have a lot of smart phones. There is a reason that SMS and USSD are corner stones of the african digital economy.
There's 5 billion people with a smartphone worldwide. The device hurdle is gone. ChatGPT is worse in languages other than English, but it's still so much better than searching on Google. The alternative before ChatGPT was watching time consuming videos on Youtube.
> There's 5 billion people with a smartphone worldwide
There are more than 8 billion people on this world. By your own account, 3 billion of them do not have smart phones.
> The device hurdle is gone.
Evidently not. There are a lot of countries that still have a smartphone penetration rate of <50%.
We're talking about people with low income here. We're talking about people that cannot even afford bicycles for transportation.
One issue I have with the widely used metric of 'Productivity per hours worked' - if I have a second apt I rent out to Bob for $1000, I have a 'productivity' of $6.25 per hour 'worked', despite nobody producing anything and nobody working for any length of time.
But that's the same with GDP: if you and me are living on an uninhabited island and grow/catch all of our food for consumption, our GDP is $0. If however I rent out my bungalow to you and rent yours instead, and I sell you all my food and buy your food instead, somehow our GDP skyrockets.
That's false equivalence. GDP being gameable does not mean the metric has anything to do with how valuable the output of each individual worker is, yet the metric suggests the opposite - if you made a metric like 'Number of Nobel laureates per large bodies of water', that would imply the presence of large lakes somehow aids or hinders top-tier scientific research, which is obviously a false premise.
My only argument is that both metrics are gameable in a way.
> if you made a metric like 'Number of Nobel laureates per large bodies of water'
That's a strawman argument.
Same as cooking for youtself or family/friends, taking care of family/friends, exercising, anything that doesn't involve accounting transactions basically.
A pack of cigarettes in Norway is virtually unaffordable.
If 200 USD / month would replace a developer, then it would be a bargain.
If 2000 USD / month would replace a developer, then it would be a fair price.
Problem is that none of existing models are capable to replace a developer. They all need babysitting. And I am saying that as a probable customer who would actually pay something between 200-2000USD for AI junior developer which can actually work and understand what it is doing. Either those models will start doing what has been promised by tech CEOs, or AI winter is upon us.
The morons who want to replace developers with these so called tools already feel like developers need babysitting. Partly because they suck at giving requirements or sticking to them or both. It will be fun to see them finding out the devs do more that bashing code.
This argument would make more sense if machine learning inference had zero (or close to it) marginal cost, which is the case for intellectual property among other things.
However, inference infrastructure is anything but free, and I'd estimate that cost to currently dominate the fraction of dollar per token.
This is a real problem but I think the author has it backwards.
As LLM subscriptions become more expensive simply hiring folks from low income countries becomes a more cost effective solution.
> low-income
someone has to pay for the servers at the end. are you asking for openai to subsidize ChatGPT Pro for low-income countries? Since OpenAI is for-profit entity focussed on profits, I don't think it might be a wise idea financially for OpenAI to do so.
Well it was founded as a non-profit
Well, technically they still haven't profited har har har
It might be better for OpenAI, but this disparity only increases immigration. If the US is serious about keeping immigrants out, they should subsidize access to AI.
I think one step at a time.
There are still many countries where digital literacy is very low.
So first step is to ensure everyone has internet/mobile phone/laptop and knows how to use them.
I take one look at our smartphone/social media addicted society and the damage that does and think of the parallels of forcing opium on a population in the 19th century
Not that the rural third world don’t already have phones. Whether they are engineering to be as addictive as crack like in the west I’m not sure.
Oh, they are already as addictive as crack because Zuckerberg has successfully pushed his social media onto them. And it's very hard to compete with Zuck. The network effect is a very wide moat.
A warning about ChatGPT Pro. The 128k tokens context claim is deceptive advertising.
Messages above ~65k tokens are rejected. Messages between about 50k-65k are accepted, but the right-side of the text is pruned before the LLM call is made. Messages just below ~50k are accepted, but are then partly "forgot" on any follow up questions (either the entire first prompt is excluded, or the left-side of the text is chopped off).
Realistically, it's a 55-65k token limit (40k token question, 15k token response).
They want you to attach your context so they can use RAG.
I can't even be bothered filing a bug report, because I know this shit is intentional. The mistakes always run in a favorable direction.
(GPT-5-Pro is a genuinely good model however, and usage limits are generous)
What utility does the guy running a bonbon tè stall get from a frontier AI model anyway?
… who cares?
1. the companies don’t get any cheaper compute because a user is from a low-income country.
2. this is an AI subscription, it’s purely a luxury product. We do not need this to survive. If you can’t afford it, don’t buy it.
The internet used to be a luxury too. No, the internet doesn't cost 100 dollars a month everywhere in the world for 300 mbps. In developing countries it's as cheap as 20 dollars a month for the same speed.
Nonsense.
Albeit I'm no economist, I'm quite sure you should compare salaries to costs, not gdp/capita. Whether unemployed/retirees and children can afford a ChatGPT pro subscription seems irrelevant.
Let's take Madagascar, GDP per capita is $ 538. But the average salary is above $ 150/ month.
What interests us really though is not really the average salary in the country, rather the white collar (the end user's) worker's one.
In Madagascar software engineering salaries seems to range from an average $ 850/month for junior roles to well beyond $ 2000 per senior/specialized roles.
And this further ignores that such expenses are generally paid by employees, often with bulk discounts compared to B2C customers.
Which leads us to conclude that if ChatGPT Pro is such a performance multiplier, it is worth the price even in the poorest of the poorest countries in the world.
Yes, obviously. It's advanced technology. Are we just dividing two numbers now?
Median people from the Democratic Republic of Congo will have to work for 6 years to pay a PG&E bill.
Their power bills are at least an order of magnitude cheaper. Pricing is relative. They also consume fewer kilowatts of energy.
People in low-income countries have low incomes. News at 11:00.
Seriously, what is the point of this observation? Few if any workers earning low wages have any use for a ChatGPT Pro subscription?
I believe it's more about witnessing a deepening AI-divide between rich and poor countries. Or the maturing of the AI market. But in the end, this will only matter if we see these more expensive tools/subscriptions actually having a market advantage that cannot be mitigated by low income countries. (Certainly no new startup will be able to train new models in low income countries, hire world class AI professionals etc.)
Ok..?
So what? My salary pays for 2-3 developers in the UK, even more in India, and I am not necessarily any better than those developers. Should I feel bad for that as well?
So, sell the same thing at the same price everywhere? Things are cheaper in the developing world so US AI can either follow their pricing or get ignored.
Stop, if for just a moment:
Have a look at industrial accident data globally, considering underreporting in developing nations.
- Same, environmental accidents.
- Same, WMD proliferation, including chem, bio and nuclear.
- Same, malicious cyber.
Now, ask yourself if we have enough problems aligning & regulating AI at the moment?
Are we sure that in the name of laudable egalitarian ideals that we are prepared for the second and third order effects of broad global accessibility to AI, including frontier models?
Ok
Wait till you find out how much an apartment costs!
This is a really stupid blog post, I’m not sure why someone decided to post it here other than to ridicule it.
A person in a low-income country would also need to work for 38.6 months ($2400?) to afford to hire one of my electricians for two days of labor. Things in high-income countries are expensive, who would’ve guessed??
If I lived somewhere that the average income is $200/month, there’s a lot of things on Maslow’s Hierarchy that come before ‘ChatGPT Pro’… like um a stable electrical grid, clean water, sewer system, etc.
The running tech bro narrative is that Americans are lazy and not snart.
But tbe chart on that page shows very high productivity
[dead]