Ian Welsh

The horizon is not so far as we can see, but as far as we can imagine

Category: AI Page 1 of 3

Maybe Early AI Adoption Is Stupid?

By

On August 11, 2026

14 Comments

The thing about AI is that it takes previous costs (your engineers and their tools) and makes them more expensive. Often much more expensive:

if you built your saas business before ai, your entire business was designed around one assumption which was that software has ~zero marginal cost.

ai blows that up. every ai action costs money. incumbents now have to create more expensive tiers or introduce usage based pricing. both force customers to make a new purchasing decisions while revenue doesn’t automatically increase just cuz the product became more expensive to operate. so your cost per seat rises, your gross margins compress, & adoption remains uncertain.

meanwhile, the model labs are subsidizing ai usage like crazy, so they can undercut you. ai native startups are burning venture capital to acquire users, so they can undercut you too.

(lack of capitalization in original)

I really don’t know what to think of AI for coding. I’m no longer a coder, haven’t been for almost thirty years now, so I don’t have enough firsthand knowledge, though I did have an AI generate basic sorting and random number generation code just to get my hands a little wet. Among devs I trust, opinion is split. Some love it, some hate it, some are on the margins. Here’s one in the hate category:

Are companies actually seeing massive productivity gains from their AI adoption?…

… All of the AI projects we have observed as a team are failing. Every single one – we have seen 0% success in a year and a half, not only amongst projects we have been asked to participate in2, but even within projects that we have observed in passing while doing totally unrelated work. Even if you grant that AI tooling accelerates specific workloads, the method and scale of the current investments is senseless. Frequently the failure is not related to AI itself, but rather that companies are terminally bad at running software projects effectively, and as I have remarked previously, AI projects are subject to all the failure modes of normal projects plus you can get everything right and then still fail because of the method’s novelty. Very few companies are so good at shipping software that they can afford the extra risk profile…

…

our team has quickly learned while on an engagement not to ask anything about ongoing AI projects in any context – by the time that project has started, it is too late for the management team, and intervention is not possible until a crisis point is inevitably reached. There is no conceivable positive outcome. The failure rate is so high that even basic inquiry leaves us in an untenable position. Any coherent question about how it’s going, what the goal is, who is using it, constitutes an inadvertent attack on the chain of command responsible for the work because there are no good answers to anything. Even in rare cases where my interlocutor has stated that things are going well (usually while the project is still mid-flight and failure has not had a chance to manifest), it is generally obvious that they are doomed, but at least in these cases I can simply agree and then go home to scream into a pillow for six hours straight3.

All of this is to say that I am very confident that almost every report at a company about “massive AI productivity gains” is untrue as a matter of brute fact. Even if some companies are seeing clear gains, this is the exception, not the norm.

It’s a long post and the entire thing is worth reading. This is a high end engineering consultant who gets a view on a lot of companies.

My feeling/guess is that AI is not mature, that the use cases exist, but that it introduces significant risks of new failure modes because AIs aren’t reliable and aren’t actually intelligent: they don’t know what they’re doing. If you don’t do the work yourself or audit it to the level where you might as well have done it yourself, it seems easy for errors to creep in, for hard to maintain or understand code to be created, and for new failure modes to exist.

If I were an executive in most businesses I wouldn’t be using AI for much, if anything yet, though I’d let a few people have an account and test it, and would do so myself. It’s moving fast, it clearly adds risks and mistakes that we don’t know how to mitigate yet, it’s expensive and that’s with massive subsidies, so becoming dependent on it when prices are likely to increase significantly is unwise.

There are going to be exceptions: if you’re Google, not sure you have much choice. But if you manufacture widgets or sell hamburgers or build homes or are a lawyer (who can’t afford to have the AI hallucinate cites) I’d give it a pass for now, or use it very warily in exploration mode. I use it for research myself, especially as search engines get worse and worse, but I also check the sources it uses.

And, as we’ve discussed before, non open models have huge risks, since prices can easily be raised, the government can intervene to cripple the model or deny to other nations, and the company itself can make changes to the model which make it less useful to you. So even if using AI I’d be primarily focusing on Open models (which means mostly Chinese.)

But my best guess is that this is a real tech, with real uses, which is not yet near to mature and which is being deployed before people understand what it’s good at or how to use it safely and cost effectively, or, more importantly, when NOT to use it. I’d also guess that at least so far, it’s not the second coming, the next great thing, in the way that its evangelists preach.

So far it seems to increase failure rates on real projects and raise costs significantly at the same time. Unless I’m running a business where I must be in it, I think I’d take it slow.

There are very few businesses where “the way we did it in 2022” will expose you to significant risk and costs. If you’re in one of those businesses, chill a little and observe. Let other people pay the price and make the mistakes. If you need to, because it’s the current “everyone must do” just lie and say you’re into it, while doing the minimum.

This is certainly something I could be wrong about. I’m confident that Chinese AI will win, for example, I’m less certain about how useful AI will turn out to be. I do think that I’m almost certainly right that most companies adopting it in a big way, NOW, are making a mistake. Wait. See how it turns out. Let other people figure out what it’s good for, how to manage it, how to mitigate the risks and how to drive costs down.

And remember, even if it’s the true next big thing, early rushers often get badly burned, as with the dot-com bubble. Don’t buy hype, make sure what you’re investing in is real and you understand what it’s good for and what it sucks at.

What I write here is for the benefit of everyone, but alas, I live in capitalism and I, and the site, take money to keep running. If you value the writing here and can, please subscribe or donate.

Burning Down The Great Library Yet Again

By

On July 29, 2026

21 Comments

Sigh:

AI companies are bulk-buying rare books, scanning them through high-speed machines that cut the spines off, and shredding the originals. A service called ISBNdb facilitates orders of up to a million books and keeps buyers anonymous. Pre-2022 books are premium because they’re free of AI-generated text. A federal judge ruled the practice is fair use because eliminating the original means only one copy exists at a time. Anthropic hired the former head of Google Books partnerships to obtain “all the books in the world.”

My Take This got to me. A bookseller told 404 Media that rare books with almost no surviving copies are being fed into this pipeline. Books that survived wars, fires, and centuries of handling are being shredded so an AI can learn to write a better marketing email. ISBNdb’s website literally says “‘AI company destroys two million books’ is not a headline that generates sympathy,” and they still built an entire business around making it happen quietly. They offer NDAs as a feature. They coach clients to call it “digital preservation.”

I’ve covered AI companies scraping the internet, torrenting libraries, and stealing music. This is worse because it’s irreversible. You can re-upload a website. You can reprint a bestseller. You can’t replace the last three copies of an 18th-century botanical text once someone shreds them for training data. And the judge said it’s legal. So it’s going to accelerate. “We shred rare books and offer NDAs so nobody finds out” is a legitimate business model in 2026. What a timeline.

These people are barbarians. Philistines. As bad as destroying the library of Alexandria or Mongols burning libraries in Baghdad.

This knowledge is irreplaceable and it’s not worthless (if it was, they wouldn’t be scanning it.) If it were up to me this would be a capital crime. Not exaggerating for effect. Everyone involved would go for the long jump.

(To give credit where credit is due, apparently Musk does not shred books for his AI. He’s scum in many ways, but one should praise the correct actions of bad men.)

We all have a limited time here, to destroy that which could live much longer than us and speak of the past to the future is a monstrous crime.

Truly most of the people who run AI are scum.

 

What I write here is for the benefit of everyone, but alas, I live in capitalism and I, and the site, take money to keep running. If you value the writing here and can, please subscribe or donate.

America’s Losing the AI Race Hard

By

On July 21, 2026

32 Comments

Amazing stuff, and sooner than even I expected. China’s Kimi is now about equal to GPT and Claude:

For the first time, a Chinese model Kimi K3 has taken on the Frontend Code Arena and is scoring at or near the frontier on other benchmarks.

And the majors are noticing:

Meanwhile Trump is talking about banning Chinese AI. 

Alternatively, the smart lads at Open AI have a regulatory plan:

I would guess that the Trump Administration will at some point realize that their best strategy here would be to create large amounts of regulatory risk around the use of open-weight Chinese models. You don’t need to “ban open source” (one of the dumber motifs of AI policy discussion). You just need to direct every agency to issue soft law that creates FUD. “A Federal Reserve Advisory Bulletin found that there may be backdoors in Chinese AI models.” It needn’t be that well justified. You just create enough regulatory risk that every regulated enterprise backs off. You probably don’t want to create so much regulatory risk that you scare off the hyperscalers from serving Chinese models; this will just drive startups to sketchier providers. There’s a happy middle ground here. I’d assume they will do some version of this.

Even more fun, the Feds want to approve all AI releases and not allow everyone the full models.

The Trump administration has taken new steps to assert more control over the rollout of future artificial intelligence model releases by dictating which companies and entities are allowed access to the latest frontier models, two people familiar with the matter told CNBC…

….

This week, the administration launched its own program, dubbed “Gold Eagle,” aimed at collaborating with the private sector to find and fix cyber vulnerabilities.

The so-called clearinghouse would put the White House in charge of greenlighting which companies can access new AI models, according to a person familiar with the matter, who spoke on condition of anonymity in order to discuss information that is not public.

China is going to eat America’s lunch on this. If they ban Chinese AI (harder than it seems, given it’s open source) all that means is writing off the rest of the world. And since American models are handicapped, smaller American companies will be stuck with worse AI.

Since Chinese AI is far cheaper, as well, I’d expect American companies to set up subsidiaries overseas to use it, rather than be stuck with American AI.

The entire situation is a complete clusterfuck. Major companies have taken on serious debt in order to build data centers which have a lifecycle of five to seven years, and often less (since new generations of GPUs are much better.)  But the Chinese product is cheaper, open source, lacks nearly as much sovereign risk and I’ll bet multiple models will soon be about as good as Anthropic and OpenAI’s.

Where’s the business case that spending all these trillions of dollars is going to produce enough revenue from US AI to pay for all of it?

There isn’t one. It doesn’t exist.

And that means that, at least in America, this is an AI bubble. All bubbles burst and this will not be an exception. If the government bails them out it will be the last major US bailout.

This is also very likely one of the last major tech revolutions which will start in the US (which it did.) Going forward they China will produce the vast majority of them.

This is the endgame. The turning point where China obtains not just the industrial base but the absolutely undisputed tech lead. From now on China will like America in the 1950s — it’s where almost everything new is created, the dynamic center of the world, and soon people will be competing to move there, because everyone knows it is the future.

 

What I write here is for the benefit of everyone, but alas, I live in capitalism and I, and the site, take money to keep running. If you value the writing here and can, please subscribe or donate.

 

Xi Lays Out His Principles for AI Development

By

On July 17, 2026

16 Comments

 

There were rumors recently that China was going to put export restrictions on AI. That now seems… unlikely. Here are the principles Xi laid out:

First, adhering to the principle of openness and win-win cooperation while boosting innovation-driven development. Xi highlighted the importance of encouraging open-source, openness, collaboration and sharing to facilitate technological innovation, industrial development and scenario-based application of AI. (My emphasis.)

Second, strengthening risk awareness and ensuring that AI is secure and controllable. Stressing the need to ensure that AI is always under human control, Xi urged all sides to jointly oppose overstretching the national security concept in the field of AI or placing one country’s security over that of others.

Third, encouraging inclusiveness and promoting mutual learning among civilizations. AI development and its application should not erode or undermine the diversity of world civilizations or the uniqueness of cultures of different countries, according to Xi.

Fourth, advocating solidarity and improving global governance. The important role of the United Nations should be recognized, Xi said, calling for further alignment and coordination on AI development strategies, governance rules and technical standards.

I’ve predicted, for a couple years now, that Chinese AI models will be the main models used in most of the world, including in much of the West, assuming they aren’t banned outright, because they’re open and cheap. Costs of running them are about twenty times lower than the US frontier models made by OpenAI and Anthropic. They’re almost as good, and they aren’t that far behind.

The problem with US models is not just that they’re expensive (though that’s huge, there are tons of reports of AI use being cut back) but that they are CLOSED: meaning you can easily be cut off, or have prices raised, or have the model changed on you with no recourse. Open models you can adapt the model, you can run it on your own servers, or various server companies can, will and do run them for you on their servers which you rent.

It’s clear that Xi gets this, and thus that the CPC understands it as well. Open Source isn’t a liability, there’s a reason why Linux runs most of the world’s servers: closed tech is the liability. Open Source is the advantage.

Notice the second bit: on AI always being under control. I wonder if Xi is thinking of Israel and the US for military targeting and how that has possibly contributed to hitting civilian targets like schools. (Possibly because Israel and the US are run by psychopaths and I bet they’d do it anyway. But no human in the loop may make it even worse.)

The third principle is about avoiding US (and Chinese) cultural hegemony. A nice thought, and open source certainly could be adapted to different nations and cultures, so that AI models aren’t all giving the same generic results.

Finally, the fourth principle. I’ve always found it interesting just how much China plays up the United Nations. I don’t know if the respect is genuine, but the words are consistent. Honestly, I think the UN should move its main HQ out of America. America’s been pulling stunts like denying diplomats visas. Not sure if it should go to China (though if it did Shanghai or Hong Kong seem like good fits) but they’d be better stewards than the US, and in any case, if the UN is going to be in the most important great power, that’s now China. (That said, I’d favor something more neutral. Perhaps Singapore.)

China just keeps coming across as smarter, more strategic and more human than the West. It’s sad, in a way, but it is what it is.

And I remain convinced that Chinese AI will be the winner over American.

What I write here is for the benefit of everyone, but alas, I live in capitalism and I, and the site, take money to keep running. If you value the writing here and can, please subscribe or donate.

 

The Old Gray Lady Runs RussiaGate 2: They’re Coming for OpenAI

By

On July 10, 2026

12 Comments

Guest post by Nat Wilson Turner.

New York Times: "China, Russia and Others Seek to Inflame Debate Over A.I. Data Centers"

Thursday’s New York Times brings back their old RussiaGate spirit with a front page banner headline about “foreign interference” and data center opposition.

Here’s a key quote:

…a push by foreign adversaries to seize on what polls have shown is deep ambivalence — verging at times on hostility — about the spread of the data centers needed to power A.I. in the United States and elsewhere.

China, Russia and, to a lesser extent, Iran have sought to use state media outlets to turn the controversy over data centers in the United States into “a domestic fracture point,” according to a new analysis by Alethea, a threat intelligence company, which identified scores of articles and posts on social media this year.

These campaigns, whose impact on public opinion remains to be seen, have raised alarms in Washington, where A.I. is seen as a top issue heading into this year’s midterm elections.

The foreign efforts appear intended to stoke the debate over data centers that has united political figures across the political spectrum — from Senator Bernie Sanders of Vermont, a progressive, to Stephen K. Bannon, the erstwhile adviser to President Trump.

“Foreign actors aren’t manufacturing American debates over the future of A.I., they are exploiting them,” said Jessica Brandt, a former official with the Office of the Director of National Intelligence who tracked foreign influence efforts during the Biden administration.

The goal, she added, is to “deepen our divisions in order to dent our appeal and weaken us from within.”

Interesting sources, some company called Alethea and a former Biden admin DNI spook as our sources.

We’ll come back to them later, but first I’m curious as to why the NYT is only now covering this story when OpenAI put out a press release saying basically the same thing on June 10 except focused solely on China.

After all, OpenAI’s report was convincing enough to sway such luminaries as Senator Tom Cotton (R-AR), Republican Leaders on the House Energy and Commerce Committee, Rep. Brett Guthrie (R-KY), Interior Secretary Doug Burgum, The Bitcoin Policy Institute and prominent tech investor Kevin O’Leary, per WIRED.

Read More

Microsoft To Offer Deepseek Based AI Copilot

By

On June 17, 2026

4 Comments

Regular readers will know that for a couple years I’ve been saying that Chinese open source AI would win the AI “war” because it’s cheaper and non proprietary (prices can’t just be raised suddenly, or capacities taken away.)

Over the last few months there’s been a lot of screams coming from regular AI users. OpenAI and Anthropic moved to token based billing, which is to say “you pay based on how much you use.” They still weren’t charging full rate, but they were charging a LOT more and users were not happy. One company spent 500 million by mistake: they forgot to put limits on how much their employees could spend.

Oops.

Nor are ordinary users exempt:

I Went From $3,000/Month on Claude to $5/Week on DeepSeek

And honestly? 80% of my work is identical.

For the past two months, I was burning $3-5K monthly on Claude Code. Every idea from design to development to testing – full end-to-end automation, even simulating users to test my products and provide feedback. Extremely token-intensive.

But Claude’s caching sucked, making it insanely expensive. Then I discovered DeepSeek V4.

The numbers: • Claude: $5 input, $25 output per million tokens •

DeepSeek: $0.28 input, <$1 output (with their current discount) • DeepSeek cached: $0.0002 – literally less than a penny The caching optimization is game-changing.

Once DeepSeek has seen content, it basically stops charging tokens. My result: $5/week vs $1,000/week for the same workload.

So now Microsoft has created their own minor Deepseek fork, and will run it on their servers to power Copilot. You can still use a version run by US labs, but if you can’t afford, or justify that, you can use the Deepseek version.

Driving the news: Microsoft says companies using Copilot Cowork will pay based on how much compute they use.
  • The company tells Axios it is exploring a fine-tuned version of DeepSeek V4, or another open-source model, as a lower-cost alternative to the Anthropic and OpenAI models now powering Copilot Cowork.
  • Microsoft says it expects to make a lower-cost model available in the coming weeks and will confirm its choice then.

Worse than this, there’s beginning to be serious pushback on whether AI is all that useful. Uber’s COO opened the door back in March:

In perhaps the most high-profile example of this growing concern yet, Uber COO Andrew Macdonald acknowledged during a recent podcast appearance that gains in productivity simply weren’t being reflected in the oodles of cash the company has been shelling out on AI.

“That link is not there yet, right?” he told Rapid Response host Bob Safian. “I think maybe implicitly there is more that is getting shipped, but it’s very hard to draw a line between one of those stats and, ‘Okay, now we’re actually producing 25 percent more useful consumer features.'”

“If you’re not actually able to draw a direct line to how much useful features and functionality you’re shipping to your users that trade becomes harder to justify because it’s not free,” he complained. “AI is not free.”

As far as I can tell there’s little evidence that US priced AI is more cost-effective than the employees who were laid off because it was so great. I rather suspect that in most cases, it’s less cost-effective.

But more importantly we have the “it’s better to be wrong with the crowd” effect moving against AI. In almost all positions, including executive ones, if you’re wrong in the same way that everyone else is wrong, it’s no big deal. If you’re wrong against the crowd (say not getting into AI when the rest of your industry is) and it turns out that AI is the next big thing, well, you’re fired.

So much of the AI mania was driven by this and a relentless hype cycle. Now that important people are beginning to push back on it, it’s no longer required to be all-in on AI. And that’s bad for Anthropic and Claude.

AI is not the next coming. It is not going to make it to general AI (not this generation of large language models anyway) and while it does have some utility the US frontier models cost far more to operate than any conceivable return most of their customers will receive. It isn’t the “get rid of three-quarters of your employees” super app corporate leaders were promised.

And to the extent it is useful, well Chinese open source models are more cost effective. As good? Generally no. But they keep catching up, and paying 70 to 97% less makes up for being somewhat behind.

So to the extent that AI is a real industry, odds are high China’s going to win the race. Since the models that will win will be built off open source models that’s not a crisis for anyone, it’s a good thing, far better than a proprietary future.

BUT it does mean that US AI expenditures are probably going to turn out to be the biggest misallocation of resources in centuries: bigger than the housing bubble and bigger than the dot-com bubble (which at least did have a world changing technology behind it.) Not quite the Dutch tulip bubble, but at least the Dutch got lots of pretty flowers of that, instead of massive ugly data centers.

Business is driven by stupid people engaged in group think, especially in the West, far more than most people will admit. Everything Silicon Valley does these days is someone trying to create a monopoly or oligopoly so they can be insanely profitable, while China actually competes on price, and that’s why China keeps eating the West’s lunch.

I’d cry, except that an open source AI world is a far better one than a proprietary one, and every tear some Silicon Valley tech bro cries over a lost opportunity to make a monopolistic buck an angel gets their wings.

What I write here is for the benefit of everyone, but alas, I live in capitalism and I, and the site, take money to keep running. If you value the writing here and can, please subscribe or donate.

 

AI: Make Them Stupid, Then Sell Them Brains

By

On June 9, 2026

30 Comments

The evidence on AI’s effect on those who use it has been coming in, and it’s not good. While it doesn’t effect everyone, it seems to effect most people, and the worst affected, it seems, are the young. Olds have the advantage of growing up in world where they had to learn how to do things themselves. To be sure, phones and social media seem to have had a negative effect on attention span and learning ability, but AI is yet another assault, and it hits the young hardest.

This excerpt comes from a larger piece from a university professor on the inability of his students to read. The whole thing is worth reading, and the decline is truly precipitous: fundamentally most of them can’t read an entire book, and struggle even with long articles, and they can’t pull out the arguments made. The bit on AI follows:

Another reason for the decline in student reading capability is increasing reliance on generative AI. In June 2025, Nataliya Kosmyna and colleagues at the MIT Media Lab released a preprint titled “Your Brain on ChatGPT.” They divided 54 participants into three groups writing SAT-style essays — one using ChatGPT, the second group using a search engine, the last group using nothing — and monitored brain activity with a 32-channel EEG. The ChatGPT group showed the lowest neural connectivity of the three, with up to 55 percent reduced connectivity compared with the brain-only group, and “consistently underperformed at neural, linguistic, and behavioral levels.” Eighty-three percent of LLM users could not quote a single line from essays they had written minutes earlier. When the LLM group was forced to write without AI in a follow-up session, their brain activity did not bounce back to baseline; the researchers coined the term “cognitive debt” for the lingering deficit.

The fundamental strategy of a lot of tech startups has been to degrade pre-existing infrastructure by under-pricing, for years if necessary, until the old methods are so diminished that they can start charging monopoly pricing. Uber is the classic example: Ubers were far cheaper than taxis for about a decade. Now they’re often more expensive, if the taxis exist at all. Certainly where I live in Toronto, the Taxis did somewhat survive, and cost less.

But overall the strategy was a success, taxi companies were devastated and Uber’s doing great now. All it took was years of losses and predatory pricing: their model wasn’t superior, their product wasn’t superior except having a good app, but they had far more access to patient money, willing to take losses for years to get to the oligopoly pricing end-state.

Neither Anthropic nor Open AI are remotely profitable. Every single query costs more to run than is charged, even to paying clients. A recent increase in prices, still far below running costs, has hit users with massive bills. There’s no evidence AI is better than humans at most tasks, and the real cost (and sometimes, even subsidized, the current subsidized price) is higher than just having employees. AI is often faster, but it makes mistakes humans don’t, and needs to be checked.

But if you make your employees use it they’re going to be degraded and lose the ability to do their jobs well. The more you do something, the more your body and brain optimize for it. The less you do it, the worse you get.

AI’s strategy for replacing workers is threefold: first, sell executives on getting rid of pesky workers for AI, because it’s supposedly easier to manage.

Second: Subsidize while companies lay off the workers and replace them with AI. Once the workers are gone, jack up prices; and,

Third: by encouraging companies to force workers to use AI and to replace workers with AI in some cases, make the workers less capable: stupider. Over time as more and more people become dependent on AI to think and work for them, they will lose the ability to do the work themselves. AI may be shitty, but it will be better than the dullards AI makes its users into.

It’s an ingenious strategy, really. Make people stupid, and replace them with a product which costs more and is inferior to them for most tasks before they were made stupid.

The longer term issue will be that AI isn’t creative: it uses the embodied creativity of past humans, in terms of their writing and their discoveries to simulate intelligence. But as humans produce less and less new creative work, AI will be reduced to eating its own results, and indications are that leads to model collapse: AI’s are dependent on human, and by making humans redundant and stupid they will themselves become stupider and less effective over time.

We live in a time where we can’t look ahead, ever, at technology and make even the smallest effort to control the end results, it seem. At least in the West. Or, rather, we refuse to deal with obvious negative issues if doing so means a few people won’t be able to get as filthy rich.

Dumb.

And soon we’ll be even dumber.

What I write here is for the benefit of everyone, but alas, I live in capitalism and I, and the site, take money to keep running. If you value the writing here and can, please subscribe or donate.

Western AI Investors Are the Dumbest Money In The World

By

On May 27, 2026

34 Comments

Regular readers will know that ever since Deepseek came out, with token costs of three to five percent of US models, and open source which you can run on your own servers, I’ve been saying that China would win the AI war.

Here’s another data point:

I Went From $3,000/Month on Claude to $5/Week on DeepSeek

And honestly? 80% of my work is identical.

For the past two months, I was burning $3-5K monthly on Claude Code. Every idea from design to development to testing – full end-to-end automation, even simulating users to test my products and provide feedback. Extremely token-intensive.

But Claude’s caching sucked, making it insanely expensive. Then I discovered DeepSeek V4.

The numbers: • Claude: $5 input, $25 output per million tokens •

DeepSeek: $0.28 input, <$1 output (with their current discount) • DeepSeek cached: $0.0002 – literally less than a penny The caching optimization is game-changing.

Once DeepSeek has seen content, it basically stops charging tokens. My result: $5/week vs $1,000/week for the same workload.

The Chinese have been optimizing for efficiency. Frontier models are a little better, at each generation, but the Chinese aren’t sitting still on quality either. What will happen is convergence: where all the models are about equally good. Until then, the Chinese models may be 3 months to 6 months behind, but that’s all and when they have the advantage of being, not just 95% cheaper as before but over 99% cheaper, and when they’re open source, so the vendor can’t just increase prices or restrict usage without users having any ability to run the model themselves, well, only an absolute moron would plan on using Anthropic/Claude or Open-AI.

The West is building vast numbers of data centers, far more than the Chinese, because nothing is optimized for efficiency. Our models use far more electricity, far more GPUs, and far more water. Additionally the Chinese are working hard on real-world AI uses: robotic AI, in other words, so that their AI can be used for actual production, and pushing on humanoid robots so they can take care of their old people, do household work and so on: Chinese AI is optimized to do shit work so you can read and write and paint, while Western AI is optimized to do creative work so you can shovel manure, do your own laundry and clean toilets.

I simply cannot fathom; I literally spent 5 minutes repeating to myself, “how can they be so fucking stupid?” what American tech leaders and the people throwing money at American AI companies are thinking. They lose money with every single query. Starbucks just got rid of their inventory AI because it couldn’t tell the difference between oath milk and cow milk even after months of trying to make it work. The only large use-case for American AI is programming and even there quality questions are rife.

But even if it winds up being as great as they say, they’re still going to lose, because it’s not going to be enough better than Chinese AI (if it’s better at all in a few years, which I doubt) to make up a 95 to 99% cost difference plus the safety that open source provides.

We are ruled by morons. Our rich people, whom we celebrate as brilliant, are idiots and fools. The moment Deepseek came out, the response should have been to figure out how it was so much more efficient and add that to American AI. No one even considered it, they just threw hundreds of billions of dollars more money at doing the same thing they were doing before.

You can’t look at American AI, as anyone who uses is, and not realize it’s massively subsidized and that at some point they’re going to have to charge the actual, real cost. As with everything else, the plan is to destroy alternatives, then once they’re damaged beyond repair (like Taxis and “rise sharing”, raise prices.”

But this won’t work with AI because the Chinese aren’t playing along and it’s software that can be run on any server, if you have access, and, again, Chinese models are mostly Open Source and way, way cheaper.

I’m shaking my head as I write this. This is the greatest mis-allocation of resources I’ve seen in my entire life. It makes the housing bubble (out of which we at least got some homes) look brilliant and wise.

Morons. Our leaders are morons. And this isn’t even a case of “well they’ll do well out of it and they don’t care about anyone else”, it’s a case of “they’re going to lose hundreds of billions of dollars.”

One more bailout. Maybe. This is the last time.

And the West is DONE.

Because we aren’t even ruled competent evil people. We’re ruled by utter idiots.

(I wanted to write that they’re the dumbest money in world history, but, alas, that still goes to the Dutch tulip bubble. Though, again, at least they did get a lot of beautiful flowers out of it, which is more than we’re going to get.)

What I write here is for the benefit of everyone, but alas, I live in capitalism and I, and the site, take money to keep running. If you value the writing here and can, please subscribe or donate.

 

Page 1 of 3

Next

Powered by WordPress & Theme by Anders Norén