> Because it wasn't about "engineers should write 1M LOC per month of product code" it was "we want to scale automated porting of code to safe languages so that 1 engineer managing 1M LOC of automated conversion can work"
These are one and the same. Whether it's ported code or not doesn't change that. The framing device also doesn't matter, because it's the exact "Oh it's our goal" shtick that executives use in the former's case.
"It's just a measure" doesn't cut it in a world where every single AI measure immediately gets turned into a target by executives greedy for efficiencies that don't exist.
EDIT:
Right, I forgot. This is HN where everyone is a galaxybrain and "Port a million lines of code per month" is a totally reasonable goal for a single individual.
I can easily game writing 1M LOC per month by having the LLM write code in more verbose ways, with useless indirections and abstractions thrown in for good measure. I could even ask claude to write code that does nothing but just takes up line.
In contrast, converting 1M LOC of code per month is a much more solid measure, as long as you measure LOC of the source, not the new code. Sure, in the short term you can pick the easy/verbose things to port, but it's hard to do sustainably. A 5M LOC code base would still be expected to be ported in 5 engineer months.
Granted, you can still rush the work, not test properly, neglect good planning and engineering. Ported lines of code should not be the only measure (just like with any other measure). But it's a much less problematic measure than coding 1M LOC
Define "realistically". You're basically saying attention is all we need indefinitely into the future and all other gains come from more compute or scaffolding around current architectures.
Attention is all we need because it is currently the best parallelizable way to model long-range dependencies on current hardware constraints, not because flat tokens yield some natural law of intelligence inherently.
Who's to say we won't find a way to encode provenance or privilege natively into models such that the tradeoff changes?
It's hard to say what the solution will be. If I knew it, I'd build it. But it's even harder to sustain that the current architecture is a crystalized global optimum.
Aside from LLM architecture, that already is a complex issue, an issue is that training data is unstructured text.
An LLM able to structurally separate context and instructions, should logically need separated data to train, and we don't have it.
Moreover, while an equally powerful LLM architecture solving this may exists, there are no guarantees at all that we are able to come up with it in a reasonable timeframe.
Without some signals moving in that direction, the most pragmatic and realistic way of looking at the problem is that it will not be solved in the near future
I agree this doesn't mean we shouldn't try to address limitations with the current architecture. I just mean that I expect the root cause to be solved eventually if we ever really want to take steps towards AGI.
The other comment got the answer already, but yes. It's a cost problem.
LLMs are designed this way so they could be trained off unstructured text, which critically can be obtained by just scraping things off the internet.
The moment you change anything about this, you incur the trillion dollar cost of needing to manually curate the training data.
There's some attempts to get around this problem with synthetic data, but they're running into problems with model collapse (Maybe severe performance degradation is worth the security tradeoff?) and the politics of AI; All major AI companies highly restrict using their systems for synthetic data & AI training, and they're too busy themselves to investigate exotic approaches.
Hence: Realistically, this is just a problem AI will have for the foreseeable future. There's no fine tuning that can fix this, nor can a new model be easily trained with these properties. The costs are just enormous right now.
This might sound crazy but I think embodying the AI will be the long term solution here. When AI robots use language to relate their experiences and make predictions about the real world they are walking around in, it will prevent the model collapse problem. Their language might diverge from human language, but since we live in the same world translation should be possible.
Edit: Actually, I think that with a fairly small amount of auxilliary data, it could be ensured they keep the ability to speak English.
I think you misunderstand why tech debt lingers around. It's not a capacity or capability problem.
Organisations just don't want to deal with the accountability involved with "touching cold code". Whether it's a human or "AI agent" doesn't change the "It worked in prod, you touched it, you broke it, never touch anything again" dynamic.
That's one dimension of it, but in the context of this thread we are talking about how maintainable a codebase is for other humans. If your codebase is messy you depend on a few key employees and it might be hard to onboard new ones, so there has always been financial incentives to reduce tech debt.
Um, no, actually AI makes it better because the cost is lower now. I'm not sure what point you're trying to make here, obviously organizations already fight against tech debt all the time through a variety of means?
The point there is that it is MUCH easier to get corporate to agree to something when the cost is nebulous and being paid anyway. If you get a senior dev to clean up some tech debt, how much did that cost the company? The dev will have some multiple things at the same time, so you can't cleanly assign a number of hours, maybe multiple people are involved. It's practically just an unknowable. Practically, $0.
> Sure, but it remains a big enough problem that human intervention and review is still necessary for any serious work across all use cases and industries.
Another important consideration: Hallucinations getting less common/severe but not (as-good-as) solved makes them worse.
LLMs used to very obviously get things wrong. And people wouldn't trust them. Now they're good enough that people blindly trust them.
Now people just directly PR AI output with little to no manual review. We even have clowns calling for the complete abolition of directly human-authored code.
Whatever gains were had in better AI code output over the past two years I lose in having to review much more thoroughly.
> He would begin somewhere–statistics on AI demand, say–and then walk the calculations carefully over to the next step–maybe revenue needed for profitability by AI companies–and you could follow the argument.
That's exactly what the first (titled) section does?
Haha thought you were referring to the upsell at the start asking to subscribe to the newsletter for $70 / year. But yes it does call out the unprecedented amount of money getting dumped into AI.
What turned me off though was this paragraph:
> This is a hysterical era perpetuated by liars, cowards, imbeciles, craven boosters and the easily-fooled. Those excited about generative AI are either the victim or the perpetrator of a con centered around a technology to ingratiate at the highest cost possible.
That's a very bold claim. Really anyone excited about generative AI dude? That's just an absurd claim, and makes it sound like he hasn't used an LLM since GPT 3.5. It's just the language is so hyperbolic and angry that it's giving me more rant vibes that really hurt the tone and damage the (many valid) claims he's trying to make.
Really tried to read through this all the way, but man I'm just not in love with this guy. I feel like the frustration is clouding his judgement. This line is another one with a fact that isn't really grounded:
> so, you know, they only need to grow by 496% by the end of 2029!
Which isn't wrong, but also Anthropic's revenue increased from $1 billion in Dec. 2024 to $47 billion May of 2026. Which of course doesn't guarantee that it will continue to grow at that scale, but it's clear that there is a strong demand for what they are creating.
Idk, not really sure what my point is here. There are just so many facts and numbers quoted in here... It's a bit exhausting to refute a piece like this, when parts are genuinely correct, and parts are maybe subconciously exaggerated due to some emotional leaking into the argument.
> ... Anthropic's revenue increased from $1 billion in Dec. 2024 to $47 billion May of 2026. Which of course doesn't guarantee that it will continue to grow at that scale...
My 2 year old is 80cm tall now, and was just 50cm tall 18 months ago! So it's not wrong for me to expect him to be over 2 metres tall when he's 8...
I just woke up and THIS! ... you almost owe me a new keyboard! I love it!
This statement cleanly encapsulates the entire problem with all of the frontier models' companies' pre-IPO numbers.
They have something-something "new technology" and we don't know anything about how the market is going to settle on the ethics, the utility, the human capacity opportunity cost impacts of not training and/or mis-educating an entire cohort of intern-engineers for a few seasons to a generation, the full environmental costs of hardware and operations necessary for the training each new larger model, ... and we cant even quantify the unknown-unknowns - the risks we cannot forsee.
To predict market revenues for the next few years based on the curves, that they self report without external disclosure of the underlying numbers, is just like expecting your 2 yr old to continue growing at the same pace in the future and in the past - laughable. Good thing it was just water not coffee and it didnt quite come "out my nose" :- ) Thank you kind stranger!
Dividing by 12 you still have the same problem. They're projected numbers as opposed to real ones as well as being grossly skewed by any short term fluctuations.
Divide both by 12 and you do not get the projected numbers. You get monthly revenue, a real measured number. It is the number being reported * 12 when they state a new ARR.
E.g. When Anthropic stated $1B ARR (an extrapolated value) what they were actually reporting is $(1/12)B Monthly revenue. If it helps their current monthly revenue is 47 times that, for a grand total of $(47/12)B per month in revenue.
Yes it is the current monthly revenue which is a projected number as far as the other 11 months go. That's fine if the overall economy has low volatility, your sector is well established and predictable, and your company isn't undergoing any significant changes. Absolutely none of that applies to the frontier AI labs.
Well, he dismisses any value whatsoever to GenAI. That's immediate bozo bit criteria to me. And, well, if Anthropic revenue doesn't grow 5x between now and the end of the decade, I'll be pretty surprised. But, sure, if it doesn't, then someone will keep them around anyway. AMD almost died in the 2010s as one example, but they kept getting propped up and now they're back in the game swinging. There are people who can see alpha beyond the next 10Q. Ed Zitron isn't that sort.
> Well, he dismisses any value whatsoever to GenAI.
I didn’t read it that way. I see a lot of value in it.
I just don’t see us justifying the amount of infrastructure being built or current valuations. Or in the unlikely event that we do, the societal upheaval is going to take away the ability to monetize it meaningfully.
OpenAI and Anthropic may make it through. But that is different from saying valuations are justified or that all this infrastructure will pay off.
"Those excited about generative AI are either the victim or the perpetrator of a con centered around a technology to ingratiate at the highest cost possible."
How else would you read the above statement? He's just preaching to his own choir IMO.
My take: like any gold rush, a lot of dumb ideas will get backed and they will all fail. And then we'll keep the ones that worked. SSND. Good luck picking the winners a priori.
I read it in context as being about the market prospects of genai.
The problem is, when there is so much overinvestment, everything gets wrecked. In the aftermath of the dotcom boom there was at least a bedrock of fiber and still useful equipment to build upon amid the rubble. This time we are going so much further; also many of the durable assets are misplaced bets and the depreciating ones will depreciate more steeply.
Someone should do the analysis of a decade and a half of Nvidia datacenter GPUs from Fermi to Kepler to Maxwell to Pascal to Volta to (Turing) to Ampere to Hopper to Blackwell and generate some hard depreciation numbers. Fiddling around a bit, 16-20% annual depreciation (so 5-6 years total and then any further revenue is bonus goods) it would appear, but that's a fiddle number.
But confounding this, K80s and V100s are still offered by cloud providers 13 and 9 years after their releases and academia still loves their GTX 1080 Pascals in their desktops. At companies, the beancounters take a computation and find the best architecture !/$ for that calculation. It does not need to be brand new shiny. It's Nvidia's job to make that case, not them. But anyway, the real data is right there. And those old GPUs demonstrate the dark fiber is already in place (and it's not so dark or they'd pull their racks).
AI is the special case. New GPU generations are the only way to access HW implementations of last year's research on precision modes and matrix math. If that slows down, that would be the first real bellwether of a slowdown. It hasn't happened yet. I'm a little surprised myself, but I also think coding agents are the vanguard of general design agents and that's going to hit a lot of industries at once.
So as long as the next generation of GPU halves the price of tokens and doubles throughput (or better), the demand for tokens will continue to rise IMO.
What I don't think is that AI can come for anyone's job successfully no matter what the C-suite sorts insist.
In summary, if you're a bear, you can point to the depreciation cycle and scream the sky is falling. And if you're a bull you can point to GPUs staying in production for a very long time despite the depreciation. Guess we have to wait for 2030.
Sure, according to an unnamed "GenAI principal at Alphabet" of whom "We could not verify the name of the person who describes themselves as 'GenAI principal architect at Alphabet' and therefore we cannot 100% trust their claims."
But let's run with Deep Layer(tm)'s hot take from 2024, GPUs all die in 2 years, no exceptions*. Poor guy just spent $400,000 on a DGX with 8 B200s, each of those B200s generates a piddly ~$3,000 in profit monthly spewing tokens, netting $576K in 2 years, that's a pathetic 20% annual return. Oh no... Won't someone call Michael Burry!
*Never mind the 3-year warranty or any extended service contract, that GPU is D E D and you're S O L.
You gotta include 50k+ in power plus other expenses. Still looks like an ok return on capital, but you are betting heavily that cost of computing doesn’t fall much.
If token prices fall a bunch then it may not even be worth leaving on, depending on your facility’s relative power, cooling costs.
If we push far into oversupply eventually a bunch of firms building this infrastructure are going to lose out.
Chinese providers realized that LLMs have peaked and have started trying to reduce the price per token. Deepseek pro v4 can easily add tests to my complicated code and costs cents for a million tokens.
I can ask Claude or ChatGPT architecture questions and then use Deepseek for the rest.
How are these businesses going to pay to price of energy and GPU depreciation again?
I love nonsense like this. If using a larger model to plan a chain of thought task for a smaller model works, what makes you think Anthropic can't do the same thing and offer it as one of their effort level settings? It reminds me of all those fallen AI ASIC startups insisting they can crush Nvidia until they found out the hard way that the rules of the game are dynamic.
The real challenge IMO is whether enterprise will want to run the models on-site for 100% security and privacy, but even then, what stops Anthropic from offering such an option on-prem or in the cloud?
He implies $400 billion in revenue by the end of 2029 is unrealistic when in fact it's very doable if you look at the trajectory of this technology since ChatGPT 4.0 launch. Google and Meta bring in around $500 billion in ad revenue between two of them annually. ChatGPT will easily bring 100s of billions in ad revenue if fully monetized given 1. it has billion weekly active users 2. ChatGPT conversation provides even better context for ad targeting vs search or social media. Enterprise AI revenue is going through the roof already, and with computer use companies will literally be able to fire large percentage of white collar workers and replace them with AI agent without updating their software infra.
Does that '100s of billions' come from a big bucket somewhere called 'spare cash', or does it correlate to a commensurate reduction in the 'around $500 billion in ad revenue' that Google and Meta are extracting?
Do your assumptions - " if you look at the trajectory " - factor in a slowing economy, a slowing growth in quality improvements in the tech, and/or the asymptote of market saturation for punters happy to stump up more than $50 a month?
What about a few hundred billion in salary and benefits reductions due to mass layoffs?
Not saying this would be good (qualitatively) or even good business in any sense, but we’ve already seen companies willing to sacrifice headcount to cover CAPEX for these models.
A few hundred billion in salary and benefits reductions equates to millions of layoffs. At minimum, we'd be looking at something about the same magnitude as the 2008 financial crisis. That scale of workforce reduction would have profound implications for the broader economy.
In a consumption-driven economy, businesses need consumers. Any gains from these layoffs would be short term at best.
I mean it almost certainly won't increase unless a major company takes out substantial debt, in which case we just kick the can and have conversations about bigger numbers. I don't quite think you understand, where will these hundreds of billions come from? By 2029 we will be well into a hardware glut and people will run their own models. Anthropic doesn't have the data flywheel to compete with OpenAI or Google. They went all in on special purpose AI and hit a brick wall and had a "do as much evil as possible" strategy which didn't pay off. Hopefully they fail before they get the entire industry regulated.
> Haha thought you were referring to the upsell at the start asking to subscribe to the newsletter for $70 / year.
People like you would be why I put "(titled)" in the reply.
> That's a very bold claim. Really anyone excited about generative AI dude? That's just an absurd claim, and makes it sound like he hasn't used an LLM since GPT 3.5. It's just the language is so hyperbolic and angry that it's giving me more rant vibes that really hurt the tone and damage the (many valid) claims he's trying to make.
The premise is that AI is significantly more expensive than current subscription & token fees. Within that framing, yes basically all AI users are getting conned. Tricked into redesigning their workflow around an unaffordable technology, in the hopes there will be too much sunk cost and they'll just eat a thousands-a-month fee.
> Which isn't wrong, but also Anthropic's revenue increased from $1 billion in Dec. 2024 to $47 billion May of 2026. Which of course doesn't guarantee that it will continue to grow at that scale, but it's clear that there is a strong demand for what they are creating.
"Doesn't guarantee it will continue to grow" is an understatement.
Let's take a generous assumption of the average subscription; $1000/month/seat. This will be quite a bit higher than pretty much everything but hardcore software dev, we'll re-do the math with $200 in a moment. Let's also grab Ed's $60B figure for both Anthropic/OpenAI, as it's more generous.
That's 30 million subscribers for Anthropic, 30 million for OpenAI, 60 million total.
They need to 5x. So 240 million extra subscriptions.
... Are there 240 million people left on the planet who can afford $1000/month?? (Either directly, or their employer) This kind of scaling is already hitting the limits of people on the planet. That sounds ridiculous for "240 million people" against 8 billion, but remember that $1000/month is a lot of money and a lot of jobs just do not benefit from AI. 2/3rds of employment in the US is stuff that happens in the physical world. Claude won't restock shelves, manufacture goods, construct buildings, cook food, or wipe geriatric asses.
Go again with $200/month. While this monthly fee is much more palatable, the sub-count inflates to 300 million subs needing to grow to 1.5 billion. They'd need to sell a sub to everyone in Europe and North America.
(And while there's loads of people in Africa and Asia, most of those are low income. You're not getting expensive AI subscriptions out of them or their employers either. China's obviously not gonna buy US AI, India has a GDP-per-capita of $250/month.)
>They'd need to sell a sub to everyone in Europe and North America.
Yep. Every man, woman and child, and even then provided we include Russia, Mexico, Cuba, Haiti etc, and, out of desperation to get to 1.5 billion, Turkey, which is in Europe a little.
> There's plenty of noise about banks holding large amounts of bad private credit debt.
This is still only big enough to cause funny banking collapses not actual 2008 scale financial disasters. Banks hold a lot of bad debt, but it's isolated from consumer accounts. Might not want to hold equity in SoftBank though.
> There's so much uncertainty and the combination of war, high oil prices, and uncertainty about tarriffs that the market struggles to value anything as international fear drives investment into the US and high prices confusing whether growth is growth or just inflation.
The big concern lies in what the Trump admin will do. Things could end up merely a bad recession, like the Dotcom and Telecom bubble.
Or they can attempt to keep the bubble going once it collapses, crashing interest rates, and doom the US economy.
On other hand private corporate credit freezing might take down lot of business that need credit lines to operate regularly. Even the not so bad zombie companies. Tightening up and not being able to revolve credit anymore could lead to bankruptcies.
>This is still only big enough to cause funny banking collapses not actual 2008 scale financial disasters. Banks hold a lot of bad debt, but it's isolated from consumer accounts. Might not want to hold equity in SoftBank though.
Banks are lending to these private funds that are packaging questionable loans into securities (as opposed to banks giving loans or companies issuing bonds). This is the post-2008 place for people to get highly leveraged loans and they probably need to be better regulated.
But yes it doesn't seem like private credit alone will cause problems, the concern I'm trying to outline is a few of these things happening at the same time causing a kind of collapse.
TACO uncertainty is strangely propping up asset values as there's always a credible thought that whatever is happening is pretend or going to be reversed soon. And the expectation that the fed isn't independent any more and will make decisions to prolong the bubble resulting in a bigger crash ambiguously far into the future. Few want to start shorting because they have no concept of how long the market can stay irrational or if 20% inflation might be around the corner instead of a popped bubble.
> the big providers are charging full freight for inference.
Except they're not. Anthropic's claims of temporary profitability line up exactly with when SpaceX is giving them discounted compute, OpenAI's such a shitfest they threw the CFO off the glass cliff for daring to push back against the IPO. "Profitable on inference" is an unsubstantiated rumour.
Just look at the copilot changes. Demand switching to other providers immediately when prices rise, and there's not even certainty that the new copilot prices cover costs.
> They might not make back the money from training
This is an understatement. With all the datacenter buildout, they need trillions. For the investors get their money back and the bubble to not implode, they functionally need to unemploy everyone in the US.
Unemploying everyone was what openai described as their success condition when it was founded a decade ago. There was a q&a on their website that said "How will you know when you have reached AGI? When the system performs most or all economically valuable work." Lots of people thought they were joking, or it was marketing, but they were 100% serious from the first.
I think they've since changed that definition, but the reason for it is their agreement with Microsoft which stops granting MSFT IP rights to OpenAIs tech once they reach "AGI". So, it's in OpenAI's best interests to be able to claim "AGI" ASAP.
The pricing on Open router is clear. Anthropic, OpenAI, and Google all garner a massive premium over deepseek and qwen. There's no other realistic explanation except that they're making bank.
Why do you think Chinese companies can do that? It's government subsidising price they do it with literally every ibdustry.
Home grow a bunch discount them federally, let them wipe the foreign markets.
If AI is threatened by china why would US NOT do the same? If they did they're in a much stronger position to do so than china. Cheaper energy, more cash, stronger industries.
Infrastrucure is thr kind of thing that only a foolish US admin would let fall apart to their advesary.
> Home grow a bunch discount them federally, let them wipe the foreign markets.
US is doing the same and was doing that for decades now. American companies operate on loss for astonishing amounts of money and consider it completely normal. One gotta love complains about Chinese companies selling under price coming from American tech industry.
This is just silly. Deepseek has published so much regarding speeding up and making cheaper inference and people are still harping on the government subsidies thing.
So what's all the project Stargate stuff? Subsidies only work when China is doing it?
Deepseek is actively sacrificing performance for cost, which is very clear in their latest model releases. They are not attempting to get to number 1 in benchmarks, and they say it clearly in their own publications.
Furthermore, being open weight, anyone can sell qwen and deepseek compute, not just Ali and deepseek themselves.
And yet they are not profitable on an ongoing basis, and aren’t even claiming to be.
The supply is currently constrained because 50+% of data center plans were cancelled as a result of the impossibility of the buildouts happening in a timely fashion, and subscriptions are charging a small fraction of the actual cost of inference, leading them to all bleed money, hence the rush to IPO to get one last infusion, since many of the past investors have publicly stated they aren’t putting any more money in until they see an ROI.
Companies are hitting their budgeted limits for AI tokens less than half way through the year and reporting that they aren’t seeing enough benefit to substantially increase that budget, and so they are scaling back use and asking people to be prudent rather than token maxxing.
In the meantime subscriptions still exist in the form of chatbots and it’s easy to exceed the inference cost of the provider by simply using your daily, weekly, and monthly limits.
The reality is that we just don’t seem to be at a point now where people are willing to pay full price for the perceived value. Perhaps we’ll get there within another generation or two of hardware and software improvements.
Can 5% of the population even pay for that? Some kind of huge increase in prices for compute and inference and companies maintaining large bills for AI assistants for key employees or teams (1000-2000$) seems most likely to me.
75$ a month for a cell phone? I pay 18$ monthly for mine. This is Southern Europe where the average monthly wage is 1000$. I dont know who could afford yet another 75$ expense.
Anthropic is a five year old startup, if they can be profitable that quickly in the AI space, even if only temporarily, I'm not really seeing the problem?
These companies are going all in and growing rapidly, because they want to dominate the market and since it is difficult to differentiate between competitors, even being third place is a terrible place to be in the consumer facing AI space.
An open mind to not having a knee jerk reaction to “this is vibe coded therefore it’s not production-ready.” The delta between the prototype and production is likely far lower than you think, and many engineers are sneering at the prototype without actually looking at it or attempting to spend a few weeks applying engineering discipline to bring it up to scratch.
What is being sneered at is these prototypes being put into production. What is being demanded is that additional engineering time to make sure it's actually up to scratch.
> That first pillar is still there. Maybe the author isn't aware of the impact they have, but I know, with the evidence of reverted PRs, that when I step outside my area of deep knowledge I can no longer call BS on the agents. Our most capable agent, with access to the same kind of distributed systems the author talks about, is regularly wrong, frequently myopic, and just outright dumb constantly. It's the expertise of engineers on the team that push it back on track.
I'd posit there's another layer. You have domain knowledge, certainly. But more valuable still is the wisdom to find more.
Anthropic and OpenAI can stick financial regulations in the training data all they want, but the AI systems will never learn to anticipate the future, or reach out to clients, partners, or regulators in complicated situations.
> Because it wasn't about "engineers should write 1M LOC per month of product code" it was "we want to scale automated porting of code to safe languages so that 1 engineer managing 1M LOC of automated conversion can work"
These are one and the same. Whether it's ported code or not doesn't change that. The framing device also doesn't matter, because it's the exact "Oh it's our goal" shtick that executives use in the former's case.
"It's just a measure" doesn't cut it in a world where every single AI measure immediately gets turned into a target by executives greedy for efficiencies that don't exist.
EDIT:
Right, I forgot. This is HN where everyone is a galaxybrain and "Port a million lines of code per month" is a totally reasonable goal for a single individual.
reply