Crazy how five years ago this level of AI would be seen as scifi, and now there are people out there who think it's trash because we can trick it if we ask questions in weird ways.
> there are people out there who think it's trash because we can trick it if we ask questions in weird ways.
Some of this sentiment comes form wanting AI to be predictable and for me stumbling into questions that the current models interpret oddly is not uncommon. There are a bunch of rules of thumbs that can be used to help when you run into a cases like this but no guarantee that they will work, or that the problem will remain solved after a model update, or across models.
There are a lot of rules of thumb you can follow to avoid getting bitten by a rattlesnake, but the easiest way is to just not pick up random snakes. I don't know where I'm going with this, but I am going for a walk.
> Anthropic CEO has stated they have high margins on inference, so training is the big cost center.
I'm pretty sure that in corpo-speak "inference" excludes the cost of datacenter construction, GPUs and other hardware, manual data cleaning, R&D, administration, etc - basically everything except the power bill for inference.
I have absolutely no problem with companies that run inference only - plenty of them offer open models as a service - they're usefull and their accounting can be believed... but they don't have near $ Trillion valuations and they don't misallocate capital on a vast scale as the frontier models do.
The point of the OP is that closed models don't pay for themselves and, on the scale of the US economy, they provide minuscule economic advantages compared to the enormous investments they consume.
They've raise 70-ish billion (which they have not spent all of) and have a run rate of 14 billion/y as of now. All said and done those are great economics so far, even accounting for those extra expenses.
Your argument requires the run rate to reduce over time until OpenAI reaches profitability. However, even OpenAI has publicized that they expect their expenses to exponentially increase for their models to remain competitive.
So they are not profitable now & they have no idea of when they ever will be.
Worse, Gemini has guaranteed funding for continued training whenever the AI hype bubble pops.
Anthropic & OpenAI's only saving grace is that Google is generally terrible at product.
> Is it actually being sold at a steep discount? Anthropic CEO has stated they have high margins on inference, so training is the big cost center.
They're spending more than they're making. For the foreseeable future, saying "we could be profitable if we stopped training" if goofy, because they can't stop. If they do, no one will want to use their product because it will be overtaken by competitors within three months.
I get it that in 10 years all of this might peak and we're gonna be content using old models, but that'll be a very different landscape and Anthropic might not be a part of it anymore if they don't start making money before that.
> I get it that in 10 years all of this might peak and we're gonna be content using old models
I would personally be happy using gpt 5.3 codex for the foreseeable future, with just improvements in harnesses
IMO we're already at the point where even if these company collapse and the models end up being sold at the cost of inference (no new training), we would be massively ahead
That's a perfectly valid approach if you can balance capex and revenue. Why stop and try to be profitable when the economy is giving you the liquidity to push that down the road?
Models are already super useful, but if you can make them more useful by burning cash people are willing to hand you, why not?
Well, training isn't going to end soon if these companies keep on competing with one another whilst being neck-and-neck, so I'm not sure why you would ignore the cost of training in the ROI calculation.
Amodei says yes - each model pays for its training. But they're scaling up investment for each new run, so they're still happily in the red.
And also that may be the case for Anthropic who have fewer free users, a large enterprise business, and less generous rate limits on their subscriptions. I don't know if OpenAI or Google have commented. I suspect OpenAI is in a worse position given their massive non-paying consumer base.
They have good margins on inference at API costs, i.e. $5/$25 per mtok input/output. They are almost certainly making losses on subscriptions, at least if people max out rate limits.
In the past 30 days I have burned $78.19 in API token costs with my $20/month Claude Pro subscription. In January I burnt over $300 in API token costs.
Because the power users of the max plan are subsidized at the upper end of usage by people who don’t approach the per account limit. In other words, the power users are getting more than they pay for, because most people don’t reach that threshold. If you let the power users have dozens of accounts, it has a multiple effect on the proportion of accounts breaching the profitability line.
They are likely aiming to maximize reach/mindshare. Get as many people hooked as possible. More important than minor upside from a few multi-Max users.
EDIT: also, the casual or gym-style members that pay every month but barely use the service are of course very valuable wrt margins
We were pretty close to a collapse of the existing financial system. Maybe we’d be better off now if it happened, but the interim devastation would have been costly.
We weren't that far away from ATMs refusing to hand out cash, banks limiting withdrawals from accounts (if your bank hadn't already gone under), and a subsequent complete collapse of the financial system. The only thing that saved us from that was an extraordinary intervention by governments, something I am not sure they would be capable of doing today.
Exactly. Most people just don't know how much data is being collected on them, and probably can't know at this point. I say can't because the reality sounds so much like a conspiracy theory that a majority of people would simply reject the truth outright.
> In a way it shows how poorly we have done over the years in general as programmers in making solved problems easily accessible instead of constantly reinventing the wheel.
I just don't think there was a great way to make solved problems accessible before LLMs. I mean, these things were on github already, and still got reimplemented over and over again.
Even high traffic libraries that solve some super common problem often have rough edges, or do something that breaks it for your specific use case. So even when the code is accessible, it doesn't always get used as much as it could.
With LLMs, you can find it, learn it, and tailor it to your needs with one tool.
> I just don't think there was a great way to make solved problems accessible before LLMs. I mean, these things were on github already, and still got reimplemented over and over again.
I'm not sure people wrote emulators, of all things, because they were trying to solve a problem in the commercial sense, or that they weren't aware of existing github projects and couldn't remember to search for them.
It seems much more a labour of love kind of thing to work on. For something that holds that kind of appeal to you, you don't always want to take the shortcut. It's like solving a puzzle game by reading all the hints on the internet; you got through it but also ruined it for yourself.
> I just don't think there was a great way to make solved problems accessible before LLMs. I mean, these things were on github already, and still got reimplemented over and over again.
What kranner said. There was never an accessibility problem for emulators. The reason there are a lot of emulators on github is that a lot of people wanted to write an emulator, not that a lot of people wanted to run an emulator and just couldn't find it.
"I mean, these things were on github already, and still got reimplemented over and over again."
And now people seem to automate reimplementations by paying some corporation for shoving previous reimplementations into a weird database.
As both a professional and hobbyist I've taken a lot from public git repos. If there are no relevant examples in the project I'm in I'll sniff out some public ones and crib what I need from those, usually not by copying but rather 'transpiling' because it is likely I'll be looking at Python or Golang or whatever and that's not what I've been payed to use. Typically there are also adaptations to the current environment that are needed, like particular patterns in naming, use of local libraries or modules and so on.
I don't really feel that it has made it hard for me to do because I've used a variety of tools to achieve it rather than some SaaS chat shell automation.
And can come with hidden gotchas. I remember dealing with one bit, presented as an object but I thought that was simply because it was in an object oriented language, it was simply a calculation with no state. Many headaches later I figured out it had some local state while doing a calculation, causing the occasional glitch when triggered from another thread. They didn't claim thread safety, but there sure was no reason for it not to be thread safe.
Ah yes people were making emulators because emulators weren't a solved problem...
That isn't why people made emulators. It is because it is an easy to solve problem that is tricky to get right and provides as much testable space as you are willing to spend on working on it.
There has always been a tension between "take the time to build something you know will work" and "prioritize speed over all else and hope you get lucky and it doesn't fall over too fast" in software. AI is making the difference in speed between the two schools of thought larger and larger, and it's almost certain to make the latter philosophy more financially attractive.
Yeah, I can see that. But that didn't really feel like what the original article was arguing. It felt more that it was arguing that even people who care about good code, if they use agentic tools at all, can't produce good code, and it was the advantage in the velocity of agentic tools as a whole over the production of good code as a strictly separate category that was the problem?
It can be, but there are lots of reasons to believe it will not be. Knowledge work was the ladder between lower and upper classes. If that goes away, it doesn't really matter if electricians make 50% more.
Its not really a matter of some great shift. Millennials are the most educated generation by a wide margin, yet their wealth by middle age is trailing prior generations. The ladder is being pulled up inch by inch and I don't see AI doing anything other than speeding up that process at the moment.
> So if this time is different, I’m not seeing it yet
Compare those numbers to 5 years ago. Remember, this is the timescale of a country, not a beagle. America has run on the strength of the dollar for decades, and the symptoms of that collapsing are likely to play out over decades.
> It's mostly in forms of treasuries purchased in USD that pay in USD - this means the indebtedness creates a huge amount of dollars abroad that foreigners have to then spend on US services, driving demand.
Strangely enough, this is exactly the opposite of how it works. The dollars abroad tend to stay abroad, as either a more stable alternative to local currencies, or a reserve currency. Likewise, treasuries held abroad tend to stay there as reserves. This is how the US is able to run both a huge debt, and a huge trade deficit. If the dollars were being repatriated, the trade deficit would close, and the influx of money would cause hotter inflation. Same with treasuries, yields would spike as demand fell.
There are lots of second order effects there, good and bad, but, basically, those dollars not coming home has funded America for quite some time.
I, a foreign entity, have sold something to an american and now have 10 dollars and zero treasuries.
I purchase a treasury. I have zero product, zero dollars and one treasury.
At some point in future i have zero product, maybe 12 dollars and zero treasuries. Presumably i now either repeat the cycle or use my winnings to spend on us output.
GP’s version checks out, your assertion about dollars staying abroad doesnt track? What am i misunderstanding - How did these dollars get abroad, how did they repatriate to buy treasuries, how did a treasury become a reserve, how did the dollars still exist abroad after being exchanged for treasuries?
> I, a foreign entity, have sold something to an american and now have 10 dollars and zero treasuries.
Or you sold something to a non-American entity in a dollar-based market, eg. oil. The dollars do come from America to begin with, but once they get "out there" they work as a medium of exchange for whoever wants to use them for that purpose.
Interesting - The petroyuan was not on my radar at all.
https://ipr.blogs.ie.edu/2025/06/27/geopolitics-of-oil-how-c...: This article explores case studies such as Russia, Iran, and Venezuela, illustrating how the petroyuan has been implemented to bypass sanctions and reduce dependence on US financial systems.
> Presumably i now either repeat the cycle or use my winnings to spend on us output.
What you are missing is that these dollars can be circulated indefinitely in the global economy without ever repatriating, because they are valuable and useful as actual currency. They may never come back to the US.
Because the world trades using US dollars. Country A needs to buy something from Country B. Country A needs to buy/get dollars to buy stuff from Country B. Country B will not accept anything but dollars or gold for its products because it also needs to buy other stuff like oil in dollars from other countries.
It could accept any credible currency if it was connected enough. Euros, yuan, rupees and yen aren't going anywhere for at least 20 years. Each one is a separate system and countries mostly connect to just one, which is USD, but that doesn't have to be the case forever.
India won't accept euros because it's not part of the ECB, not because it doesn't believe in their value. But India has accounts at US banks in dollars.
Banks do this, not countries. Most banks in the world have accounts at US banks to accept dollars with, they don't have accounts at eurozone banks to accept euros with, or Japanese banks to accept yen with. It doesn't matter in everyday practice because it's easy to exchange euros in eurozone banks or yen in yenzone banks with dollars in dollarzone banks. There's plenty of infrastructure for that. It matters in long–term economic trajectories because all those banks are holding US dollars and the US exports inflation to them and they're not holding euros and then ECB can't.
Most of the time this is exactly (foreign or not) institutions do.
Think about it, if the 10-dollar treasury is due and you got your money back, the US debt will go down by 10 dollars. However, in our reality, the total amount of US debt almost never goes down.
Of course some interest will be used in other ways, like spending on the US goods or staying as cash to provide liquidity. But at the end of the day, the most popular way to spend the money got from due treasuries is... to buy more treasuries.
> Strangely enough, this is exactly the opposite of how it works. The dollars abroad tend to stay abroad, as either a more stable alternative to local currencies, or a reserve currency.
I think you have it backwards. The US dollar can be handled as a more stable alternative only if it's actually stable. As soon as the US government starts to deteriorate goodwill and outright be hostile towards the world, not only does the US dolar lose its value as a stable alternative but foreign governments start to be motivated to dump their assets, which further tanks its value. In the past couple of weeks we started seeing countries outright dump their investments in US dolar to derisk their portfolio, and they did it at the tune of billions of dollars. You also started to see political pressure for foreign governments to demand their gold reserves are pulled out of the US, which means this pressure goes way beyond US dollars.
If dollars were being repatriated, but as investment into financial instruments and real estate instead of purchases of goods and services, then that would not affect the trade deficit, right?
reply