What you're describing is people who fail to follow the most basic principles of academic research. (Check existing academic literature, mention and give credit to prior work.) This would be fine if these people didn't claim to be doing scientific research, didn't boast their academic credentials, didn't publish their finding as original work and didn't demand credit for their work in academia. Of course, they do all of these things. They benefit from a system they're actively denigrating (and in some ways degrading).
To put it more simply, people with academic credentials should not demand acknowledgement of their current intellectual work while denigrating and ridiculing the importance of very similar work done in the past.
It's an interesting project, but the discussion on HN looks weird. It gets brought up every few weeks[1] and everyone just spams comments with messages about how "fast" it is.
DuckDB is fast for some specific workloads. If you use it for most other things, it is at least an order of magnitude slower than SQLite. It also has some limitations in terms of what SQL it will currently run (e.g. I immediately ran into an issue with recursive queries). That will probably get better with time.
[1] If you search HN for "sqlite" and "duckdb" you get 4,310 hits and 2,398 hits respectively. That's a very heavy skew, considering SQLite is everywhere and had been around for a quarter century, while DuckDB effectively appeared on the scene two years ago.
I'm going to sound like a broken record but... different use cases. They're analogous in the comparison "sqlite for analytics" but completely different architectures and implementations. Part of this is the fault of the developers, but I feel they were trying to highlight the similar focus on in-process, zero dependencies, simplicity and test coverage - not a direct "vs" comparison. IME recursive queries in analytical workflows are not very common; they typically work against the fundamental data layout on disk.
SQLite is awesome and I would love to see more posts about it, but the reality is one of the major reasons it's awesome is the no-drama/stability/it just works. DuckDB is seeing a lot of development on many fronts so there's a lot more to learn and talk about right now.
Yes, it's specifically promoted as DBMS for OLAP workload. And it's usually compared to ClickHouse, another analytical DBMS. So people who use it know why it's good.
The article has an explanation for what kind of database it is. After reading that one sentence you wouldn't write the second paragraph.
> it's optimized for the kind of queries that scan millions of rows to filter, aggregate, and join — not the kind that look up a single record by primary key
I’m sure the use of duckdb may seem weird for normal developers, but for data people it really is game-changing, especially for data scientists or business analysts.
The fact that people here are looking at these numbers and saying "this is fine" is absolutely bonkers.
Basically, it's a company that's not sustainable for two separate reasons. The first one is that they have an extremely high overhead. SG&A of 55% is really bad. The seconds reason is that their R&D costs are truly astronomical. They could probably cut those costs to some extent, but they're not going to cut them to nothing. They're already losing ground to Anthropic even with this much R&D.
To put it differently, even if OpenAI cut its R&D and inference costs by half, they would still be leaking money like a sieve.
This is the venture model now though. Spend until profitable. Uber did it. It seems OpenAI could do it as well given we seem to be in a 2 horse race for foundation models and having capital to get better pushes them further ahead.
Uber’s situation was different, though. The reason Uber were bleeding money is because they purposefully made all their rides cheap to undercut the taxi businesses. People used Uber because it was cheaper than renting a taxi.
Now you can’t really find taxis anywhere, even at airports it’s a lot more difficult than it used to be.
Once the taxi business was disrupted enough, Uber’s pricing skyrocketed and customers had basically no other options for competition on pricing.
OpenAI basically created a new market. There is no AI chatbot incumbent to disrupt and swallow.
Uber/Lyft takeover had little to do with price (though, yes, they were cheaper) and everything to do with reliability and overall quality of service. Even though ride sharing industry lost money in subsidy arms race and side bets it was fundamentally sound in major metros since early on (similar to how Amazon was fundamentally sound from early on, despite not recognizing profit for a long time). Popular "analyses" kept equating Uber/Lyft with firms losing money on every sale with no path to fix it but the demand was always there as riders had already left taxis and transit on reliability and convenience grounds.
Some humans will need to interpret the thinking and apply it somewhere and take some responsibility for those decisions. If you think AI can do all that end to end it’s a different question but we’re nowhere near that right now.
Definitely, I’m not saying that AI can entirely replace humans. But AI is definitely replacing parts of many jobs. If AI companies raise their rates to be profitable, and it turns out that paying for profitable AI is not worth it vs paying for humans, that might be a sticky situation.
There will always be a competitor that can undercut the inference market. There is no "moat" given that you can self host decently capable LLM agents like Qwen3.6 on not super expensive hardware, like an AMD R9700, and still get competitive speeds to most cloud interfaces.
If you can self host it that easily, any Joe can scale it out much like shared web hosting, and shared web hosting or even dedicated rented boxes has always been cheaper than the big cloud providers.
I don't think OpenAI or Anthropic can reasonable compete in the long term if they can't achieve "AGI", and they won't, no matter what shareholders desire.
Actually the point is total cost wise outside of subsidy it is not cheaper than humans. the bigger problem is as the parent said open AI created a market. It is selling a commodity service with investor funds. There is no moat. your second sentence soon you won't be able to find human thinkers is on its face absurd, assuming the human race continues. Thinking is the human ecological niche.
For now, businesses are getting addicted to cheap tokens. As the screws get turned, business will debate whether they should spend budget on humans or tokens. What's further devastating is that humans are also becoming addicted to cheap tokens. Much human output is nowadays a token slopfest. People are becoming dumber too. So the real business question will be spending budget on token monkeys or tokens.
Which doesn't work the same way at all. With taxis, making them unprofitable leads to a long-lasting lack of taxis. When lots of jobs are lost, it actually becomes easier to hire someone with the right experience.
It might work very much the same. Discourage a cohort of CS grads into following another career path. Give businesses enough time to fully commit to “agentic workflows” such that they don’t have the expertise for in-house engineering anymore. Completely spaghettify every code base such that only AI would be willing and able to implement new features in it. Let customers lower their expectations of quality to meet what AI can product. By the time they crank up the token price, it may be hard or impossible for businesses just to switch back to human engineers.
It depends on how long you can keep those people un- or underemployed. I think engineers are rapidly bleeding experience even while being employed if all they do is prompting.
But when lots of jobs are lost, consumer spending is lost, and it becomes harder to sustain a business (whether B2C or B2B) and afford to hire someone...
If you knew about how much man power it takes to maintain, evaluate and improve agentic workflows, I don’t think you would write such a thing. In this context, AI is a jobs program for permanent employment.
Japan too. never thought I'd see it here but a taxi driver took the long way after a work drinking party. I guess he thought we were too drunk to notice. Well my boss sure did and lost his mind at the guy.
Likely the continued existence of taxis are keeping Uber's prices in check in the Australian market.
Uber will be running an optimisation model and be charging the maximum market can sustain, with additional goals such as eliminating competition and not being shut down by regulators.
My family and I have gone back to using car services for rides to the airport b/c "Uber XL" seems to include a WIDE variety of vehicles in terms of size and cleanliness.
A car service is about the same cost, the car looks brand new and clean and the driver is helpful.
Uber's situation is exactly the same. OpenAI is offering inference for a bunch of industries at prices that make it more competitive than hiring humans to do the same work.
If the break-even price to actually provide the service wasn't actually economic compared to humans, would there be nearly as much of a market? That's the real question. OpenAI is basically betting that they can live long enough that AI systems get built around them, which creates enough of a lock-in that they still have customers when prices increase by a lot.
I think you underestimate the price by a few orders of magnitude where it makes sense to pay a model instead of a human. If someone earning 200,000 a year gets replaced by paying 500 a day to Anthropic or OpenAI their employer comes out ahead.
There's likely always going to be value in limiting the number of $200k+ SWEs you have to pay. But that's not the interesting case.
What about the $10k/year offshored employees that are getting replaced by AI call centers? If that were the break even, then once you close down the whole building and develop the systems to not need them, then how much would inference costs have to go up before all that gets unwound and handed back to humans? It's more than you think - there's real margin there.
The uber situation was even more insidious than that. It wasn't like college students were calling cabs to go to bars in 2013. Uber created a market. It was essentially a mind virus. Gee now I can go to this place all for $7. Chum the water, establish the new pattern of living that people won't ever back away from, then twist the knife and raise prices knowing they won't revert back to whatever Old Way now long forgotten or not even engaged with by the upcoming generation.
Before Uber did it, Amazon had been doing it for almost two decades. It's nothing new. There is a difference between 1 billion and 20 billion in losses, though. Amazon in, I forget, 2014? Ran a profitable quarter with I think $1 in profits, simply to prove they were in control of their finances, and "we can stop any time we want". Sam gets a lot of shade, but he's been around the YC block once or twice, I suspect whatever risk they're taking on is at least somewhat measured.
Amazon structured their entire operation to look like this but as you indicated, could have switched to a porfit-making, dividen-paying company more than a decade ago, that just wasn't their strategy. The same can not be said for OpenAI. Even if they slashed their R&D, their marketing and sales costs are extremely high for a tech company. On paper they look more like a utility and those are not worth double-digit multiples; they compete with t-bills and GICs
Looking at the fact that third parties are making a profit offering XYZ third party open models on OpenRouter, it stands to reason that OpenAI could turn off their R&D, marketing, hype jedi, legal departments and just sell GPT9.999 and turn a profit.
Again like in the Amazon analogy, I don't think they're done growing, and unfortunately, I think they've positioned themselves (perhaps intentionally) as too big to fail, and need to continue growth at all costs.
I'm glad I'm not OAI's CFO sounds like a stressful job trying to justify/account for whatever Sam says to the board, or whatever the board demands. Sam hasn't said hardly anything since about February so I'm guessing the CFO simply bends to the will of the board these days. But that's speculation.
it stands to reason that OpenAI could turn off their R&D, marketing, hype jedi, legal departments and just sell GPT9.999 and turn a profit.
That rests on 2 assumptions:
1) That inference on OpenAI's frontier models is actually cost competitive with open models. Their high SG&A suggests otherwise.
2) That slashing R&D won't lead to a marketshare collapse when everyone (remaining) moves to Anthropic to get on their frontier models. All evidence suggests otherwise again, with Anthropic already exerting enormous competitive pressure on OpenAI's marketshare.
I think OpenAI is in a terribly tenuous position: they're getting squeezed from Anthropic (on the high end) and open models on the low end. A lot of companies in a lot of industries suffered this fate. Getting stuck in the middle is not a good thing!
It wouldn't surprise me if they have an unadvertised model router. Like, it's extremely clear you're chatting with a lobotomized model if you use voice mode in their app. Wild speculation here but I'm reasonably confident that as a $20/mo user I am not getting the same level of max thinking model that my enterprise account gets for the same question, despite both being labeled as the same model. Nowhere in their literature does it say $20/mo users get the exact same model/thinking effort, either
I’m not sure Sam is actually well-regarded around the YC block? Didn’t he lie about being the chairman of the board of YC? That’s what it says on his Wikipedia page.
Let’s not forget the whole fiasco where he was already almost fired from OpenAI.
His business track record is basically one failed location-based social media company.
I think it’s starting to become clear that OpenAI is going to be the first casualty of the AI race, and I think these undisciplined operations are a big part of the reason.
A major tell is how Apple Intelligence is seemingly steering away from OpenAI and is embracing Google instead.
Anthropic has the most useful B2B tooling and found their product niche, and they have the model leads in that niche.
xAI gets financially shielded by being a part of a gigantic financial instrument and the Elon Musk reality distortion field. Cursor has a similar product market fit as Anthropic and gets to consolidate with xAI.
Google and Microsoft get to use AI within their highly profitable ecosystems.
Apple gets to mostly sit out but act as one of the biggest toll booths for everyone else.
You've seen Sam Altman's interviews, yet you still think him a competent man? I think he's rather the embodiment of the death of meritocracy as an idea.
The problem here is that open weight models are already good enough for a majority of process automation and intelligence tasks and that is where a good chunk of efficiency corporate dollars are. So there's an ever shrinking slice of inference that will hit the frontier models and inference is where the insane margins are. Now to be fair, Claude co-work and Claude code/Codex do seem magical today and these potentially will continue to be high margin/leverage plays. Frontier models are also likely to push towards decision making - so we'll have to see how it shakes out but the bottom end is already commoditized and it is getting bigger and bigger.
Or Claude is better at getting people to move to more expensive membership tiers. From reading here, it seems like Claude still has a lot of users. If Claude has lower limits for their $20 plan, it stands to reason that people are paying for more expensive plans to get similar levels of usage. This assumes they aren’t reducing demand through the throttling, which is a big assumption.
I’d love to know what Anthropic’s comparable numbers look like.
In other words, Codex likely has a lower demand:capacity ratio compared to Claude. Which can mean either OpenAI did a better job at building out capacity, or the demand for Claude outstripped Anthropic's projections. Or both.
huh, I felt like (gpt5.5 < opus4.8 << fable) in terms of code quality, and (g ~= o < f) in terms of pure pass rate; which one did you mean? curious about your typical workflow/tasks
I want to like gpt5.5 but it's like an evil genie: bugs are fixed, features are implemented, you are now a proud owner of a 2kloc file with a single function that makes you wish you had keybindings for horizontal scrolling
GPT 5.5 regularly got things correct where opus 4.8 got things wrong. Just today, it was launching an executable and putting the environments after the launch. It should have been
Env1=val env2=val ./executable ....
But instead it did
./executable
Then set the envvars. Chatgpt 5.5 made no such mistake.
These companies are clearly calling things that are R&D that aren't R&D.
If you're building a model that lasts a few months before it's no longer the most current one, and maybe a year before it's completely unusable by anybody, then that should just be COGS.
Doing that, however, would betray the real problem with this business model.
They are also likely overestimating the useful lifespan of the hardware. They keep extending the number of years on the GPUs to make the accounting look better.
Presumably when the power consumption costs more than the cost of replacement.
It’s not so much that these GPUs stop working after 3 years, but that newer GPUs can handle more requests with less power for the same purchase price. So the useful value of the GPU degrades until eventually it’s cheaper to replace than to keep running.
If the supply side constraints remains the same, I doubt they'll be releasing their GPUs as they could be considered strategic assets. Their current moat is largely hardware right.
In few years open weight models may be good enough for anything but advanced usecases. With right hardware, competition may grab the lower end of market using open models. There's also potential loss of interesting training data from real conversations.
Not here to entirely disagree its bonkers, but Tesla lost about 6b until 2022 when it got profitable and has since returned a healthy multiple of its prior losses as profits.
Amazon and Uber are other examples of businesses that looked like total basket cases for years. I remember reading, and at the time being persuaded by, articles arguing that Uber was doomed because there are no real economies of scale in livery services, and so the minute they began to achieve a dominant position and hike prices, countless competitors would easily swoop in and undercut them. Didn't turn out that way.
Yeah insane that people think it'll be okay in the long run but wondering how much different the financial status of other such company would be? Not much I guess.
I can't imagine there's a large variation in costs per token for inference and training costs between companies, and since they're all basically doing exactly the same thing, and competing on price... yeah.
I'd agree for OpenAI and Anthropic, but Google is a different story, because they're renting TPUs to both OpenAI and Anthropic, and presumably making reasonable margins on them given how supply constrained the entire industry is.
>"It’s easy to forget, but for most of 2025, the idea that AI-generated code was slop and might always be slop was not only a reasonable position to hold, it was the default, mainstream position.
That question was answered decisively last November."
It's easy to forget that people said this exact thing about every model after GPT 3.5. This is a standard trick the industry uses to invalidate negative experience with LLMs. 'You are prompting it wrong' becomes 'you are using Gemini, but you should use Clade' which then becomes 'well, all of your criticism is now irrelevant, because everything is fixed in this new version'.
This "discussion" about capabilities is set up to be asymmetrical and basically non-falsifiable.
You undermine your own point when you misrepresent the situation like this. Real human mathematicians, including at least one Fields Medal winner, have validated and complimented the result.
You interpretation of my comment is either very biased or just made in bad faith.
The claim made by Open AI has two necessary components. The first one is that the conjecture had been disproven. (This is what had been verified by "real human mathematicians".) The second necessary part of their claim is that the work to disprove the conjecture was done mostly by their AI model rather than by people employed by Open AI.
Funny thing is that even the explainer on OpenAI's own website points out the issue:
"This result does not show us all the times AI has claimed to have a proof of something and been wrong."
"I believe if the level and type of human expertise that is represented on this note had been assembled to find a counterexample to this conjecture a month ago, and those people put in similar amounts of time working on it than they did to reading and thinking about Chat GPT’s solution, the mathematicians would have found a counterexample."
You seem to be saying model capabilities aren't improving. They are. The fact that many mathematicians have looked at the result and confirmed it and solved some other problems with the technique elevates this above claims.
i mean i am very much still waiting for it to not be slop, but fable actually i think made a bit of headway in this direction, the code it writes what little of it i saw, makes me want to fall over dead slightly less than other models.
This "how to adapt" prop-slop is getting tiresome. So much of it is obviously intended to project and demoralize rather than provoke thought or give legitimate advice. The trick of this kind of writing is to skip arguing that something rather questionable will happen and go straight to giving advice about how everyone should adapt to the new, totally inevitable reality. This isn't even a particularly sophisticated method of manipulation.
AI seems to be ruining every single major thing that drove economic growth for the past 4 decades. PCs, the Web, software in general, high-capacity servers, Raspberry Pis and so on. The next thing to be affected will probably be smartphones. All of these things are foundations of profitable businesses right now and we are destroying them on the mere promise to get to some idiotic utopia in the future.
AI is a nothingburger, it's just the latest capital flight. Undeployed capital needs to go somewhere and this time it's AI, last time cryptocurrency, before that metaverse. One might reasonably ask why these bubbles keep getting bigger and bigger and the answer to that is almost entirely USA government policy which is to squeeze labor and expand capital at every opportunity.
The premise of the article is faulty. "Nerds" became "cool" as technology became more important, so sociopaths in leadership positions stated to pretend to be nerds. It's as simple as that.
I think a far more important question is why we no longer have more reasonable public figures. Who are the modern equivalents of Isaac Asimov or Richard Feynman?
What this shows me (again) is that the whole system where vulnerabilities need to be constantly discovered, reported, analyzed, then patched, then the new version distributed to every singe user - again and again - is quite obviously unsustainable. The industry must come up with some alternative system for dealing with bugs and security issues. Currently the industry prefers to play dumb and turn its own failures into a profit (rent seeking) opportunity.
IMO Writing correct software the first time around - so formal methods.
But the tooling isn't there yet (though lightweight versions, e.g. strong type systems like rust's, are and significantly reduce the security issue load).
First, it's not made up. I've heard numerous stories of this sort (maybe a bit less extreme) from people I have known for many years working for large and mid-sized companies. Second, large companies have entire teams for monitoring social media and they absolutely will try to zero in on you and fire you if you speak too much about their crappy internal processes. This goes double for anything in S&P 500.
To put it more simply, people with academic credentials should not demand acknowledgement of their current intellectual work while denigrating and ridiculing the importance of very similar work done in the past.
reply