66 comments

[ 0.26 ms ] story [ 11.4 ms ] thread
I'd imagine that OpenAI would try to stagger their announcements, rather than publishing them recently close to each other. Is this because their previous post (about navier stokes problem) was met with controversy?
They are desperately rushing to IPO before the bubble bursts.
AI just solved a millennium prize problem. In a matter of days. Because of a rumor that someone else solved the same problem with AI.

What exactly would AI have to do in order to not be called a bubble?

You don’t understand what a bubble is. How good the technology is is irrelevant. That has nothing to do with an economical bubble. It’s all about massive capital misallocation driven by a frenzy of FOMO, which is specifically the case for AI investments. Economically speaking what is happening is the most obvious bubble possible, it follows everything that would be expected from a bubble where companies are chasing an ill-defined grandiose dream, based on a new technology we don’t understand and has very dubious ROI, selling some vague future utopia, allocating massive amount of capital to build infrastructure dedicated to a very early versions of that technology.

As things mature there will be a correction, ie the bubble will pop.

I would recommend to read « Boom and Bust: a global history of financial bubbles » https://pure.qub.ac.uk/en/publications/boom-and-bust-a-globa...

You should see all the misallocation that was put towards valve-based computing. Why didn't they just all arrive at the correct answer without investing in discovery first?
As long as the data centers are utilized and generating revenue, I see no reason for a correction or any misallocation of capital for the infrastructure buildout.

And today these data centers are fully utilized. OpenAI tweeted today that they may need to disable new signups for the Pro subscription in the near future due to capacity constraints.

A year from now, who knows what the situation is going to be like. It seems quite possible that robotics, self driving, research, etc. drive even more demand and revenue.

Stating with any certainty that allocating capital to build infrastructure is a mistake and that there is a correction coming seems unserious.

> A year from now, who knows what the situation is going to be like.

That cuts both ways, we are building datacenters for an immature technology that is quickly evolving. We have no idea what AI will look like in the next 5-10y. Everything that is planned to be built is based on the demand we see right now, not what it will be in the future. That means different GPUs that require different cooling systems, different power supplies, etc. NVIDIA already broke backward compatibility with their new cards, which requires a different infrastructure.

What is unserious is the opposite position: believing that we already know what will be valuable in the future and bet the entire economy on it, without any proof of positive ROI.

Directionally we are seeing demand for more compute.

It's a reasonable assumption that data centers that are set up for large power usage and cooling will be valuable.

Claiming the opposite based on, well, nothing at all, in order to forecast a correction, seems less reasonable.

"Compute" is not one generic commodity. Filling your datacentres with ASICs that do nothing but compute SHA256 for Bitcoin mining was a smart move in 2015 but today that hardware is worthless e-waste.

What does the depreciation curve look like for nvidia cards purchased today? How long will it take to recoup the investment on this buildout? Will those datacenters pay for themselves before they're scrapped?

It's an interesting question. Some napkin math:

Let's say we serve a Fable class model on 8x B300.

From Kimi K3 metrics, with 8x concurrent streams, we would achieve 55-60 tok/s per stream, matching Fable 5.1 throughput.

432 tok/s x 3600 => 1.555M output tokens/h x 50$/M API price = $77.76 revenue per hour.

Assuming total API billing at 2.06x output token bill = $160.2 / hour or ~$20 per B300.

A server with 8x B300 could be $461.5k.

At an obviously unrealistic 100% utilization we would look at 4 months of revenue to match the cost of the server.

About how model serving works at scale and actual utilization I know little.

And for all we know Anthropic could serve their model with 64 streams on the same hardware instead of the 8 we assumed here.

How these things are even connected?

The current prices the largest players set for their models are not profitable, they bleed money. Eventually they will "fix" it. It could end up making their services less affordable and it could cascade other businesses and services that are dependent on them go out of business

> The current prices the largest players set for their models are not profitable, they bleed money.

How are the open-weight Chinese models staying ~6-12 months behind on widely distributed / commodified hardware, and serving for even lower prices?

By distilling the US SOTA models, which is cheaper than creating from scratch.
Its resource allocation problem, same happened with .com bubble, lots investors put tons of money into dark fibers. Were they useless? No its very useful.

If you want check a example company from the .com days check cisco, their stock peaked at 75 then crashed hard and only managed hit that again thanks for the AI bubble.

If non AI companies, in particular non tech companies start making unprecedented amounts of profit, inflation adjusted I will concede.

That being said I think LLMs are impressive, still.

Per Anthropic's prospectus: generate $30T (~94% of 2026 US nominal GDP) in revenue.
You really believe there will be less demand for AI in the future?
No, I think AI models will be cheap commodities available from hundreds of providers for a few dollars, like a Linux VM is today.
This is just a blogpost, rather than a major announcement (FLT proof was closer to the latter than the former)
At the rate this field is accelerating maybe this IS staggering
It will be fun if the AI realize that quantum is not needed and just simulate the results in a classical computer
It will be really fun if it turns out that AI simulating quantum effects in a classical computer genuinely is quantum and only collapses to classical with human observation.

(yes, I know why it doesn’t work that way, but it would be fun if it did)

I had qubit bring up and calibration fully automated with Python in 2011 including full spectrum measurements, lifetime characterization, Rabi/Ramsey measurements, calibration of single qubit gates and two qubit swap gates and full quantum process tomography, so not sure if AI is really needed there, curve fitting and some data logging is enough for this. Fluff piece. Still of course cool, I guess today I would just let Codex loose on some experiment goals but in the end my ability to produce results was mostly limited by the chip itself and the qubit lifetimes and theres no magic trick AI can apply to make these go up by a factor of 10. Still would’ve saved me a lot of time for routine programming tasks I imagine and that seems to be the main takeaway of the article. I guess name dropping quantum computing makes this sound cooler but in principle it’s just automation that you can apply anywhere, nothing quantum computing specific here.
Even with AI you'd want the AI to be writing python scripts instead of following an analysis.md.
Relevant paper [0] "Replication of Quantum Factorisation Records with an 8-bit Home Computer, an Abacus, and a Dog"

[0]: https://eprint.iacr.org/2025/1237.pdf

My C64 is going to absolutely smash this and my dog (Darby) is a frequent barker too! I'll vibe up an abacus emulator as a PWA.
That's a nice article, but it's not relevant here.

The article you linked is about using quantum computer for fake big number factorization. You pick a very big number with a well known easy factorization and then use a quantum computer to factorize it, that is easy because you choose the number very carefully.

This article is about using a LLM to calibrate a quantum chip. It replace the work of a junior researcher (or something like that). In another comment, someone claims that is using a python script for this same task.

I like to think that the python script is an "expert system" that is 1980 AI, and the main article is an "large language model" that is 2020 AI. My guess is that for now the python script is better, but the LLM are advancing very fast and will catch up soon.

Some body compared AI as something like a big boulder rolling down a mountain, decimating anything that lies in its path. After giving it a bit of thought, I feel that what humanity is doing with LLM is exactly that.

The problem is not that it is super smart. It is that it is super dumb and super powerful. Like a boulder falling down.

Humanity is like this bunch of utter morons who has rolled a big boulder up a big mountain and let it loose at the top, and standing at the bottom is clapping and cheering seeing it coming down, guided by random collisions in its path, and with real probability that it will land on them....

Let's go back to caves, this whole "civilisation" thing is not working out.
The trees, man, the trees. We never should have left the trees.
Is reductio ad absurdum the only argument pro AI crowd can come up with?

Like, why not stick to those products of civilization that don't actively try to replace humans who just so happen to comprise this very civilization?

I saw the GP as a snarky riposte against the doomerism of the anti-AI crowd.

Technology has been augmenting or replacing human work for as long as the first slab of rock was shoved up an inclined plane (wedge) and popped into its spot in the Great Pyramid.

It seems that humanity would have went on a lot longer if we never left the caves, so you might have a point there.
(comment deleted)
Something about the look of quantum computers takes me right back to the ENIAC.
My neighbour - a few hundred metres further up the road - has two dogs, one of which a surefire genius as it has no problems barking out 334543333-bit RSA factorisations. It must have been solving NP-complete problems for years by now, night after night.
(comment deleted)
(comment deleted)
I wondered what the birds were singing about every morning.
So new business model meta is: acquire enough compute that you can burn millions of dollars on patentable scientific breakthroughs with unused capacity and on models nobody else has yet. That might actually kind of make sense.
If you had a Genie, would you only use it to make more slightly less capable genies?

This has always been the end game

Same meta as many businesses for hundreds of years: use capital investment to hire labor, use labor to produce goods (including breakthroughs), sell goods and leverage patents. How many patents do IBM/etc have?
Another one of openAI's stolen 'discoveries' for the IPO warm-up ?
Another glorious random number generator experiment?
I find it funny to see all these framings as AI "freeing" people from some work so they are able to do some other work. People should be more prepared for the prospect of AI doing any conceivable sort of work, and this sort of formulation lulls them into a false sense of security.
Along the same lines, I asked ChatGPT-5.6 Sol to evaluate the feasibility of a neural network running natively on quantum computing, you can see the study here:

http://taonexus.com/publicfiles/sep2026/quantum_neural_netwo...

In short it's not really feasible and it suggested classical coherent photonics and in-memory compute as more viable approaches.

(Off-topic aside: these days I am more interested in funding Social Security Trust Funds (OASI & DI Solvency) - if anyone at OpenAI can help reactivate my account: rviragh@gmail.com it would let me do further studies that directly support this important goal, currently my chatgpt account was deactivated. I apologize for any mistakes I made earlier, it won't happen again. Please reactivate my account - thank you.)

Results of prompts are like farts. We each have our own and aren't interested in the ones from other people.
OpenAI is on a quite offensive lately with pathbreaking discoveries.
Dumbest thing I have ever witnessed was the quantum bullrun end of 24 start of 25 and the surge in stock prices of rgti/qbts/ionq and the absolute usless stuff they have since produced beside pr trash.
Every time I open one of the multiple daily posts about Anthropic or OpenAI and read the comments, I get this weird feeling.

Astroturfing is obviously common in political spaces, whether to normalize certain views, manufacture consensus, or shift public opinion. It seems to me that HN would be a prime target for tech companies to do the same thing, and lately I can't shake the impression that there's a lot of it going on here.

This site might be dying.

The worst for astroturfing I think is Kagi. Everytime a post comes up l, every comment is like "Kagi is so great yada yada", often totally unrelated to the article, with little comments mentioning negatives or commenting on the article.

Like, who even use a search engine at this point, let alone pay for it.

If this happens with a minor search engine I'm pretty sure it happens with a lot of other products, starting with AI models.

(comment deleted)
Funny you should mention them but not the multiple daily posts about the Chinese models and labs.
LLMs are advancing the frontiers of knowledge. Is it really that hard for you to believe that people are simply genuinely fascinated about these developments?

You need to touch grass. People are burnt twice shy and have become comically cynical. Your discernment has been completely shot.

The whole thread without a single technical post about the significance (or not) of the announcement. Who want to follow me to create a new site that focus on the substance of the news?

I'm not abandoning HN, I can come back here to make fun of posts I don't like, post shallow comments, snarky ones, vent my conspiracy theories, complain about the evils that are S, A or E, post my personal anecdotes, advocate for my beliefs which I'm convinced as the only truth. Last but not least, feel smug posting about AI sounding posts. HN is a great place for all of that.

Is it a version that has not yet been nerfed? Since yesterday I noticed Sol and Astra have become much dumber.
I was wondering about this! ORCA is high-quality software with S-tier docs, but it's very slow.
I am wondering about using ML/AI for quantum chemistry; ORCA is S-tier software and docs, but is very slow!
I feel like, given how poorly techies view OAI and its executive team, they'd be better off not releasing PR (ie., IPO-farming) announcements like this.

If you've got something objectively monumental to announce then sure I guess (I mean, most of us will just assume OAI is lying/exaggerating), but nobody takes the OAI propaganda seriously. You might even say this is not how you make your first billion.