Hacker News new | past | comments | ask | show | jobs | submit
They have a huge moat, unfortunately. The last models are steps above any other company out there.

The newer Chinese models get somewhat close, but OpenAI & Claude got better at stopping them from copying the model so the Chinese models will not make such big jumps anymore. And most Western companies will not use Chinese models if they are not open weight.

If they keep going like this huge parts knowledge work sector (legal, marketing, software, ads, health) economy is AI driven in 2-4 years and OpenAI & Claude will own it. Another huge shift of wealth from the Western World to Silicon Valey.

I don't know if people even need the best models anymore. Some people I've worked with have expressed a preference for older versions.

Things have gone very very fast. Two years ago, you couldn't reliably have AI write you a medium sized code project. It would spin in circles and apologize. Now, an old model can write it with only a little direction, and the programmer can have a reasonable preference against the newer model.

It seems to me open-weight models will put a floor on the pricing. They're slightly behind, but slightly behind last year meant nothing worked, now it means "just as good".

I agree - for the majority of people, there really is a hard limit on the technology. (Can it do what I want it to do in a reasonable amount of time with reasonable accuracy)

Many of my coworkers liked Opus 4.8, it was the moment they could actually use the models more reliably.

We're there, and moving forward doesn't really get us anything better. Most people are using them to format documents, spit out spreadsheets, pul together data points, and ask the models search questions.

We can already do that, it's only a matter of time before we realize we're paying 200/Mo for Jane in accounting to handle some of her work, when 5$ / Month would render the same benefit.

MOST people simply do not need frontier level intelligence.

The new models can pretty much run without human review. So yeah that flips the perspective: do we still need a human for that task, not which model can the human use.
loading story #50565038
In general I think there's going to be diminishing returns on any given aspect of LLMs because not everything can be infinitely improved upon. If something can write a medium sized program near flawlessly, and I'm writing medium sized programs - then nothing else it can, or cannot, do matters.

Maybe bigger models will always just get progressively better at writing out bigger and bigger programs to the point that we're one-shotting fully functioning operating systems with wide driver support in the future, but again these features just don't matter if that's not what you want to do.

I think we're already limited by our own desires when it comes to coding.

The reason you can't one-shot something complicated is that you haven't generally considered all the important choices. You have to be asked for every choice, and complicated things have a lot of choices. But I think you can do something waterfall-esque, which used to be a big no-no in programming, if you just list out all the decisions meticulously.

I've had some success with doing that, just talking it over and thinking a lot about design decisions, making a document, and then letting it loose on the implementation. Inevitably, you find that there were contradictions or missing choices in the design, and you have to answer some questions.

So, not one-shotting, but it's my fault, not the AI's fault.

You can of course tell it to just fill in the gaps with whatever it thinks is reasonable, and then you'll have something potentially quite different from what you imagined.

loading story #50560486
loading story #50562394
I think LLMs can be improved much more but most likely significantly diminishing economic returns on the investment for that improvement. Ultimately, it becomes too expensive for a few % more improvement at increasing cost. I believe many of the LLM providers are approaching that problem. So it's time for them to commercialize viable long term business, maybe keeping a team still trying to dig up improvements without blowing the bank.
Exactly. After horses, the Ford Model T [1] looked awesome. After a few decades, supercars started coming out. While it is nice to have a very fast Corvette, most people prefer to use a more basic SUV or sedan that is good enough for their day to day priorities.

[1]: https://en.wikipedia.org/wiki/Ford_Model_T

You beat me to it!

I was going to comment that most of us don't need a McLaren supercar (or a Hummer) for our grocery runs -- that Prius actually is Good Enough.

> put a floor on the pricing

ceiling

> They have a huge moat, unfortunately. The last models are steps above any other company out there.

But "they" here is Anthropic and OAI together, and you probably need to add Google onto that list too, now that they appear to have got themselves back on track.

If it was just Anthropic, sure, this would be valid. But with 2-3 companies, we've got constant price competition bringing token costs down (as we are currently seeing) and no guarantee that Anthropic will ever come out with a clear advantage. If it's a three-way split, then that list price seems hard to justify.

And this is all assuming that the Chinese models don't work out how to get back to near-parity, which is also a massive gamble.

Well there is also the rest of the world. And me as an EU citizen doesnt want all this power concentrated in US silicon valley.
A moat would be something that would allow them to slow down. They can't do that. A moat is something that can't be replicated. The performance of their current models will be replicated at some point and sooner rather than later. They can't stop because I don't think they even know how long it will take them to get some better model.

What they have right now is enough money to to work through that. But they don't have enough revenue to keep this up forever and they can't stop either.

I agree they have a moat. But it's not due to model outperformance as much as first-mover advantage turning into a distribution advantage.

Claude is a known brand and installed product for a decent fraction of the economy. To avoid churn, Anthropic doesn't need to stay way ahead of everyone else. They just can't fall behind. The latter is much easier than the former.

Note: I don't have a strong view on this IPO. Don't trade on it based on my comments.

{"deleted":true,"id":50561726,"parent":50558486,"time":1795103632,"type":"comment"}
> The last models are steps above any other company out there.

Google and Microsoft just released competitive models within the last month. Opus 4.8 was just released in May, and there’s probably 4+ non-Open AI/Anthropic models that have overtaken it. You can’t build a moat around a six month technical lead.

I agree with you that the two are way ahead of other companies (apart from google) with regards to model performance. but I doubt everybody will always need the smartest models - for most of the rudimentary daily work in the economic current models are already more than good enough. if openAI and anthropic can use their advantage to offer this rudimentary capability at much lower prices than other competitors then they have a real moat. but just having the smartest models by itself isn't gonna justify these valuations.
Unless prices go crazy, it does.

I have my own little app / ui that I work with and have all models, also round robin to test models, but GLM and Qwen just make weird mistakes & misunderstand the repo like Opus did last year when it first came out.

Where Fable 5.2 and Opus 5.5 are just always spot on. I dont need context, I write garbled prompts and it just gets me always. And it knows better and better, where it should push back & go into more detail and when it should just only do what I ask.

If Open weight models can reach Opus 5.5 level of consistency, I agree with you. If they can't I don't.

It’s not if it’s when. Opus made these same sort of mistakes until relatively recently.
> but I doubt everybody will always need the smartest models - for most of the rudimentary daily work in the economic current models are already more than good enough

Yep. Actually, I do most of my coding work with GPT Luna. It's cheap and good enough.

I don't think they have a huge moat.

because remember good enough is better than "better".

my take Microsoft will be the bigger win, followed by Palantir due to Microsoft understanding how businesses small, medium & large enterprises use software. Microsoft now has its own dedicated FDE org - & for "a.i" to pay off you need a lot of FDEs - since the gains are all on the application layer. the models have become largely undifferentiated

so whether microsoft uses their own models or openai's - if their own models are good enough it doesn't matter they suck by a few percentage points against the frontier they will win.

Google is top of the benchmarks at the moment, and Meta is doing very well too, with this just being the first iteration of the Wang-era models. Grok is lagging a bit, but probably nothing money can't fix.

For many use case the models that already exist are more than good enough, if not entirely overkill to be replaced by cheaper alternatives like Jev.

I don't think having a better model is actually a moat
It used to be when things moved slower. Now it is a temporary advantage which just makes it an arms race.

The companies that win are those that workout what keeps users/orgs that spend money and want to keep spending (or integrate so it is hard to get out)

I agree, as a private user of Claude I don't find the newer models rather useful. But rather a better moat would be something like - a better deal (more credits, cheaper etc.)
Thats just silly. Yeah having a slightly worse model doesn't matter if its still capable. But the latest claude models are getting so good that they actually can run without human supervision in highly complex tasks. I haven't used them enough, but my guess is OpenAI is also getting close to that. For every test I run with Chinese models, it just randomly goes the wrong direction every now and then, making it untrustworthy. That is a huge difference.
Having a good product doesn't mean you have a moat. Think of a car manufacturer. The only moat Ferrari really has is their branding, and maybe a few intangibles (e.g. patents).

Anthropic doesn't even have a brand worth protecting.

A big difference isn't the same as a moat. A moat would be a reason that others can't turn around and copy them.
It depends on what tasks end up producing the most revenue I suppose? Will it be these long running complex tasks, or a series of short tasks, where the difference is less apparent?
loading story #50558787
Is this the first time in history that every time someone finds a new valuable use for this technology, I can use that very same technology to copy the moat?
Maybe oscilloscopes? Decompilers?
> If they keep going like this huge parts knowledge work sector (legal, marketing, software, ads, health) economy is AI driven in 2-4 years and OpenAI & Claude will own it.

Exactly. If you haven't embedded AI in your workflows yet then it's just a matter of time or you're going to crash hard.

It's clear that it can do the work of several people, faster, cheaper and better and that will never not be the case.

They could triple the price (please don't) and I wouldn't flinch.

Again, I think people overestimate how much these models disrupt other jobs. It is good at code and technical writing, but in other fields I am not sure it is as helpful. And in those fields tripling the price would make it an absolute no.
No opposite, coding is at the forefront, everything where AI can verify if it's output is correct, and law is very easy, health also. Will all be as good as coding is now.
> They could triple the price (please don't) and I wouldn't flinch.

> They could triple the price ... and I wouldn't flinch.

> (please don't)

If you wouldn't flinch, why do you care if they take more of the froth? It's all worth it to you at 3x what it currently is, why object that they take what the market will bear?

>The newer Chinese models get somewhat close, but OpenAI & Claude got better at stopping them from copying the model so the Chinese models will not make such big jumps anymore.

I missed that, do you mind giving more details about how it is harder now for Chinese model? What was put in place but more importantly which proof we have it actually is effective?

That's not a moat, I can use Claude at 9am, Codex at 2pm and Kimi at 5pm without any issues.
I don’t think they have any moat… I think they have a noose…

Cheaper open weight models have a potential to undercut them especially when people realise what they can do with these models

This leaves OpenAI and Anthropic to always be chasing the frontier which is going to be increasingly expensive and they’ll never get to stop

>> They have a huge moat, unfortunately. The last models are steps above any other company out there.

A "huge moat" is not being incrementally better than one's current competition, not in tech. Anthropic's moat could be eliminated by one great model from any number of competitors, which means it isn't much of a moat. A proper "huge moat" would be a bundle of patents meaning nobody else was allowed to compete (apple/android) or a decades-old network to lock in users (windows). Anthropic has fanboys but most users would drop it in a hot second if a slightly better model appeared at a similar price.

Moat for what? People don't need cutting edge models.
> economy is AI driven in 2-4 years and OpenAI & Claude will own it

Not really though... OpenAI and Claude will own the US economy in less than 5 years is crazy talk.

My experience is that small businesses have defined computing needs and dislike tech changes. If they get something that works they don’t need progress.

On the one hand this will help the ai companies reduce their tech spending but on the other hand smaller specific models is all the customers need.