Top
Best
New

Posted by therepanic 3 hours ago

I love LLMs, I hate hype(geohot.github.io)
234 points | 126 comments
SwellJoe 1 hour ago|
This line: "this is my main argument against the valuation of frontier labs. It’s not that AI won’t create that much value, it’s that they won’t capture it."

That is a very astute and concise way to explain everything about how the frontier labs are behaving and how they're trying to push more people to pay token rates for the best models. At the current subscription prices ($100 or $200 a month for a generous, though bounded, amount of tokens), frontier models are a no-brainer, most folks and companies will use them. But, at token rates, 10x or 100x the cost of open models or what I was spending on the frontier models a month ago? That is a harder question to answer "yes" to. I certainly wouldn't spend $1000 a month for the best model, much less $10,000; my employer might pay $1000/month, but definitely not $10,000. The frontier labs need everyone to answer "yes" to spending 100x what they currently spend to justify the valuations, and it's just not going to happen as long as everyone knows how to make these models.

Both OpenAI and Anthropic are trying to figure that out now. Anthropic, in particular, has their finger on the trigger...they want to push people to usage-based billing for Fable. But, OpenAI released 5.6 Sol, competitive with Fable (or close enough), and it's available via subscription (even the $20 subscription!), and there's no moat keeping someone from switching. If Anthropic really does end Fable access on the subscription plans in a few days, I predict a large market move back toward OpenAI.

The market isn't going to bear the cost of making the frontiers investment make sense.

andy99 33 minutes ago||
The funny thing about Fable, is we all but know it will be obsolete in a month or two. Between their embargo shenanigans (which IMO they could have avoided just by not pretending it was dangerous) and continuing to give access, whatever marginal advantage it had was essentially wasted.

It would have been an interesting experiment to charge more for it right away and see what the market would bear, rather than tease it for long enough for it to be presumably superseded any time now by whatever is next.

overgard 1 hour ago|||
Yeah, I've started reaching for local models more. I'll use frontier models at the current cost for tasks that the local ones aren't great at, but when the rug pull inevitably comes, to me they're not worth 1000-2000 a month. And honestly, for my purposes I don't really need models to advance a lot. Like, I tried fable a couple of times and there just wasn't much there to justify its use to me. Opus did the same thing much cheaper.

I think an interesting question is going to be, if models are a commodity, who is going to want to foot the very expensive bill to train them? I'm sure training cost will drop.. eventually, but I doubt it will happen fast enough for any of these companies.

hintymad 1 hour ago|||
> But, at token rates, 10x or 100x the cost of open models or what I was spending on the frontier models a month ago

And we can't ignore the power of "good enough". GLM5.2 may not be as good as the SOTA models, but it can be good enough for most, of not all, of our needs.

matwood 22 minutes ago|||
> It’s not that AI won’t create that much value, it’s that they won’t capture it.

Think airlines - both passenger and freight. They have never come close to capturing all the economic value they enable.

QuantumGood 1 hour ago|||
Just to clarify your implication: Fable subscription usage was just (re)extended to July 19
datadrivenangel 23 minutes ago||
And now their comment is good next week! The rug pull getting extended doesn't mean it won't come.
bluegatty 1 hour ago|||
Frontier labs will figure out all sorts of ways to wiggle into the value chain beyond being commodities.
SwellJoe 58 minutes ago||
When do you reckon they'll start doing that?
bluegatty 42 minutes ago||
They are raking in 10's of billions, growing revenues astronomically as we speak.

That's not sign of commodity actor, just the opposite.

Lt_Riza_Hawkeye 1 hour ago|||
While I do agree there will be disruption we haven't seen yet, my company is already spending >$40k/day for a "frontier model", so who knows. Then again, they're not using that for coding
edot 50 minutes ago||
What are they / you using it for in such quantities?
arscan 1 hour ago||
Who is going to end up capturing all this value being generated is going to be very interesting. Back in 1980, who’d have thought MS would capture the majority of the value from PCs over the next 3 decades, and not IBM?
SwellJoe 1 hour ago||
So far, it seems to be the reverse of that disruption. Hardware companies, Nvidia, Apple, AMD, Intel, ARM, memory companies, are all having record-setting quarters, and it's actual profits, not subsidized by investors and circular investments (though the hardware companies are investing in the AI companies to keep the hype train rolling).
hamandcheese 2 hours ago||
> where’s all this new magical software that the productivity improvements should imply?

It's running, privately, in my homelab.

I think we are entering what I call the "have it your way" era. If an open source project doesn't do exactly what you want it to do, fork it, or create a new version. It's too easy.

This makes me a bit concerned about the future of open source. Upstreaming used to be worth it, since maintaining a fork is effort too. But now the balance has shifted significantly. Especially with many projects becoming a lot stricter about contributing, and some becoming outright hostile to AI. I can't blame them. But I think the effect will be that improvements are less likely to make it back to the community as AI adoption increases.

atomicnumber3 2 hours ago||
Remember: code is free as in "free puppy". FOSS communities were never valuable because of the code. It was the shared written and oral traditions that make the software useful, usable, and updated.
hobofan 2 hours ago||
> that make the software useful, usable, and updated

There is a lot of OSS software out there (e.g. in scientific communities) that I would say would barely qualify for each of those three attributes. The main reason it's valuable for the respective communities, is because it's the only thing that's available.

bayindirh 1 hour ago||
Developing scientific software is disproportionately hard though. Making it usable, useful and keeping it updated is even harder.

There's two reasons for that. The math is generally very unorthodox and alien for a seasoned developer, and software development practices are equally alien for the scientist who can understand and evolve the math behind it.

I have written a boundary element method evaluator for my Ph.D. not only math was alien, the required coding techniques for making it fast is very different for a standard developer. You have to have the perseverance and interest to do that. I chose that path intently and I do not regret a millisecond of it.

The problem is, if you don't have a dedicated team to continue that codebase (e.g.: like the Eigen team), your code is basically done and done. If somebody doesn't share the same passion, it's almost impossible for someone to take and carry it forward.

Oh, due to the math and optimizations, the code's structure need to be both documented and the next batch of developer(s) have to be tutored by the person who's giving the code to them.

blauditore 2 hours ago|||
You will likely end up in maintenance hell soon. This will likely not be much easier with AI because coding is not the hard/annoying part, it's the fact that you need to dust off every little project every time a tiny fix is needed, and that's a lot of toil in the long run.
pianopatrick 22 minutes ago|||
Seems to me this would get easier or harder depending on how you write the code. Like if you write the code in something standard and unchanging like POSIX shell scripts or C99 or ES5 javascript, at least the ecosystem won't change out from under you. If you use rust or python or a bunch of node.js dependencies then you might have to edit the project just to keep up with ecosystem changes.
peab 10 minutes ago||||
yeah I had this happen to me. Except when I go to maintain it, now cursor/claude are good enough to essentially handle it on their own, so it turns out to be very low effort to maintain.
fragmede 1 hour ago||||
Maybe? I ran across an old pre-LLM project of mine recently, and past me was an asshole and didn't leave a readme for future me. Meanwhile post-LLM projects at least have a readme that the LLM generated for me or my agent to read and pick up context on. Being able to ask an agent what is this repo, what's going on here? Hey just make it do this, instead of toilsomely digging in and doing it tmmyself, seems to say that might not come to pass.

There is, of course, the question of if that's making me dumber. It might be, but there are other brain training things I'm doing outside of that to force my brain to do the thing.

dakolli 59 minutes ago||
The fact that you're even saying this it is probably an admission that you do think it's making you dumber. Most people I know, who are honest with themselves, have admitted to me that they feel like it's making them dumber or "zombifying" them. This is also well studied already, https://arxiv.org/abs/2506.08872

LLMs are poison for the brain, I'm almost certain of it, at least when used in the way most people are using them. If you drive everywhere because you don't want to walk (but you could), you're obviously going to be physically worse off than if you walked. This is the case with llms, if you have them do all the thinking, planning and action you're going to be cognitively worse off than if you didn't use them.

vilos1611 19 minutes ago|||
It's pretty easy to generalize this, but it doesn't match my perception. People who are using llms to do things they could have already done, but faster, probably have atrophying skill sets. People who are using these tools to accomplish significantly more difficult or complex work than they used to are absolutely finding new ways to push themselves. The problems are just much bigger.

The average Joe can easily vibe code apps that took a small startup just a few years ago. If developers are also using AI to build the same simple apps - then yeah. They're not pushing themselves hard enough, and probably not using their brains as much anymore.

InvertedRhodium 24 minutes ago|||
Socrates thought the same of reading and writing, that it would weaken the memory and isolate people from one another.

1966 saw the peak of calculator protests, where math teachers claimed similar things of calculators.

cyanydeez 1 hour ago|||
alternatively, you might end up in 'good enough heaven' and not have to touch it for a decade because, you know, it does exactly as you need and you're not google, microsoft, openAI or antrhopic.

I'd bet there's far more 'good enoughs' than anything else out there. One of the reasons microsoft office is constantly churning subscription, etc is because they solved good enough decades ago and need to justify valuations that just don't matter for most of their user's use cases.

Not everyone is a software developer having to churn out the 101th SaaS that's just because some MBA refuses to hire a dev.

sixtyj 1 hour ago|||
Creating a fork of an active project only makes sense if you are its sole user (of the fork) and you really need exactly the modification you've been dreaming of.

I have seen so many unnecessary forks of popular projects that I think it's better to stick with the original, even if that means it won't be perfect.

arjie 49 minutes ago||
In the old world, this was because keeping in sync with upstream was hard. In the new world, it takes an hour. And because you're the only user, you can test in prod. Makes the whole thing faster. I have lots of forked and family-only software. Some are abandoned upstream etc.

As cost to software goes to zero, these things become easily possible. In the past, I'd only fork top-quality software (things like `xsv` etc. which is easy to edit. These days even complex PHP software I fork with little trouble.

With lots of software, the value is in the data model and algorithm choices. Sometimes I even just point Claude Code / Codex at an open-source thing I want to vendor some functionality into my personal setup with and it gives me what I want. The hard part for me is modeling the data well. That takes experience with encountering things and it's hard to replicate the edges. LLMs often don't get the rough bits right. But someone else's hard work usually has accounted for this.

paulryanrogers 2 hours ago|||
You still have to track upstream and merge conflicts. Or else you have to get LLMs to fix all the CVEs in your fork.
midasz 1 hour ago|||
I'm guilty of creating a fork that just goes off the rails, but still needs to keep up with upstream. I do it via a skill and seems to work good enough for now: https://github.com/midasvo/findroid-ce/tree/main/.claude/ski...
cyanydeez 1 hour ago||
I'm doing it right now to see what the cost is; I cloned the upstream and made a copy of the working directory and asked the Qwen3.6-35B-A3B model to merge my production files with the new upstream.

Since it's just a duplicate folder, I can always fall back if it fubars.

otabdeveloper4 2 hours ago|||
[flagged]
yowlingcat 1 hour ago|||
I understand the concern and it's fair but I am very curious about what happens when the two notions of "free" (free as in beer, and free as in freedom) start to diverge because the former gets easier to do.

The latter as always been more durable. Linux doesn't have the mindshare it does because it's "free" as in beer - it's because it's "free" as in freedom.

The price of freedom, of brewing your own beer, is sometimes higher than buying it from the store. But for many folks, the control over the supply chain is what makes it worth it. In LLM-land, it might take a little bit of time for folks to catch up -- or maybe a lot of that is already in motion as companies get paranoid (and rightfully so) to frontier labs getting a little grabby about data. If you need a ZDR environment, "free" as in freedom has a very high premium that you will pay and rightfully so.

thegrim33 1 hour ago|||
"This new tool allows for writing all this code ..... but every person and company, in unison, in a grand conspiracy, all decided to only write private software with it that they aren't releasing to the public in any way"

Seems reasonable

coldtea 1 hour ago||
Doesn't have to be "every person and company, in unison, in a grand conspiracy" and other such strawmen.

We could try steelmaning this argument instead: it's enough that most big companies who would otherwise have incentives to contribute.

Before FOSS got in fashion, around the early 2000s, most commercial companies wouldn't touch it as contributors and were openly avert to it, and to open sourcing their stuff. This can be the case again.

nullsanity 1 hour ago||
[dead]
TheAceOfHearts 2 hours ago||
At least for me, the jump in productivity has resulted in building stripped down one-off software for my highly specific use-cases.

You can use an LLM to create anything but you still need to know what it is that you're building, and you need to think through how everything should work or the LLM will just fill it with sausage. You can tell that the models are still quite jagged and limited by the mixed quality from a lot of the software that these presumed trillion dollar companies are putting out. The future is sausage.

lacedeconstruct 50 minutes ago|
This doesn't make sense, I enjoy making bread at home but it costs 10x and tastes like dog shit I dont want to spend my time perfecting the craft of making bread for my daily needs (maybe once in a while its a soothing activity), I want someone smarter than me to spend his entire life coming up and perfecting a solution and exerting more time and effort than I can afford and I am very happy to support him so I can stop worrying about it and focus on what I want to do
dolebirchwood 37 minutes ago||
> This doesn't make sense

Makes perfect sense to anyone good at using these models. What doesn't make sense is that analogy. Typing prompts isn't even close to as difficult to baking bread.

lacedeconstruct 31 minutes ago||
>Makes perfect sense to anyone good at using these models.

It doesn't really, because whenever I ask them what did they actually create, its always a shitty dashboard or a finance tracker or something derivative and worse than what is out there

fwlr 1 hour ago||
I get it, I want to agree, I really do like the “this is a new tool in the toolkit of the professional software craftsperson” argument…

…but consider: the Q-tip. “Don’t use it to clean your ears”, but for most people that’s all they want to do with it, and empirical observation indicates that this dynamic results in either “using Q-tips irresponsibly” or “not using Q-tips”, with “uses Q-tips properly” being a small-to-vanishing proportion of the whole.

ghthor 1 hour ago|
Qtips are made for cleaning your ears. It says not to do that so they are NOT sued every time some idiot fucks up their ear with one.
wizzwizz4 1 hour ago||
But the part of the ear that needs cleaning can be reached without a cotton bud. This is like shoving a sponge down your windpipe to remove mucus.
dom96 2 hours ago||
I love LLMs too, but I am concerned about their cost. They are all still very subsidised. Is there any guarantee that I'll be able to run a Opus 4.8-level model on my personal computer before the big AI labs decide to hike up the prices?
Aurornis 1 hour ago||
> They are all still very subsidised.

I think the opposite: I think the frontier labs have good margins on their inference unit costs.

We can already see what it costs to run near frontier-size models. There are independent business pivoting to serving these models at reasonable prices and they're competing on OpenRouter for costs much lower than frontier labs.

> Is there any guarantee that I'll be able to run a Opus 4.8-level model on my personal computer before the big AI labs decide to hike up the prices?

I would bet good money on prices going down significantly, not up.

If we get to the point where you can run an Opus 4.8 model on your local computer, it's going to be even cheaper for a datacenter to serve it on their hardware. That means prices crash, not that they're going to rise.

theli0nheart 1 hour ago|||
They may have good margins, but a few things are still true:

1. Much of those profits have to be immediately reinvested into model training runs to avoid being lapped by competitions.

2. Unit costs are irrelevant when the labs don't price per unit, and instead charge, for instance, $200 / month for $10k worth of tokens.

This isn't a steady state. Whatever the current situation is, I doubt it's sustainable.

Aurornis 52 minutes ago||
> 2. Unit costs are irrelevant when the labs don't price per unit, and instead charge, for instance, $200 / month for $10k worth of tokens.

Cost to generate all of the tokens divided by revenue generated by selling those tokens is what matters.

The subscription plans confuse a lot of people because that's what they see. They're not seeing the gigantic API bills from all of the tokens going into enterprise use cases.

The subscription plans are a small part of their income. Most users aren't maxing out 100% of their plan usage every week. I wouldn't be surprised if their average plan user was using less than 50% of their monthly quota each month.

Plans like that can produce a net increase in profit if they get consumers interested in the brand and pitching it at work. Giving them some extra token headroom on their $20/month or $100/month home plan is money well spent if it gets all of a company's developers advocating for enterprise plans with budgets exceeding $1000 per person.

dgellow 1 hour ago||||
Interestingly enough, geohot also has an article covering this: https://geohot.github.io//blog/jekyll/update/2026/06/18/pric...
Aurornis 1 hour ago||
That's commentary on company valuations.

Token prices are going down. Competition is global. A company could choose to keep their API prices high, but if another company comes in at 1/10th the price for 95% of the performance then they won't have many customers.

dgellow 1 hour ago||
You’re right, my bad, I read that too quickly
helloplanets 1 hour ago||||
The subscription based plans are heavily subsidized, but the direct API inference pricing (which larger companies need to pay) is profitable.

Using a full Claude Max 20x plan to 100% of weekly usage would easily cost you 2k through the API. While the Claude Max 20x plan is 200 a month.

Aurornis 50 minutes ago||
> Using a full Claude Max 20x plan to 100% of weekly usage

I doubt many of their customers are on the 20X plan. Of those, I doubt many of them are using 100% of their weekly usage regularly.

Comparing the 100% maximum usage scenario of their most discounted plan against the API cost has been a trap in this conversation since it came out. I bet if we saw their financials it would be a tiny sliver in a pie chart somewhere.

user43928 11 minutes ago||
[dead]
dude250711 1 hour ago|||
I thought hardware prices would always just keep going down.
esafak 34 minutes ago||
That is a great comparison. The problem is when costs become prohibitive for new entrants.
ekidd 1 hour ago|||
You can maybe run a local Sonnet-4.5-ish-level model (sort of) for less than the price of a new car, even at current massively inflated prices for fast RAM. This is probably not what you were looking for. But it's there. You could share one server between multiple developers. Maybe make a little AI co-op or something, with a pair of RTX Pro 6000 cards?

Also, DeepSeek V4 Pro is cheap via any commodity API, and DeepSeek V4 Flash is essentially free at API prices like $0.09/M, $0.18/M out. This is generally not subsidized.

For a more practical local setup, Qwen3.6 27B on a used Nvidia 3090 (US$1300) or two is surprisingly nice. It needs clear instructions and you can't use it for hands-off vibecoding, but it's actually quite reasonable for hands-on programmers.

arjie 24 minutes ago||
I’ve got a pair of those cards and DS V4F is incredibly good. I’m happy I did what I did because I like this stuff but if you just want stuff then you are absolutely better off not spending $20k on two of these cards and using the API. This guy is absolutely correct.
EgregiousCube 2 hours ago|||
Guarantee is too strong a thing to seek, but healthy competition makes it highly likely that the supply/demand curve will meet at a healthy place.

You're always guaranteed that you can stash away the open models!

talloaktrees 1 hour ago|||
Currently, because of the subsidies from the frontier models, demand is mostly for higher intelligence.

If subsidies do end, demand for price efficiency per unit of intelligence will go way up.And because there's many players in the market, this demand should be met by at least some of them.

fragmede 1 hour ago||
GLM-5.2 is runnable and downloadable today on a MacBook studio that costs a stupid amount of money. No one can take that away from you except by force though, if you want to set it up today.
andy99 1 hour ago||
He says he might have been too harsh in his “eternal sloptember” post from may: https://geohot.github.io/blog/jekyll/update/2026/05/24/the-e...

I wonder what he thinks was too harsh, still seems pretty bang on, I think it’s going to age well.

fragmede 1 hour ago|
> the adoption of AI agents into software development will be one of the most costly mistakes in the field’s history. Agents cannot program, and it’s taking longer and longer to realize that they can’t.

I think he now thinks agents can maybe program a little bit.

kenforthewin 2 hours ago||
I felt the same way in 2024-2025. Then Sonnet 4 was released, and things started feeling different. Opus 4.5 was another step change for me. Everything feels like it's accelerating, and timelines are getting crunched. I guess in some ways I envy OP, who would "bet everything" against ASI - the truth is I don't know, and I don't think anyone knows, where this ends.
lukeschlather 1 hour ago|
He didn't say he bets everything against ASI, he said he bets everything against ASI being a flash of light in the sky which destroys our chance of getting access to the wealth it creates.
kenforthewin 1 hour ago||
That's a much less generous interpretation of his writing. "Yes we will birth superintelligence, but everything will just sort of work out for us humans". This seems like a silly take to me.
kordlessagain 2 hours ago||
There's good reason to hate the merchants and their marketing. But builders are not merchants. They build with whatever tool is available.
pphysch 1 hour ago|
Geohot is one of the (attempted) merchants, but maybe that is not going so well and he is changing his tune.
fragmede 1 hour ago||
Not sure why you had to add the (attempted) qualifier. He started a company and is selling a box. That makes him a merchant. How successful that venture is, is a different question, but he absolutely is a merchant in this arena.
password54321 2 hours ago|
Yeah I don't think any of the labs have some secret sauce for intelligence either. It seems most of the advancements are still coming from hardware, making LLMs more efficient and throwing more compute and data at problems. And even those problems still require a lot of prompt engineering: https://cdn.openai.com/pdf/04d1d1e4-bc75-476a-97cf-49055cd98...
andy99 2 hours ago||
The secret sauce is training data. They’re not just taking advantage of more compute (which obviously is necessary but as mentions basically a commodity). They are paying billions to data labelers and making judgements about the nature of the training data they best need to make the product they want. This seems to get pushed aside as a minor point but it’s the primary differentiator of the big labs.
password54321 2 hours ago||
As a I said, compute and data. But LLMs can be distilled, so even their data is not much of a secret sauce.
reinitctxoffset 1 hour ago|||
I'm pretty sure at this point that Anthropic is training mixture models (at least in the heavy pre-train) and deploying them dense with explicit loss on thinking trace coherence.

Having a thinking trace that is legible, coherent, and immediately implies the explicit turn output and/or tool use seems difficult if not impossible to reliably get from mixture models.

I predict MoE is a transitional technology, it's got too many problems and the benefits are...kinda grandfathered into the dogma at this point.

dominotw 1 hour ago||
even meta that sucks at doing anything is releasing frontier models. making an top ai is easier than making twitter clone( threads) if you have enough money.
password54321 49 minutes ago||
I mean the problem with Threads was lack of user engagement. The same could possibly still be said about their models.
dominotw 40 minutes ago||
yes ofcourse. But engagement needs strategy and execution.
More comments...