Goddamm... This could easily have housed every man, woman, and child in this country. But no, no free housing while we squander wealth with our stupid ideas.
also before court cases for defamation. Two cases in Germany hold Google executives responsible for false claims (by the Google AI) of sexual crime claims against an innocent man.
yeah. It should be the same with Facebook. Here in Australia, it seems like nothing is against the community standards anymore (not spam, porn or even violent threats)
yeah the circle-jerk nvidia has created with coreweave and so many others just to wreck the gpu and memory markets, all so these hideously expensive AI racks can SIT IN THEIR FUCKING PACKAGING ON A LOADING DOCK because they don't have power or cooling built is mindboggling
That's why they keep trying to play the "AI is so dangerous" narrative and why you keep hearing about "rogue AI agents" which in reality someone had to instruct to break the law, and they probably used some of their no-guardrails internal-use models (so Claude Mythos rather than Fable).
Claude Mythos and Fable are the exact same model, the branding only reflects which one has more restrictive classifiers in front of it during deployment. But the weights are identical.
The issue is they are too expensive, it's not really a limitation of the technology, well it might be, but that's not the issue here. They are paying Nvidia Billions for their entire 2026/27 GPU runs and they can't get the customers to cover that. It's like $200 pm for their max plan. It's almost cheeper to hire a human in India.
Have you actually used it recently? It’s become incredibly good. Calling it a glorified autocorrect is like calling ASML’s lithography machines a glorified polaroid camera.
In 20 years we ALWAYS look back and wonder how so many were so fucking dumb. Thats literally history. Theres never a point where there isn't some major fucking bunch of absolute idiots trying to ruin everything.
Ai is just this decades attempt at ruining everything.
But it's also objectively fucking stupid to act as if this is some new historic event in human history and not just yet another dumb fuck move in a long line of dumb fuck moves.
Plenty of good reasons to be pissed off. You don't have to lie and bullshit to make up more is what I'm getting at.
Fucking yes it would be better to put hundreds of billions of dollars every year into curing cancer/HIV/ Type 1 diabetes/alzheimers, literally any disease. Curing a major illness like those would actively help hundreds of thousands or millions of people around the world every year.
What does AI help? My fucking manager in sending cryptic emails that raise more questions than answers?
Dump a few hundred billion into disease research, or fusion power, or just into housing for all of us.
AI helped me with my voting ballot. I asked Kagi's AI to research the ambiguous candidates and propositions. I had it put them into a political alignment chart - left/centrist/rightwing. Addendums for candidates included things like whether they had criminal convictions or corruption, stance on vaccinations, how they feel about "tough on crime", the homeless, ect.
For propositions, the political alignments had a fourth entry added, where the yes/no included the wealth level of backers for a position.
This is very useful, completing in a hour what would have taken a week of research. Very useful for digging into the mechanical details of a proposition, like whether it prevented the replacement or supplementing of a measure.
How do you know that anything it displayed was accurate for those candidates? Every single time I've asked it to include source links, those source links don't actually support the AI's statement.
One of the candidates was appointed to their office by a Republican governor, and elevated more recently by a Democratic one. The AI referred to official government records.
If an LLM was something you could train on your GitHub repo or an internal knowledge base, and ask it natural language questions about that, AND it didn't need an 800 acre datacenter to do it, I wouldn't have so many hang-ups about it. Or maybe even train a model in a programming language and have a model that's tailor made for writing C, or LISP, or Brainfuck or something.
But that's not what they are. They only function if you hoover up everything the light touches and feed it to the machine, to the point where rare, irreplaceable literature is being shredded apart for training data. And god help you if you try to take your ball and go play somewhere else as the AI boosters profess you must do, they will follow you there and scrape you anyway
$8B real loss on $16B revenue. ($72B is a bondholder gain (accounting loss) from conversion). $9B in training costs not counted as loss, even though it is worthless whenever next model version is released.
$561B in forced spending on compute over next 10 years, with $750B total compute costs lined up. Structural losses are baked in for a few years.
But, when the word guessing program gains consciousness, it will be capable of doing the job of every worker in the world, and eager to do those jobs, and anybody who owns Anthropic shares will be a trillionaire in an economy that totally works and is healthy. So, better invest now!
Musk giving 1 share (about $160) to all US children with a Trump account, will make the next generation cheer for skynet genocide and oppression. "For a while, it added 10-20% to my net worth."
even though it is worthless whenever next model version is released.
For most version releases I actually doubt that, I'm pretty sure they just keep training the model and release checkpoints every few months. Architectural changes, however, will indeed invalidate the previously trained models, and occasionally they have to do that to keep up with the competition.
But if you do architectural changes even just once a year, that means you'd have to depreciate those training costs over just that one year to be truthful. Meaning they might as well be losses rather than depreciating capital investments.
There is a genuine "training corpus" that is an accumulated asset and grows and feeds new models. The vast majority of training cost, that should be expensed, is the training time compute. Neither of these are expensed. There was about 13 major US model releases in September. At least 7 of these replaced a 1 month old previous model family. Open AI gpt sol went from 5.6 to 6 to 6.1 in about 2 months.
Fast model releases not only lock in structural losses, they juice revenue growth illusions. Model use rate generally drops by 66% 4 weeks after release, because of competition, and a surge of use on release for evaluation.
Yes. If you follow the money, investors in OpenAI are the usual suspects, a Kushner in there (admittedly the "best" kushner) and the early investors were guaranteed returns of up to 18% (bank CD currently around 5.3%) which stemmed from a $500B injection from the US government. Not only will it be a rug pull, but a government built one. Truly appaling, we let the fox in the henhouse.
Cant have Healthcare for our people, but can pump and dump the world economy.
The loss, up from $13 billion a year earlier, was driven by $72 billion in accounting charges tied to changes in the value of financial instruments Anthropic issued to raise money
This right here is how all of this will come crashing down. Anthropic and OpenAI, along with several associated companies, are using accounting tricks to hide their debts and obligations. These two are trying to maintain hype for investors so they can cash in on an IPO with total disregard for long-term fiscal sustainability. At some point the bill will come due, and you can bet the C-Suite will have already made out like bandits.
Don't forget, though, that there's another cashflow that they're banking on: total surveillance.
AI can't do a lot of things, but the thing it's most suited for is creating a society run like a prison to try and keep the discontented masses in check.
Not enough people realize this. They think the component shortage is a side effect, but it’s to push you towards cloud services on underpowered devices that rely on it.
Imagine the censorship when every service is monitored and controlled by them. Imagine the market capture when they can see everyone’s financials and trade secrets. When they can cut off journalists and activists with one click.
Haven't heard anything new lately, but there is a push to eliminate physical currency for, what I presume, is a similar reason -> make everything trackable.
I think this will backfire too. I think they way underestimate the amount of people already using "underpowered devices" that will just continue to do so without the new cloud services shit.
Basically everyone I know us using a 5-10+ year old laptop, for example. And they just use it fine, sure it's pokey, and it's not everyone, but it's a lot.
The Italian government has just passed an emergency decree that authorizes AI-powered face recognition in public spaces without a judge's authorization.
I'm not even sure it's great at that. Is this not a function of any industry struggling to find a usecase eventually ending up with the largest US employer and cash cow: the DoD?
LLMs ('AI'), not so great, but it can get (probabilistic, but fascists don't care) descriptions of video feeds into easily searched text. Gobs of GPUs in datacenters paired with surveillance devices all over the place however presents a greater threat, where there's an (evil) will, a way will be forthcoming. Nice of Ukraine to point out how fragile they are. There's a deathstar analogy there somewhere.
Yeah, the type of people pushing this don't even care about the inaccuracy either. If it "catches a black man" with facial recognition, even if it's the wrong person it won't matter "they are all criminal thugs anyway." See also the ICE bull shit and basically deporting anyone even remotely Mexican looking, citizen or not.
This right here is how all of this will come crashing down
No, this is pretty normal, and more importantly, doesn't actually cost the company cash. They gave lenders bonds that can be converted to stock, so this loss on their books is pure paper loss, the same way wealth in owning stock in this money losing company is pure paper wealth.
These numbers are just illustration for simplicity, but imagine a company borrowing $20 billion and telling the lender that you'll either pay back the $20 billion plus interest, or they can opt to convert it into stock worth 10% of the company. When the company is worth $100 billion, the bondholders won't convert to stock. But if the value of the company shoots up on paper to where that same stock would be worth $100 billion, the lenders would rather take the stock than the cash repayment. And because it's worth $80 billion more, the company will have to record that on the books as replacing $20 billion in bonds for $100 billion in stock, and an $80 billion loss compared to their last statement.
But look at the underlying financial cash flow. When that happens, the company actually takes $20 billion in debt off their books. They don't have to repay that loan anymore. So it improves their cash flow position at exactly the same time that they're booking a loss.
The whole thing will come crashing down for totally different reasons, relating to operational cash flow, not the actual accounting shenanigans where one kind of paper number gets converted to another kind of paper number.
While you explained this correctly (as far as I understand it), there is part of this everyone keeps making an assumption on the outcome, the "or they can opt to convert it into stock" part. I'm really curious what the interest rate they gave the early investors to entice them when the company was much further from a sure thing. If the interest rate was decent, that final number is going to start becoming awfully enticing to just collect on instead of getting stock that might collapse.
That's the tradeoff, of course. If a company liquidates, it pays its debts first, then its preferred stockholders, then its common stockholders. But each step down the line gives a greater upside in exchange for less certainty in getting paid at all. Bondholders get paid at higher priority, but have the exact amount they get paid defined by a specific mathematical formula. So they won't get any additional benefit from the company becoming super valuable. Or they can get in the back of the line to see what unlimited upside they have from the company's value going up a lot.
Also, Anthropic's bonds aren't exactly low interest. They're right there at the edge between junk bond and investment grade.
It's not entirely transparent though. It's a paper loss in terms of cash but they did have to give the stock. At their crazy valuations, even 1% is a big deal.
With the difference that all those services mostly outsource their asset costs to their drivers or have very little investment there. Unlike anthropic who have billions in costs related to all the compute they need to service their customers (obligations to other companies, datacenters and, not to forget, ever rising interest on all their debt).
I'm doing my part by constantly asking chatgpt questions without ever logging in or paying them money. Eventually they'll all go belly up and maybe we can ditch this shit.
It is obvious to me that AI computing is a COMMODITY, in that it will inevitably be something that can not be effectively gate-kept.
These two major US companies (OpenAI and Anthropic) are playing the game of “locking in” a monopoly in a product category that has yet to mature. This is a pretty old capitalist game, of course… Uber, Microsoft, facebook,and many others have played this game in the past.
But, unlike OSes or word processing file compatibility I don’t see a “network effect” in that an open source/weight model can/will do 99% of what a proprietary model can, will do it “transparently,” will eventually become so optimized that it can run locally, etc. etc.
I mean: maybe its name recondition that is worth “trillions” but a race to be “Xerox” or “Kleenex” or even “PC” (as in IBM PC) did not really work out well for these companies.
I think that even the established players (MS, Google, etc.) are biding their time, making sure that they don’t get locked out while non-proprietary solutions become feasible.
I means I know that much of the issues have to do with the limited availability of hardware, but locking down supply is not sustainable for the “big 2” and sooner or later more adequate memory and GPU supply will enter the market…
I dont think a local model will do 99% of what the big ones can, and even if it could, it would require an enormous amount of memory and cpu to do it.
People pay for a good user interface and convenience, that's well known. Netflix and Spotify are built on that idea and people starting using them even over free pirated stuff :)
Qwen3.8-flash-next broke the vram requirement with its new architecture (which is a preview of their 4.0 architecture). You can run a 177b parameter model on 12gb VRAM, 64gb ram and 80gb SSD at 30tok/s and the speed scales almost linearly with increases to vram.
Incidentally, this was proven at almost the exact time all the calls for "regulating local models because they'll spread into the toasters" started.
You absolutely can have local models that run on consumer grade hardware, and the proprietary companies know it and are scared to death.
Are these really able to be as good as chat gpt? Because every time I've tried a model like that, it has not been able to be nearly as good. I have 48 gb memory on my machine so I can't run this particular one though.
I would buy more ram and a whole new computer if the market wasn't insane.
It is good enough to be useful for programming. It can do clearly defined tasks like find the root cause of a bug.
Using it for writing code is frustrating because it is slow and produces lots of tokens. Meanwhile Opus 5.5 is good enough that you can pretty much rely on it doing exactly what you say (and not what you meant), faster, more context efficient and you get a ton of usage with the cheapest subscription.
I can see this changing, though. If the open-weight models reach the level of Claude 5.5 and Anthropic stops subsidizing they might be on pretty equal footing. I don't see how further improving the model would help, except maybe if somebody invents a different way to manage context that remembers all the important details over a long time so it actually "learns".
I won't pretend to be an expert but my understanding is that the architecture loads experts from the SSD on demand with the RAM serving as a hot cache and the VRAM running the active experts.
As I said above I was able to run 70tok/s on my 4090w64GB ram and a samsung pcie4.0 ssd.
I haven't run the benchmarks myself, but I can say the quality of response I was getting seemed on par with the frontier models I use for work (Claude opus/fable) and the published benchmarks for the model compete with those same models.
It is a q3 quantization so there is some small loss of accuracy, but from what I've understood those are modest losses in the worst case.
There are models that don't need to consume the entire worlds data infinitively. We've had then before llms, things like the pi ai camera can do ai processing on image with something like 8mb of on board storage.
The problem is anthropic, open ai, who ever are trying to do the stupidist approach and assume they can brute force money and power at the problem and don't understand that they need exponentially infinite data in a way that is unobtainable.
In an ideal world there would be a million little small local models that so specific tasks and can be ran on consumer hardware.
I’ve been consuming lots of “build your own local AI coding agent” kind of content lately. No doubt this is the same as the guys that are welding together their own versions of Shelby Cobras and stuff… but in a few years you’ll be able to buy a “server” with enough GPU and ram to host strong open models and serve them to your little office.
I work in a small company, and if AI plays out the way Anthropic sees it’d their AI bill will be $100k every year (fire 2-3 people, pay for AI so that the remaining people do more). At that level of expenditure, why wouldn’t my poss pay as much as a copy machine $25k, lets say, go give “local AI” a chance?
The stock valuations that most tech companies carry (and certainly the 2-3 trillion that is being talked about for Anthropic) speak to potential future revenue.
To me, I see computing and actual “model” advances meeting the need within a few years… if not by entrenched companies like Nvidia, then by insurgent chipmakers or new (Chinese or otherwise) chipmakers.
In 10 years you’ll have “opus” is not better running in your phone and probably something more powerful in a PC.
A local LLM is more convenient than cloud. But it is way less efficient. The cloud LLM processes multiple people's tasks at once because generating one token is mostly loading things into GPU memory, the compute required for inference is not that big. When doing multiple at once you can load one stage onto the GPU and then do that stage for a number of tokens for essentially the same cost.
I dont know, maybe, but I think either way, local models will be a fringe interest for technical people, and not something that will affect the income for these big tech Ai corps in any significant way.
Because you simply can't run a model with all the worlds knowledge on your home computer.
Someone said that maybe there will be small specialized models that only know about a certain thing, and sure, that's possible.
You also can't run a model with all the worlds knowledge on a supercomputer. That simply doesn't exist. To think that even the largest frontier models can do that is a mistake.
The real question is: will the median user be able to run basically everything they need locally, and how much compute would they need to do that? These companies are betting the farm that it will never happen, and that they'll be able to charge whatever they want in rent to a captive audience.
But I think these tech CEOs and ML researchers talking about "AI killing us all" is related via the aphorism: It's easier to imagine the end of the world than the end of capitalism. They can see the writing on the wall, but can't imagine a future for themselves.
A lot of their revenue is also investment from hardware and software vendors that are using the revenue from the AI companies for those investments.
It's like the Enron scheme, but instead of one company inventing fake companies to move money around to increase revenue and drive up prices to hide the fact they're hemorrhaging money while the C-suite tries to divest their own money from the company, it's multiple companies working together to move money around to increase revenue and drive up prices to hide the fact they're hemorrhaging money while the major investors try to find a way to cash out.
Anthropic is anxious to do their IPO because they know their business model is fucked, and their shareholders want out.
Are they? This is productive capital being rented out at a profit to the end users. And the end users are, as often as not, big public institutions with near bottomless accounts.
I might argue they are rapidly exhausting scare natural resources. And you're welcome to do the material analysis on whether dumping 50 gal of water and 100 Mbtus of energy through a digital engine to make a "Farting Peach Says It's Voting For The Green Party in 2026!" viral Facebook ad is benefiting humanity.
But merely observing that money is changing hands between big economic players doesn't let us know if something is profitable. You're describing the basic functions of a financial economy.
Anthropic is anxious to do their IPO because they know their business model is fucked
That's the popular line. People said the same thing about Tesla when it went public.
Yup, because they're still losing hundreds of dollars for every dollar spent. Their economics still rely on a delusion that the hardware outlay is a fixed cost that can be recovered after model training by selling cheaper compute. Instead, models continue to be released not only to match competition, but because models still do not do a fraction of what they are promised and being sold to do. So capital costs are not fixed.
So AI hyperscalers have a lose/lose/lose/lose choice of either growing at a rate that requires unsustainable subsidies, stopping growth and admitting models can't do what they and a thousand other startups have been selling them for, charging actual costs, or any combination of the three.
If people were charged anything near the actual costs, much less with a profit margin, demand would implode fast enough to create a sonic boom.
Add to this that incremental improvement comes at exponentially more and more cost. So they spend ungodly amounts more in hopes that the next incremental improvement will push them over the magic line beyond where the are.
Some impressive feats but the issues holding it back from the 'replace the whole economy' sort of ambition they imagine are very stubborn. Notably it still confidently declares incorrect information (a significant volume of software developement works around by rigging auto-retries until it stumbles into 'correctness', and of course the people pumping out slop art/books aren't bound by 'correctness') and they still have no clear path to autonomous control of humanoid robotics (as crazy advanced as the online GenAI experience, they are still struggling to demonstrate even the most basic manual labor tasks).
Idk man, to me, it feels like we are further from AGI than fusion reactors actually being hooked up to the grid, considering LLMs are still just glorified autocorrects, needing a ton of prompt injections from their providers. And yet, the AI companies make it seem like it's right around the corner (obviously, because $$$), whereas to me it feels like we are still at least 50 years away. Maybe AGI will be the next fusion generator chase.
I don't think they're "glorified autocorrects", this might have been true a couple of years ago. The improvement of LLMs, month by month and sometimes even week by week, is spectacular.
Ironically there was notable news in the fusion reactor space this week as Gemini 4, building on the work of Matt Landreman and GPT-6 Astra a little while ago, expanded counterexamples to Grad's conjecture which had claimed exact solutions for controlling a key aspect of fusion were impossible.
So fusion may actually be more around the corner thanks to the advances of capabilities of AI which helped prove that the fusion problems aren't as hard as they were thought to be for decades.
And if you actually believe that, I have an amazing business proposition to you. So, there is a bridge in NY, and I can sell it to you for very little money...
They did manage to do something different, though. I thought at one point that proof assistants should offer to automatically upload your proofs to a central repository so everyone can benefit from them. But such a system would require many to opt in for it to be useful.
What AI does best is giving you existing solutions. It even adapts them if they don't quite fit. Because many people now discuss their ideas with AI, ideas may spread the same way as well when the conversations are used as training data.
So no, definitely not AGI but valuable. Though as typical of explosively growing startups the value does not come from the technology but from activity bordering on criminal that lawmakers can't keep up with.
AGI has been achieved more or less because it just means self sufficient at completing tasks. An LLM is just a component these days because they are Vision-Audio-Language-Models with multi frequency 'thought' loops. The next letter is A for action where they just have high frequency outputs for direct computer and robot use.
Just because it is bad AGI doesn't mean it is not AGI right
I just recently installed Strata which runs on qwen 3.8 flash next. I'm getting over 50 toks. Now a deepseek v4 version is out along with glm 5.3 flash. It's slower but useable and very capable. All this on a 5080 16GB and 128GB ddr5.
The future for me is going to be open weights and on your own hardware.
That's how it needs to be, and how it should be. Otherwise, we are all permanently screwed by big tech, subscribed and licensed to even own a computer by 2035; MicroSlop says open-source AI is good somehow, but we will see through their actions if they lock down their new line of NVIDIA Spark laptops to an approved number of models. I heard rumours that Deekseek might be banned in America; I can't see how that would work though.
Deepseek is open weights. You can just download and use it locally. I would never run deepseek using their servers. Or z.ai servers. That I agree with China is still pretty bad.
While I have no problem with most of China's principles regarding business. I don't know about all their politics and laws, nor have I given them much thought, but they deserve more respect than they get for essentially being the whole world's industrial estate and being able to pull off the amazing things they do while having virtually no copyright laws. Open-sourcing so many things we would rather make proprietary, thus very expensive for ourselves. I'm grateful to them for deepseek! Even if they want to spy on me a little, at least I know to expect it lol =)
I mean that sounds great and I'm happy for you, but you're working with hardware that costs more than most people on here's cars when it's available at all. How viable would your workflow be on an 8gb 1080ti with 32gb of ddr4?
Running locally makes little sense anyway because computing multiple replies at once costs the same as computing one. A home setup isn't going to get utilization anywhere close to optimal.
At no point did I say it was man. Just not something most people have access to. "Got to have money to make it" painfully true as usual. I have $234 in my wallet, a late 90s Honda to crash in, and sheer force of will to try and change that (it's not working lol). The 32GB/1080ti example is the nicest computer any of my friends own, my own rig is a scavenged parts pile centered around a 10th gen i7 that will be complete when some ram falls into my lap and right now I'm posting from a phone that was $150 6 years ago.
You know, that's fair. Most people do not need a rig like mine. I would still be using my 8 year old computer if my new clients data wasn't so sensitive that I have to run local inference.
As someone who was also broke and lived in an old Honda, I know what you're going through and I've been there.
If you truly want to change your outcome, you gotta find a job that has high growth potential and a low barrier for entry. Personally I worked at a bank for almost a decade. Started as a teller(low barrier,no college needed), ended as a regional manager(high growth). I hated every minute of it. But it gave me a chance to save up and find a career I enjoy.
Good luck my dude. You will be successful. I believe in you.
Nice setup. I'm running Qwen3.8 27B at 17GB with a 32k context window on a 3090. It does most of what I need for everyday use and its all local, private, no internet connection needed.
The gap between running an LLM locally and cloud API is closing fast.
I mean, besides oil prices, the only reason why hardware is more expensive than it was before is because people are hoarding it, fearing that it won't be available in the future. But there's no reason why new and better hardware shouldn't be engineered and made available unless there is another world war.
The electronic hardware cartel has stopped designing new hardware, but it was a deliberate choice to inflate prices. They even backpedaled on 5090s
I started looking into it finally after making that comment and came across strata come back here and everyone is saying strata. I guess im trying strata?
need 64gb of system ram. qwen next architecture uses smaller experts than others, and it can keep up with ram swap. I believe it uses a file on disk (ngram) to predict what experts will be needed.
Running on your hardware is always going to be less efficient than running in a data center. You can run small models on green data centers - best of both worlds (unless you need privacy then run local).
Ever since I switched to Cortecs I'm spending like 20 cents / month for my LLM usage cuz I mostly use small models and try using them responsibly.
That's not how that works… the money "lost" isn't going down the drain. It is… going somewhere, doing something useful. That's why it is a loss to the company.
If it's going to make LLMs it's not as useful as the other things I listed. Paying a bunch of software engineers $300k each to work on LLM nonsense instead of something useful is a shame. Sure, the money isn't set on fire, but the opportunity cost is large.
I mean... A lot of the money is being spent on compute, which is just consuming electricity, which is many times made from coal or gas. So a lot literally is on fire.
[...]included a $72 billion accounting charge reflecting an increase in the estimated value of bonds that could eventually convert into Anthropic shares, rather than cash costs tied to its operations, the document said.
So dept isn't cash cost that's tied to operations? All the debt did was finance the operations. Reading this paragraph and you'd wonder: who in the hell issued those bonds in the first place?
included a $72 billion accounting charge reflecting an increase in the estimated value of bonds that could eventually convert into Anthropic shares, rather than cash costs tied to its operations
It's a very common way for startups to raise money. In this case, the bond value goes up because of a perceived increase in the value of Anthropic.
Repaying debt is cash cost. But this is convertible debt that could be converted to stock, in which case the holders of those newly converted shares don't get paid cash from the company, but instead own a chunk of the company worth more than that cash. So the conversion itself would alleviate the cash demands of the company, while also diluting the ownership stakes of any preexisting owner. That's what that paragraph means.
Companies issue convertible bonds usually because they can get better interest rates on it, in the event that they need to repay it. Or they're happy to deal with diluting their shares if the company becomes so valuable, because that means they've gotten a lot richer and just have to share their newfound riches with the lender.
Anthropic has committed to spending at least $518 billion over a decade on cloud services, computing and infrastructure.
They're betting everything on AGI. I don't think anyone there at any point has even hoped that the tiny initial revenue streams would cover the absolutely gigantic upfront costs of setting everything up.
Now, whether they'll actually reach AGI before this thing implodes is anyone's guess. The people who claim that they absolutely will are full of shit just like the ones claiming they absolutely will not. Nobody knows and only time will tell.
I'm just going to throw in my 2 cents on the AGI issue since I have a better understanding than average. I'm currently a computational neuroscience PhD student. I don't work in theoretical neuroscience currently but that's my hope after I finish my doctorate.
I don't believe current transformer architectures are capable of AGI and all big AI companies just seem to be adding more parameters to established transformer architectures.
::: spoiler Tap for spoiler
From what I understand transformer architectures act similarly to primary sensory cortices, they model complex probability densities. These companies appear to be banking on the idea that adding more training data, having smaller networks working together with different expertise, and having the train of thought (phonological loop) modeled as part of the PDF is enough to reach AGI. I disagree.
While all of those things are interesting solutions to parts of the problem I don't think they address some of the bigger issues.
in biological intelligence memory recall is built into the same system as information processing. Associations between things are learned by simultaneously recalling and processing
decision making and higher order processing do not fit into a transformer architecture
temporal encodings are only made, in the transformer system, via the attention mechanism which is much less efficient than a more biological mechanism.
:::
Haven’t there been developments in biological computers recently? Wouldn’t that be the key to AGI? I am a total idiot so I know zero about what you are doing or talking about. It just makes more sense for a computer to think like us if it has the same hardware as us is how I see it.
There are positives and negatives to using cells for computing. There are very complex systems that are responsible for making sure cells grow in the correct architecture and are supported property via blood vessels and glial cells. There is also definitely something to be said about glial cells themselves having a role in computation (whole other topic I won't get into) not just neurons.
If were just looking at the input output relationship with neurons we can model it pretty well using neuromorphic hardware which is designed to respond like a neuron. There are still gaps but and there are scientists who believe it isn't close enough.
Our current hardware can model neurons closely as well but its less efficient at it. The current AI models use point neurons which takes a lot less computing power but I don't think they're the way to AGI. Personally I'm a fan of the adaptive exponential leaky integrate and fire model or the spike response model
Good question! Unfortunately, I haven't found a good source myself. I've been trying to piece it together based on my theoretical/mathematical neuroscience background and self-study about transformer architectures. The problem is the point neurons in transformers don't map to neurons in cortex. You have to instead look at the bigger functions that are being computed by each system.
The best single source to learn more about theoretical neuroscience is Neuronal Dynamics: From Single Neurons to Networks and Models of cognition. But its super math heavy. If you want to easier to digest videos I recommend Artem Kirsanov on youtube. For transformer architecture 3 Blue 1 Brown has a great series of videos about it on YouTube as well.
I kinda feel like actual AI will not be made by humans, but by a model that is able to reprogram and upgrade itself until it has some sort of agency. That said, my experience in the field is from the 90s so I really know fuck all that's relevant today.
Generally the benchmark for AGI has been "human level intelligence" meaning that from there on AI is also the best AI researcher and human input is no longer needed in developing it. That's where the concept of singularity or intelligence explosion stems from.
The narrative here tends to be that it goes from AGI to ASI (Artificial Super Intelligence) very rapidly but in my view there's not a significant difference between the two. AI that is as intelligent as a human but not more would still process information about a million times faster so effectively it would still be superintelligent.
I disagree on the speed. The "human level intelligence" would be about as fast as a human on tasks. It might do them very differently, so it could be superhumanly fast at some things like reading a whole book.
The thing that I doubt is that this would create the prophecised feedback loop. Sure, the AI could be a bit better than a human at further developing itself but it is probably also really hard to increase the intelligence.
One thing that can't be denied is that AGI could be taught once and then cloned instead of all the work that goes into training humans.
That makes sense. That said, I'm not sure superintelligence is just a matter of processing speed. It's an interesting question - give me a complex problem and say a thousand years to work on it. I kinda feel there's some problems I'm just not equipped to solve, no matter how much processing time I have. But of course, that's impossible to test. And in computer terms, you could copy a whole team of me to work together. Hmm.
Yeah, I concur that the difference between AGI and ASI is likely moot. If an intelligent machine is able to upgrade itself to be more intelligent - well, I think we'd all do that if we could! I'm not sure that more intelligence and faster processing are the same thing, but I'm also not sure that we could distinguish between the two in practical terms.
Oh, AGI isn't for you. Its for the AI cultists to worship once they summon it. The AI cultists think they can control thier god, even thought it will be integrated into everything if it is truly super intelligent. Then they will realize that they didn't learn from all of the dystopia Sci fi warning them about summoning their god.
AGI isn't for us, it's to replace us. AGI is the pipedream they have been selling of actually reducing your workforce by offloading it all onto AGI agents. They either don't reach AGI before their debt obligations come due in 2027 causing the bubble to burst and causing the next great recession or the achieve AGI before that happens and everyone slowly loses their jobs causing the next great recession.
Yeah, I think they forgot about competition or assumed they wouldn't have it by the time it matters.
It only works if they stop training new models and can actually follow their 6 year depreciation schedule on their GPUs. But that would make Jensen Huang cry, and they do have competition right now. You can't put a bunch of effort into creating one AGI model because then a few months later OpenAI or Anthropic releases a slightly better model and all your contracts go away next year.
I am in software and I don't think it's that clear cut tbh. There's definitely a right and wrong way to use AI. I think a lot of companies are moving fast and breaking things right now, but a lot of them are going to find out that the thing they broke is the product and they won't be able to slop their way out of a problem they slopped their way into.
At a micro scale, AI is pretty good at writing workable code quickly. This obviously boosts productivity at the individual level, especially in the short term. At a macro scale, good maintainable software requires higher order thinking that transformer based models simply cannot do. So a lot of the stochastic micro decisions an AI makes lead to technical debt that is not trivial to remove, and can't really be resolved by the AI itself, at least not in a way that won't contribute to the problem.
These issues can be curbed by the programer at the point of origin, before it gets out of hand, but obviously that requires more judicious use which cuts into the productivity gains. Because writing code has never been the hard part of software.
TL;DR: fully embracing the slop machine will eventually poison your product in a way that is not really easily recoverable. It is useful, but not "too good to not use."
What good could hundred thousand of the world’s best scientists do for me if they had decades to study every question I asked and could answer them immediately?
Nothing I guess if you absolutely want to figure it all out by yourself.
AGI “use” has ethical issues of its own in its own right. But if we dismiss those (we shouldn’t) they will be extremely useful.
Imagine a personal assistant that will organize your life and manage all the trivial but tedious tasks in your life. There is a reason why crazy rich people all hire personal assistants.
I'm not knowledgeable enough for my opinion to have value, but I keep hearing that Chinese models are achieving similar results without the huge infrastructure investment - cos we aren't selling them the chips. If you can achieve similar results without spending hundreds of billions on datacentres - then it's possible that these companies have made an oopsie.
I keep hearing that Chinese models are achieving similar results without the huge infrastructure investment
yeah but it's not what you think. you think the chinese are just magically making these models appear from the dirt. the reason they're able to make their models without tremendous grunt powering them is that they distill other people's models - the theft is built into the entire ecosystem, they started by stealing from writers and artists, and are now trickling down to stealing from openai and anthropic.
at some point, someone's got to build the model with inference and that takes a lot of computation to push through.
Got a local food truck epidemic of mid overpriced food and no food safety standards, in my college town it was great, back home makes me wish they regulated more
I'm guessing the lost funds are bleeding out to energy concerns. Correct? Asides from brief construction contracts for data centers, I'm unsure how one makes such large sums of money disappear.
Anthropic told investors at the meeting that its third-quarter revenue jumped 45% to $16.8 billion from the previous quarter, and that it was profitable on an adjusted basis during the quarter.
Anyone know what this means? True or some more shenanigans?
Usually companies will publish an "earnings before interest, taxes, depreciation, and amortization" or EBITDA as a way to show whether their core business is actually profitable (even if shareholders aren't the ones to get the benefit of that business's operating profit). But that's not a standardized metric, and Generally Accepted Accounting Practices (GAAP) won't define earnings in that way.
The big problem is that the more a company has to deal with high cost of borrowing and huge capital expenditures, the more disconnected EBITDA is from actual GAAP earnings. I can build a hotel for a billion dollars and rent out a bunch of rooms, while each room transaction brings in way more money than it costs for me to have a guy at the front desk give them a key and housekeeping to clean their room, I need to actually make a lot more in order to cover the cost of all the furniture I put in the room, the construction costs on the building, and the interest payments to the bank that financed the whole thing. And if the furniture and the building won't actually last long enough to where I made all my money back before I have to reinvest a bunch of money into renovations and new furniture, then I don't have a profitable business. Or if it can be profitable on those numbers but I can't cover the interest payments to the bank, then the bank will eventually take it over and I'll lose the business entirely, so owning stock in that company isn't worth all that much.
So if they're saying they're barely profitable when you take out the very expensive costs of training new models, entering big expensive deals with data center companies to build out infrastructure, the stock they keep issuing to employees and partners, etc., that might keep things going a bit longer, but it's still not actually a path to true GAAP profitability.
It's technically true but not the whole truth. Those calculations leave out certain costs such as the value of the stocks given to employees, the cost of training new models and revenue shared with partners like Amazon. It's not that uncommon practice with technology companies when wanting to show inverstors a cleaner picture of the core business but it does not mean that the company as whole has become profitable.
thats about how much money Meta spent on trying to make the Metaverse work over a 4ish year period. Thats an insane amount of money for one company to spend in half a year. Like a mind-boggling amount of money. Even if you've made $16B in the same time frame, you're still 5 times deeper in the hole
When you want to distract the voters from something you are about to do that is massively unpopular, simply point them to a target. Chinese, communist, emigrants, you choose. You have the media machines to do it. Easy.
We will observe a substantial rise in pricing at some point; afterward, cost-effectiveness will decline, and companies will scramble to put butts back into seats.
That article reports that Anthropic told investors it expected one profitable quarter in 2026. It does not claim the company as a whole had become profitable.
Anthropic has raised large amounts of money by giving investors special instruments that can later turn into stock. When the company’s overall value rises a lot, the value of those instruments also rises. Accounting rules require the company to record that increase as a loss on its books, even though no cash actually went out the door to anyone. This explains about 72 of those 80 billion. It's not actual money spent.
Anthropic logged net loss of over $80 billion in first half of 2026, WSJ reports
October 20, 2026 3:41 PM EDT
Updated 51 mins ago
Oct 20 (Reuters) - Anthropic reported a net loss of more than $80 billion in the first half of the year and signaled it could spend as much as $750 billion in the coming years to secure more computing capacity, the Wall Street Journal reported on Tuesday.
The loss, up from $13 billion a year earlier, was driven by $72 billion in accounting charges tied to changes in the value of financial instruments Anthropic issued to raise money, the report said, citing a draft prospectus for its IPO.
The Reuters Daily Briefing newsletter provides all the news you need to start your day. Sign up here.
The figures show the hefty costs of the AI race even as investors look harder at whether the money companies are pouring into chips and data centers will pay off.
Anthropic's revenue jumped to $16.3 billion in the six months ended June 30 from $1.1 billion a year ago, while AI infrastructure spending nearly quadrupled to almost $9 billion, the Journal said.
Anthropic did not respond to a request for comment and Reuters could not immediately verify the report.
ANTHROPIC'S IPO VALUATION UNDER SCRUTINY
Reuters reported first in September that Anthropic posted a net loss of $42 billion in 2025, as part of a series of exclusives from its IPO prospectus.
The company has benefited from the popularity of Claude Code, its coding assistant that developers use to write software and build apps.
But some current investors and prospective investors who attended Anthropic's investor day last week are weighing a lower valuation for the AI company after its latest financial disclosures, four people familiar with the matter had told Reuters.
The investors who attended the event in San Francisco discussed a valuation of around $1.5 trillion, three of the people had said, a significant step down from the roughly $2 trillion valuation previously expected for the November share sale.
Anthropic told investors at the meeting that its third-quarter revenue jumped 45% to $16.8 billion from the previous quarter, and that it was profitable on an adjusted basis during the quarter.
A cooler reception for Anthropic's IPO could weigh on the wider AI market, whose near-term prospects investors see as hinging largely on the share sale's success, ahead of rival OpenAI's potential listing next year.
Reporting by Aditya Soni and Jaspreet Singh in Bengaluru; Editing by Sahal Muhammed
The full article isn't very long like a handful of paragraphs. TL;DR they lost $72 billion in Q3 despite revenue going up a lot. Same year same quarter was just $13 billion in net loss
Can you elaborate on that a bit more, I figured thats what this was trying to say
The loss, up from $13 billion a year earlier, was driven by $72 billion in accounting charges tied to changes in the value of financial instruments Anthropic issued to raise money, the report said, citing a draft prospectus for its IPO.
Pre-IPO investors fund the company but it isnt like they give all the money upfront. They do it in pieces and can increase or decrease as targets are reached, failed, or pivoted. This keeps the investors less likely to lose money. As debts and losses are shown the value of the company dropped and how much money is on tap for the future decreases. In this case 72B of funding was "lost". This is paper money because they only had it on paper they didnt have the gold in hand.
Essentially their potential value to their current investors dropped 72B.
The reported net loss is real as they used more money than they made.
Neither factor is going to bankrupt or close the company considering they still have a bunch on hand and their investors still seem to think theres a possibility of profit (just less than before).
It's not 72 billion dollars lost related to the business.
The money didn't exist and wasn't spent.
It's not clear what the instrument was exactly (I would need to look it up myself, not an accountant). But there are a range of charges that aren't actually related to profitability of a business. Value of instruments could be tied to a capital raise earlier. EG if they promised a percent of the company as part of a raise, and Anthropic is worth a lot more now. Their liability increases (the percent they owe is worth more now) and they book as a loss.
The paper loss is still critically important though.
"Oh don't worry about that 64 billion. That's just something or other the accountants dreamed up. You know what those buzz-harshing assholes are like."
It's not as important as you might think. Counter intuitively it's tied to success as a risky start up. Or arguably they should have got better terms during one of their raises - but hindsight and all that.
For the record I wouldn't by them for 1.5 trillion dollars.
Imagine being that deep in a hole and then trying to get an IPO invested. Thing is, it's all about investment. The corporate world has thrown themselves at AI to be the next best thing to cut labor and make THEIR numbers look better, so they're going to hold it up as long as they can now.
Anthropic posted a 45% revenue gain between Q2 and Q3, from $11B to $16B. If they can keep growing at that rate, the annual losses are eclipsed in short order.
Only question is whether they can actually do that or they're headed for a soft ceiling. The $80B in loses is a product of their enormous capex budget, which won't be an issue forever.
Full S-1 on Q3 losses and training costs needs to get published. Quarterly revenue growth rate was half of Q2's. Not good if it keeps halving, while their committed compute costs rising exponentially.
Key figures have yet to be leaked on S-1, which should be released fully soon. Structural losses and cash burn may be much higher than hopes for growth.
No.... The worst year Amazon ever had was a net loss of $1.4 Billion back in 2000. Most years in the early days they were at a loss of a few dozen million. At the time, they drew a lot of criticism for running red way longer. If Amazon ran at their worst year for 50 years, they still would have lost less money than Anthropic did.
In the long run you'd imagine that GPUs will become more affordable once these massive datacenters have been built. At that point it becomes far more profitable to run your own hardware. Especially if these AI companies need to increase their subscription prices by x10 to finance all this debt they've taken on.
Yeah AI is here to stay but these "founder" companies sure aren't.
Don't worry, they are already talking loudly about how much more "dangerous" those on-premise models are and suggesting that regulation is needed to make sure people only rent it from trusted facilities that can manage it.
They make it so you cannot run some models yourself without government approval, based on some criteria. For a random citizen, this would be harder than trying to stop piracy, and thus those willing to ignore the technical law may continue however they like, but it would be enough to block businesses from doing so. Kind of like how some software has a license that is a trap for businesses to have to pay if employees end up using it, but is widely popular among consumers that can get and use it freely. So the home market can do whatever if they really want, but the businesses and products those businesses launch can't ignore that sort of regulation so freely.
Another path is if they really get buy in that this is dangerous for mere peons to have: regulate hardware available for consumer purchase. Like no VRAM above 32GB, and nothing with the compute horsepower to go more than say 100 tok/s with a 32B model. Effectively setting a forever cap on home usage to current top-end 'consumer' and putting regulatory restrictions on beefier stuff. Consumer compute becomes just about improving power and cooling and form factor, no more expanding the capabilities beyond current top end.
It will definitely be profitable for the hardware makers. Companies whose entire value proposition depends on the model weights not leaking in an increasingly sloppy vulnerability-filled world of IT? Not so much.
Assuming your assessment of AI utility is correct for sake of discussion, it would not mean it is guaranteed to be profitable for any given company with any given strategy and spending.
Internet shopping and delivery is hugely profitable, but that's of little comfort to pets.com and webvan.
If a company more responsibly managed their spend and commitments and still reached the same utility as Anthropic, then Anthropic would never see profitability.
332 replies
Goddamm... This could easily have housed every man, woman, and child in this country. But no, no free housing while we squander wealth with our stupid ideas.
And that's without paying royalties to content creators as they should be required to do
also before court cases for defamation. Two cases in Germany hold Google executives responsible for false claims (by the Google AI) of sexual crime claims against an innocent man.
Good. If you want to make AI a form of communication then that's the voice of your company. You're endorsing whatever it says.
yeah. It should be the same with Facebook. Here in Australia, it seems like nothing is against the community standards anymore (not spam, porn or even violent threats)
Trash app, it's why everyone should just stop using it. I ceased doing so, 6 years ago and it's been bliss without it.
or counting the $1.5 trillion in debt the big AI players are hiding in subsidiaries
yeah the circle-jerk nvidia has created with coreweave and so many others just to wreck the gpu and memory markets, all so these hideously expensive AI racks can SIT IN THEIR FUCKING PACKAGING ON A LOADING DOCK because they don't have power or cooling built is mindboggling
ITS THE FUTURE YOU JUST DON'T UNDERSTAND!!!!!
the future of aging like fine piss lol
"The court finds Anthropic guilty and orders them to relinquish the whole of their company to the Group Representing All Artists".
Artists : "Wait -".
Dario Amodei : "Fuck yes! Good luck paying back those 80 billions, losers!"
Billion is officially the new million.
Why haven't they just asked Claude how to make money with their business model? /s
They have.
Wait for excited CEOs to fire all their staff.
Crank up the prices.
2 only works if regulatory capture in place to prevent cheaper competition from open models.
That's why they keep trying to play the "AI is so dangerous" narrative and why you keep hearing about "rogue AI agents" which in reality someone had to instruct to break the law, and they probably used some of their no-guardrails internal-use models (so Claude Mythos rather than Fable).
Claude Mythos and Fable are the exact same model, the branding only reflects which one has more restrictive classifiers in front of it during deployment. But the weights are identical.
All those loans are getting hot. They're starting to toss them around and trying to hide them.
I have strong doubts they can ever make this particular approach to AI work at all, let alone in time to cover their bets.
This path looks like a dead end. But, I don't know jack about shit.
The issue is they are too expensive, it's not really a limitation of the technology, well it might be, but that's not the issue here. They are paying Nvidia Billions for their entire 2026/27 GPU runs and they can't get the customers to cover that. It's like $200 pm for their max plan. It's almost cheeper to hire a human in India.
There's a reason so many AI projects have turned out to be Actually Indians.
My monthly salary as a software developer is 80 USD
🐤
Inference is cheap and getting cheaper. Not sure about training.
I hope it pops
An entire industry built around burning cash, water, and electricity to generate questionable answers based on glorified autocorrect. It’s astounding.
Throw another few trillion and a couple dozen rain forests at it. Completely rational behavior.
glorified autocorrect and plagiarism/nonatribition.
yep. roided up eliza's
Have you actually used it recently? It’s become incredibly good. Calling it a glorified autocorrect is like calling ASML’s lithography machines a glorified polaroid camera.
Now now this is the frediverse. Your only allowed to ignore facts and hate on ai around here.
There's plenty of good reasons to hate AI but in case you missed the notice on the door your not allowed to have nuance. It's all or nothing.
lol... it's hardly limited to the fediverse bud, people are learning why the economy is fucked: AI
why can't they buy a console? AI.
OR A GPU.
OR AN SSD.
OR RAM.
OR A FUCKING SPINNING DISK HDD.
why does the food in the adverts look like nightmare fuel? AI.
why does the president think he's popular? AI.
Tim Sweeney castigates Kenny - why? AI.
Musk can't synthesize a personality so he snorts ketamine and AI.
where's the upside, ASSHOLES? Your garbage is fucking up the entire planet, great fuckin work FANBOYS, shove that up your ass for NUANCE
In 20 years - if there's still a civilization - they're all gonna look back and wonder how so many were so fuckin dumb.
In 20 years we ALWAYS look back and wonder how so many were so fucking dumb. Thats literally history. Theres never a point where there isn't some major fucking bunch of absolute idiots trying to ruin everything.
Ai is just this decades attempt at ruining everything.
ok, so it's like asbestos and nuclear waste runoff, that's not enough reason to hate AI for you?
instead of stopping the stupidity you're like: find the nuance.
[email protected] - lol doubt [email protected] -
listen kids, if you're not brave enough to let chatgpt give you a riposte don't fucking participate ya little shits
That's entirely enough reason to hate it.
But it's also objectively fucking stupid to act as if this is some new historic event in human history and not just yet another dumb fuck move in a long line of dumb fuck moves.
Plenty of good reasons to be pissed off. You don't have to lie and bullshit to make up more is what I'm getting at.
bUt It CoUlD cUrE cAnCeR
alphafold actually COULD have and still got dissolved https://www.aimodeling.com/en/news/slug/deepmind-dissolves-alphafold-team-anthropic-pivot
Not with an LLM you won't.
obligatory...
https://m.youtube.com/watch?v=9JwpQHbMS9c
You act as though those billions of dollars spent towards other things (like cancer research) would somehow be better.
?
Fucking yes it would be better to put hundreds of billions of dollars every year into curing cancer/HIV/ Type 1 diabetes/alzheimers, literally any disease. Curing a major illness like those would actively help hundreds of thousands or millions of people around the world every year.
What does AI help? My fucking manager in sending cryptic emails that raise more questions than answers?
Dump a few hundred billion into disease research, or fusion power, or just into housing for all of us.
AI helped me with my voting ballot. I asked Kagi's AI to research the ambiguous candidates and propositions. I had it put them into a political alignment chart - left/centrist/rightwing. Addendums for candidates included things like whether they had criminal convictions or corruption, stance on vaccinations, how they feel about "tough on crime", the homeless, ect.
For propositions, the political alignments had a fourth entry added, where the yes/no included the wealth level of backers for a position.
This is very useful, completing in a hour what would have taken a week of research. Very useful for digging into the mechanical details of a proposition, like whether it prevented the replacement or supplementing of a measure.
How do you know that anything it displayed was accurate for those candidates? Every single time I've asked it to include source links, those source links don't actually support the AI's statement.
My experience too. The source never backs up what the AI claims. It always just makes shit up.
One of the candidates was appointed to their office by a Republican governor, and elevated more recently by a Democratic one. The AI referred to official government records.
Well, that's one candidate. And a reference to an official gov doc doesn't meant the info was extracted or summarized accurately.
How about the others?
...yes?
Remember to never miss the /s
Marking your sarcasm is the coward's way out
With a .ml account you need to. I've heard way more absurd statements said with a straight face from .ml users
It's cowardly to be understood?
Okay, boss. 👍
But have you considered that it's like REALLY good at programming? /s
What's stupid is, you could even just keep just the programming part, and it would become much more efficient and not ramble on with explanations etc.
If an LLM was something you could train on your GitHub repo or an internal knowledge base, and ask it natural language questions about that, AND it didn't need an 800 acre datacenter to do it, I wouldn't have so many hang-ups about it. Or maybe even train a model in a programming language and have a model that's tailor made for writing C, or LISP, or Brainfuck or something.
But that's not what they are. They only function if you hoover up everything the light touches and feed it to the machine, to the point where rare, irreplaceable literature is being shredded apart for training data. And god help you if you try to take your ball and go play somewhere else as the AI boosters profess you must do, they will follow you there and scrape you anyway
You are describing local models and they're getting pretty damn good, especially at coding.
I mean, local models with RAG exist.
$8B real loss on $16B revenue. ($72B is a bondholder gain (accounting loss) from conversion). $9B in training costs not counted as loss, even though it is worthless whenever next model version is released.
$561B in forced spending on compute over next 10 years, with $750B total compute costs lined up. Structural losses are baked in for a few years.
But, when the word guessing program gains consciousness, it will be capable of doing the job of every worker in the world, and eager to do those jobs, and anybody who owns Anthropic shares will be a trillionaire in an economy that totally works and is healthy. So, better invest now!
Musk giving 1 share (about $160) to all US children with a Trump account, will make the next generation cheer for skynet genocide and oppression. "For a while, it added 10-20% to my net worth."
For most version releases I actually doubt that, I'm pretty sure they just keep training the model and release checkpoints every few months. Architectural changes, however, will indeed invalidate the previously trained models, and occasionally they have to do that to keep up with the competition.
But if you do architectural changes even just once a year, that means you'd have to depreciate those training costs over just that one year to be truthful. Meaning they might as well be losses rather than depreciating capital investments.
There is a genuine "training corpus" that is an accumulated asset and grows and feeds new models. The vast majority of training cost, that should be expensed, is the training time compute. Neither of these are expensed. There was about 13 major US model releases in September. At least 7 of these replaced a 1 month old previous model family. Open AI gpt sol went from 5.6 to 6 to 6.1 in about 2 months.
Fast model releases not only lock in structural losses, they juice revenue growth illusions. Model use rate generally drops by 66% 4 weeks after release, because of competition, and a surge of use on release for evaluation.
New models are trained on older models, so even if the older model is not longer at par, without it the new model couldn't exists
I'm totally spamming this across the thread with copy/vaste.
Look man I just need a little liquiditiy... 😏 😈
Stop_trying_to_make_AI_happen.jpg
This IPO is going to be one of the biggest rug pulls in history.
Yes. If you follow the money, investors in OpenAI are the usual suspects, a Kushner in there (admittedly the "best" kushner) and the early investors were guaranteed returns of up to 18% (bank CD currently around 5.3%) which stemmed from a $500B injection from the US government. Not only will it be a rug pull, but a government built one. Truly appaling, we let the fox in the henhouse.
Cant have Healthcare for our people, but can pump and dump the world economy.
So far
This right here is how all of this will come crashing down. Anthropic and OpenAI, along with several associated companies, are using accounting tricks to hide their debts and obligations. These two are trying to maintain hype for investors so they can cash in on an IPO with total disregard for long-term fiscal sustainability. At some point the bill will come due, and you can bet the C-Suite will have already made out like bandits.
Don't forget, though, that there's another cashflow that they're banking on: total surveillance.
AI can't do a lot of things, but the thing it's most suited for is creating a society run like a prison to try and keep the discontented masses in check.
Not enough people realize this. They think the component shortage is a side effect, but it’s to push you towards cloud services on underpowered devices that rely on it.
Imagine the censorship when every service is monitored and controlled by them. Imagine the market capture when they can see everyone’s financials and trade secrets. When they can cut off journalists and activists with one click.
Haven't heard anything new lately, but there is a push to eliminate physical currency for, what I presume, is a similar reason -> make everything trackable.
People have been trying to eliminate physical currency for decades in not longer. That shit predates computers.
This is quite obvious, it hasn't happened at a global scale yet because people have to agree on which currency it would be
I think this will backfire too. I think they way underestimate the amount of people already using "underpowered devices" that will just continue to do so without the new cloud services shit.
Basically everyone I know us using a 5-10+ year old laptop, for example. And they just use it fine, sure it's pokey, and it's not everyone, but it's a lot.
The Italian government has just passed an emergency decree that authorizes AI-powered face recognition in public spaces without a judge's authorization.
I'm not even sure it's great at that. Is this not a function of any industry struggling to find a usecase eventually ending up with the largest US employer and cash cow: the DoD?
LLMs ('AI'), not so great, but it can get (probabilistic, but fascists don't care) descriptions of video feeds into easily searched text. Gobs of GPUs in datacenters paired with surveillance devices all over the place however presents a greater threat, where there's an (evil) will, a way will be forthcoming. Nice of Ukraine to point out how fragile they are. There's a deathstar analogy there somewhere.
Yeah, the type of people pushing this don't even care about the inaccuracy either. If it "catches a black man" with facial recognition, even if it's the wrong person it won't matter "they are all criminal thugs anyway." See also the ICE bull shit and basically deporting anyone even remotely Mexican looking, citizen or not.
Ai is a panopticon
This is my take as well.
Modern companies are only profitable in late stage capitalism if they fall into one of they two categories: scams or gambling.
No, this is pretty normal, and more importantly, doesn't actually cost the company cash. They gave lenders bonds that can be converted to stock, so this loss on their books is pure paper loss, the same way wealth in owning stock in this money losing company is pure paper wealth.
These numbers are just illustration for simplicity, but imagine a company borrowing $20 billion and telling the lender that you'll either pay back the $20 billion plus interest, or they can opt to convert it into stock worth 10% of the company. When the company is worth $100 billion, the bondholders won't convert to stock. But if the value of the company shoots up on paper to where that same stock would be worth $100 billion, the lenders would rather take the stock than the cash repayment. And because it's worth $80 billion more, the company will have to record that on the books as replacing $20 billion in bonds for $100 billion in stock, and an $80 billion loss compared to their last statement.
But look at the underlying financial cash flow. When that happens, the company actually takes $20 billion in debt off their books. They don't have to repay that loan anymore. So it improves their cash flow position at exactly the same time that they're booking a loss.
The whole thing will come crashing down for totally different reasons, relating to operational cash flow, not the actual accounting shenanigans where one kind of paper number gets converted to another kind of paper number.
While you explained this correctly (as far as I understand it), there is part of this everyone keeps making an assumption on the outcome, the "or they can opt to convert it into stock" part. I'm really curious what the interest rate they gave the early investors to entice them when the company was much further from a sure thing. If the interest rate was decent, that final number is going to start becoming awfully enticing to just collect on instead of getting stock that might collapse.
That's the tradeoff, of course. If a company liquidates, it pays its debts first, then its preferred stockholders, then its common stockholders. But each step down the line gives a greater upside in exchange for less certainty in getting paid at all. Bondholders get paid at higher priority, but have the exact amount they get paid defined by a specific mathematical formula. So they won't get any additional benefit from the company becoming super valuable. Or they can get in the back of the line to see what unlimited upside they have from the company's value going up a lot.
Also, Anthropic's bonds aren't exactly low interest. They're right there at the edge between junk bond and investment grade.
It's not entirely transparent though. It's a paper loss in terms of cash but they did have to give the stock. At their crazy valuations, even 1% is a big deal.
I dunno, isn't this the same scheme that the ride share companies and food delivery services?
With the difference that all those services mostly outsource their asset costs to their drivers or have very little investment there. Unlike anthropic who have billions in costs related to all the compute they need to service their customers (obligations to other companies, datacenters and, not to forget, ever rising interest on all their debt).
Definitely, the cost of doing business for these AI companies is orders of magnitude higher than Uber or Lyft
You mean the ones who had to massively pivot their entire business model, because they were relying on self-driving cars being practical in 2019?
The ai crash of 2026 can’t come soon and fast enough
The RAM cartel is figuring out a new excuse for a price hike before allowing it.
I'm doing my part by constantly asking chatgpt questions without ever logging in or paying them money. Eventually they'll all go belly up and maybe we can ditch this shit.
80 billion could have bought a whole lot of Chinese EV's to resale and they could have made a profit.
It is obvious to me that AI computing is a COMMODITY, in that it will inevitably be something that can not be effectively gate-kept.
These two major US companies (OpenAI and Anthropic) are playing the game of “locking in” a monopoly in a product category that has yet to mature. This is a pretty old capitalist game, of course… Uber, Microsoft, facebook,and many others have played this game in the past.
But, unlike OSes or word processing file compatibility I don’t see a “network effect” in that an open source/weight model can/will do 99% of what a proprietary model can, will do it “transparently,” will eventually become so optimized that it can run locally, etc. etc.
I mean: maybe its name recondition that is worth “trillions” but a race to be “Xerox” or “Kleenex” or even “PC” (as in IBM PC) did not really work out well for these companies.
I think that even the established players (MS, Google, etc.) are biding their time, making sure that they don’t get locked out while non-proprietary solutions become feasible.
I means I know that much of the issues have to do with the limited availability of hardware, but locking down supply is not sustainable for the “big 2” and sooner or later more adequate memory and GPU supply will enter the market…
It will never reach the scale of use they claim unless it becomes a commodity.
Commodities, a famously low-margin game. I wonder what the new companies that pick over the corpses of Anthropic and OpenAI will be like!
And if it does, it won't ever cost as much as they need it to, to justify their valuations
I dont think a local model will do 99% of what the big ones can, and even if it could, it would require an enormous amount of memory and cpu to do it.
People pay for a good user interface and convenience, that's well known. Netflix and Spotify are built on that idea and people starting using them even over free pirated stuff :)
Qwen3.8-flash-next broke the vram requirement with its new architecture (which is a preview of their 4.0 architecture). You can run a 177b parameter model on 12gb VRAM, 64gb ram and 80gb SSD at 30tok/s and the speed scales almost linearly with increases to vram.
Incidentally, this was proven at almost the exact time all the calls for "regulating local models because they'll spread into the toasters" started.
You absolutely can have local models that run on consumer grade hardware, and the proprietary companies know it and are scared to death.
Interesting.
What an excellent time to have 8gb VRAM and 32gb RAM with the current pricing -.-
Are these really able to be as good as chat gpt? Because every time I've tried a model like that, it has not been able to be nearly as good. I have 48 gb memory on my machine so I can't run this particular one though.
I would buy more ram and a whole new computer if the market wasn't insane.
It is good enough to be useful for programming. It can do clearly defined tasks like find the root cause of a bug. Using it for writing code is frustrating because it is slow and produces lots of tokens. Meanwhile Opus 5.5 is good enough that you can pretty much rely on it doing exactly what you say (and not what you meant), faster, more context efficient and you get a ton of usage with the cheapest subscription.
I can see this changing, though. If the open-weight models reach the level of Claude 5.5 and Anthropic stops subsidizing they might be on pretty equal footing. I don't see how further improving the model would help, except maybe if somebody invents a different way to manage context that remembers all the important details over a long time so it actually "learns".
i would not recommend using the SSD to run local LLMS,its slow
best: VRAM only
good: VRAM + RAM
bad: Swapping
I don't think they're insinuating running inferences from the SSD. But you need somewhere to store the weights for the GPU to load.
i am trying to say: you need the SSD to store it but not swap to it due to its slowness
I won't pretend to be an expert but my understanding is that the architecture loads experts from the SSD on demand with the RAM serving as a hot cache and the VRAM running the active experts.
As I said above I was able to run 70tok/s on my 4090w64GB ram and a samsung pcie4.0 ssd.
I haven't run the benchmarks myself, but I can say the quality of response I was getting seemed on par with the frontier models I use for work (Claude opus/fable) and the published benchmarks for the model compete with those same models.
It is a q3 quantization so there is some small loss of accuracy, but from what I've understood those are modest losses in the worst case.
There are models that don't need to consume the entire worlds data infinitively. We've had then before llms, things like the pi ai camera can do ai processing on image with something like 8mb of on board storage.
The problem is anthropic, open ai, who ever are trying to do the stupidist approach and assume they can brute force money and power at the problem and don't understand that they need exponentially infinite data in a way that is unobtainable.
In an ideal world there would be a million little small local models that so specific tasks and can be ran on consumer hardware.
Maybe not “today” but today is still early days.
I’ve been consuming lots of “build your own local AI coding agent” kind of content lately. No doubt this is the same as the guys that are welding together their own versions of Shelby Cobras and stuff… but in a few years you’ll be able to buy a “server” with enough GPU and ram to host strong open models and serve them to your little office.
I work in a small company, and if AI plays out the way Anthropic sees it’d their AI bill will be $100k every year (fire 2-3 people, pay for AI so that the remaining people do more). At that level of expenditure, why wouldn’t my poss pay as much as a copy machine $25k, lets say, go give “local AI” a chance?
The stock valuations that most tech companies carry (and certainly the 2-3 trillion that is being talked about for Anthropic) speak to potential future revenue.
To me, I see computing and actual “model” advances meeting the need within a few years… if not by entrenched companies like Nvidia, then by insurgent chipmakers or new (Chinese or otherwise) chipmakers.
In 10 years you’ll have “opus” is not better running in your phone and probably something more powerful in a PC.
There is a lot of this going on in your assumptions
A local LLM is more convenient than cloud. But it is way less efficient. The cloud LLM processes multiple people's tasks at once because generating one token is mostly loading things into GPU memory, the compute required for inference is not that big. When doing multiple at once you can load one stage onto the GPU and then do that stage for a number of tokens for essentially the same cost.
maybe they can do like 60% of the big ones?
I dont know, maybe, but I think either way, local models will be a fringe interest for technical people, and not something that will affect the income for these big tech Ai corps in any significant way.
Because you simply can't run a model with all the worlds knowledge on your home computer.
Someone said that maybe there will be small specialized models that only know about a certain thing, and sure, that's possible.
You also can't run a model with all the worlds knowledge on a supercomputer. That simply doesn't exist. To think that even the largest frontier models can do that is a mistake.
The real question is: will the median user be able to run basically everything they need locally, and how much compute would they need to do that? These companies are betting the farm that it will never happen, and that they'll be able to charge whatever they want in rent to a captive audience.
But I think these tech CEOs and ML researchers talking about "AI killing us all" is related via the aphorism: It's easier to imagine the end of the world than the end of capitalism. They can see the writing on the wall, but can't imagine a future for themselves.
Can't wait to end up holding the bag for this because they're "too big to fail" again.
So much of their revenue is government contracts. Especially with the Pentagon.
Really can't be overstated how they've wormed themselves into the Military Industrial Complex.
A lot of their revenue is also investment from hardware and software vendors that are using the revenue from the AI companies for those investments.
It's like the Enron scheme, but instead of one company inventing fake companies to move money around to increase revenue and drive up prices to hide the fact they're hemorrhaging money while the C-suite tries to divest their own money from the company, it's multiple companies working together to move money around to increase revenue and drive up prices to hide the fact they're hemorrhaging money while the major investors try to find a way to cash out.
Anthropic is anxious to do their IPO because they know their business model is fucked, and their shareholders want out.
Are they? This is productive capital being rented out at a profit to the end users. And the end users are, as often as not, big public institutions with near bottomless accounts.
I might argue they are rapidly exhausting scare natural resources. And you're welcome to do the material analysis on whether dumping 50 gal of water and 100 Mbtus of energy through a digital engine to make a "Farting Peach Says It's Voting For The Green Party in 2026!" viral Facebook ad is benefiting humanity.
But merely observing that money is changing hands between big economic players doesn't let us know if something is profitable. You're describing the basic functions of a financial economy.
That's the popular line. People said the same thing about Tesla when it went public.
It's always us normies that end up footing the bill and feeling the pain (sometimes literally to the point of dying).
Every fucking time.
Excellent use of the meme template.
"we lose money on every sale but we make it up in volume!"
Yup, because they're still losing hundreds of dollars for every dollar spent. Their economics still rely on a delusion that the hardware outlay is a fixed cost that can be recovered after model training by selling cheaper compute. Instead, models continue to be released not only to match competition, but because models still do not do a fraction of what they are promised and being sold to do. So capital costs are not fixed.
So AI hyperscalers have a lose/lose/lose/lose choice of either growing at a rate that requires unsustainable subsidies, stopping growth and admitting models can't do what they and a thousand other startups have been selling them for, charging actual costs, or any combination of the three.
If people were charged anything near the actual costs, much less with a profit margin, demand would implode fast enough to create a sonic boom.
Add to this that incremental improvement comes at exponentially more and more cost. So they spend ungodly amounts more in hopes that the next incremental improvement will push them over the magic line beyond where the are.
Some impressive feats but the issues holding it back from the 'replace the whole economy' sort of ambition they imagine are very stubborn. Notably it still confidently declares incorrect information (a significant volume of software developement works around by rigging auto-retries until it stumbles into 'correctness', and of course the people pumping out slop art/books aren't bound by 'correctness') and they still have no clear path to autonomous control of humanoid robotics (as crazy advanced as the online GenAI experience, they are still struggling to demonstrate even the most basic manual labor tasks).
"A MINOR SETBACK ON THE PATH TO THE SINGULARITY!" -
i wish them ruin, you made my day
Idk man, to me, it feels like we are further from AGI than fusion reactors actually being hooked up to the grid, considering LLMs are still just glorified autocorrects, needing a ton of prompt injections from their providers. And yet, the AI companies make it seem like it's right around the corner (obviously, because $$$), whereas to me it feels like we are still at least 50 years away. Maybe AGI will be the next fusion generator chase.
I don't think they're "glorified autocorrects", this might have been true a couple of years ago. The improvement of LLMs, month by month and sometimes even week by week, is spectacular.
Ironically there was notable news in the fusion reactor space this week as Gemini 4, building on the work of Matt Landreman and GPT-6 Astra a little while ago, expanded counterexamples to Grad's conjecture which had claimed exact solutions for controlling a key aspect of fusion were impossible.
So fusion may actually be more around the corner thanks to the advances of capabilities of AI which helped prove that the fusion problems aren't as hard as they were thought to be for decades.
And if you actually believe that, I have an amazing business proposition to you. So, there is a bridge in NY, and I can sell it to you for very little money...
Building on other people's work to provide examples for something is actually one of the things an LLM can be good for.
They did manage to do something different, though. I thought at one point that proof assistants should offer to automatically upload your proofs to a central repository so everyone can benefit from them. But such a system would require many to opt in for it to be useful.
What AI does best is giving you existing solutions. It even adapts them if they don't quite fit. Because many people now discuss their ideas with AI, ideas may spread the same way as well when the conversations are used as training data.
So no, definitely not AGI but valuable. Though as typical of explosively growing startups the value does not come from the technology but from activity bordering on criminal that lawmakers can't keep up with.
AGI has been achieved more or less because it just means self sufficient at completing tasks. An LLM is just a component these days because they are Vision-Audio-Language-Models with multi frequency 'thought' loops. The next letter is A for action where they just have high frequency outputs for direct computer and robot use.
Just because it is bad AGI doesn't mean it is not AGI right
Repeating the DotCom burst.
Please don't make the tax payer pay for it. The tax payer is gonna pay for it, right?
The rich know the poor will cushion their fall
Sucks to be US American.
Indeed. But we are also known for dragging the rest of the world down with us. So brace yourselves.
I mean, we cut basically all of our external aid style programs so it sucks for non Americans too.
Personally, I would MICH rather my tax dollar go to teaching African kids about LGBTQ+ topics than being given to some stupid AI company.
Good.
I just recently installed Strata which runs on qwen 3.8 flash next. I'm getting over 50 toks. Now a deepseek v4 version is out along with glm 5.3 flash. It's slower but useable and very capable. All this on a 5080 16GB and 128GB ddr5.
The future for me is going to be open weights and on your own hardware.
That's how it needs to be, and how it should be. Otherwise, we are all permanently screwed by big tech, subscribed and licensed to even own a computer by 2035; MicroSlop says open-source AI is good somehow, but we will see through their actions if they lock down their new line of NVIDIA Spark laptops to an approved number of models. I heard rumours that Deekseek might be banned in America; I can't see how that would work though.
Deepseek is open weights. You can just download and use it locally. I would never run deepseek using their servers. Or z.ai servers. That I agree with China is still pretty bad.
While I have no problem with most of China's principles regarding business. I don't know about all their politics and laws, nor have I given them much thought, but they deserve more respect than they get for essentially being the whole world's industrial estate and being able to pull off the amazing things they do while having virtually no copyright laws. Open-sourcing so many things we would rather make proprietary, thus very expensive for ourselves. I'm grateful to them for deepseek! Even if they want to spy on me a little, at least I know to expect it lol =)
I mean that sounds great and I'm happy for you, but you're working with hardware that costs more than most people on here's cars when it's available at all. How viable would your workflow be on an 8gb 1080ti with 32gb of ddr4?
Running locally makes little sense anyway because computing multiple replies at once costs the same as computing one. A home setup isn't going to get utilization anywhere close to optimal.
I mean, it's not really a toy. It's a business expense. I use it everyday to make money.
At no point did I say it was man. Just not something most people have access to. "Got to have money to make it" painfully true as usual. I have $234 in my wallet, a late 90s Honda to crash in, and sheer force of will to try and change that (it's not working lol). The 32GB/1080ti example is the nicest computer any of my friends own, my own rig is a scavenged parts pile centered around a 10th gen i7 that will be complete when some ram falls into my lap and right now I'm posting from a phone that was $150 6 years ago.
You know, that's fair. Most people do not need a rig like mine. I would still be using my 8 year old computer if my new clients data wasn't so sensitive that I have to run local inference.
As someone who was also broke and lived in an old Honda, I know what you're going through and I've been there. If you truly want to change your outcome, you gotta find a job that has high growth potential and a low barrier for entry. Personally I worked at a bank for almost a decade. Started as a teller(low barrier,no college needed), ended as a regional manager(high growth). I hated every minute of it. But it gave me a chance to save up and find a career I enjoy.
Good luck my dude. You will be successful. I believe in you.
Nice setup. I'm running Qwen3.8 27B at 17GB with a 32k context window on a 3090. It does most of what I need for everyday use and its all local, private, no internet connection needed.
The gap between running an LLM locally and cloud API is closing fast.
Yeah but getting your own hardware is getting expensive too.
Only because there is artificial scarcity in play
There's plenty of actual scarcity to go around too
I mean, besides oil prices, the only reason why hardware is more expensive than it was before is because people are hoarding it, fearing that it won't be available in the future. But there's no reason why new and better hardware shouldn't be engineered and made available unless there is another world war.
The electronic hardware cartel has stopped designing new hardware, but it was a deliberate choice to inflate prices. They even backpedaled on 5090s
How do you run qwen 3.8 flash next on 16gb vram, if you don't mind me asking ?
At 3 tokens a second and calling it useble I would guess?
I see people claiming that shit all the time and it's baffling.
They said 50- i looked up strata, and it uses smaller quantizations/optimizations of that model depending on your system.
I have 48g vram and 192g system ram and get ~15t/s on qwen 3.8 flash next at Q4. Off the top of my head. I think Q6 is 12.
I started looking into it finally after making that comment and came across strata come back here and everyone is saying strata. I guess im trying strata?
https://sh.itjust.works/comment/28041846
Try it out and let me know if you can get higher.
https://sh.itjust.works/comment/28041846
I didn't believed it until I started using it.
need 64gb of system ram. qwen next architecture uses smaller experts than others, and it can keep up with ram swap. I believe it uses a file on disk (ngram) to predict what experts will be needed.
https://github.com/niko1221/strata
The model I'm using is a Q4. Unsloth UD-IQ4_XS
Running on your hardware is always going to be less efficient than running in a data center. You can run small models on green data centers - best of both worlds (unless you need privacy then run local).
Ever since I switched to Cortecs I'm spending like 20 cents / month for my LLM usage cuz I mostly use small models and try using them responsibly.
That’s why you do it at work on a secured laptop you own that’s on their network. Modern problems. They pay electricity and all the jazzzzzzzz.
That's a lot of money that could have been spent on healthcare, food security, housing, transit, climate change problems...
That's not how that works… the money "lost" isn't going down the drain. It is… going somewhere, doing something useful. That's why it is a loss to the company.
If it's going to make LLMs it's not as useful as the other things I listed. Paying a bunch of software engineers $300k each to work on LLM nonsense instead of something useful is a shame. Sure, the money isn't set on fire, but the opportunity cost is large.
I mean... A lot of the money is being spent on compute, which is just consuming electricity, which is many times made from coal or gas. So a lot literally is on fire.
Your confusing fact with opinion.
Factually yes it's useful to the company.
Your opinion is that the usefulness doesn't justify it and this isn't useful.
The real question is, is it useful enough. Which is likely no.
That's at least 3 "your"s in this thread when it should be "you're". If you're wondering why you're getting down voted....
I'm not wondering nor do I see the point of downvotes? I will never understand the weird fixation with down votes.
Also I don't even see these downvotes your talking about in the first place. My post appears to have no downvotes and one upvote to me.
So either something is broken here or your bullshiting for a reason I don't understand.
MORE! </ADAM DRIVER>
See here's the thing: it isn't really their money, so they don't give a shit. If it was, things would be... ✨different✨
this goes in line with thier aggresive pandering to the govt to be used mass surveillance.
So dept isn't cash cost that's tied to operations? All the debt did was finance the operations. Reading this paragraph and you'd wonder: who in the hell issued those bonds in the first place?
It's a very common way for startups to raise money. In this case, the bond value goes up because of a perceived increase in the value of Anthropic.
Repaying debt is cash cost. But this is convertible debt that could be converted to stock, in which case the holders of those newly converted shares don't get paid cash from the company, but instead own a chunk of the company worth more than that cash. So the conversion itself would alleviate the cash demands of the company, while also diluting the ownership stakes of any preexisting owner. That's what that paragraph means.
Companies issue convertible bonds usually because they can get better interest rates on it, in the event that they need to repay it. Or they're happy to deal with diluting their shares if the company becomes so valuable, because that means they've gotten a lot richer and just have to share their newfound riches with the lender.
I'm totally spamming this across the thread with copy/vaste.
Look man I just need a little liquiditiy... 😏 😈
A billion here, a billion there. Over some time it might become quite a sum.
No wonder they're panicking so hard over open weight Chinese models
Sound investment. Sounds like the echo down a dark tunnel.
C’mon. Implode already.
They're betting everything on AGI. I don't think anyone there at any point has even hoped that the tiny initial revenue streams would cover the absolutely gigantic upfront costs of setting everything up.
Now, whether they'll actually reach AGI before this thing implodes is anyone's guess. The people who claim that they absolutely will are full of shit just like the ones claiming they absolutely will not. Nobody knows and only time will tell.
I'm just going to throw in my 2 cents on the AGI issue since I have a better understanding than average. I'm currently a computational neuroscience PhD student. I don't work in theoretical neuroscience currently but that's my hope after I finish my doctorate.
I don't believe current transformer architectures are capable of AGI and all big AI companies just seem to be adding more parameters to established transformer architectures.
::: spoiler Tap for spoiler From what I understand transformer architectures act similarly to primary sensory cortices, they model complex probability densities. These companies appear to be banking on the idea that adding more training data, having smaller networks working together with different expertise, and having the train of thought (phonological loop) modeled as part of the PDF is enough to reach AGI. I disagree.
While all of those things are interesting solutions to parts of the problem I don't think they address some of the bigger issues.
That's perfectly reasonable. I mostly take issue with the people claiming absolute certainty either way.
Haven’t there been developments in biological computers recently? Wouldn’t that be the key to AGI? I am a total idiot so I know zero about what you are doing or talking about. It just makes more sense for a computer to think like us if it has the same hardware as us is how I see it.
There are positives and negatives to using cells for computing. There are very complex systems that are responsible for making sure cells grow in the correct architecture and are supported property via blood vessels and glial cells. There is also definitely something to be said about glial cells themselves having a role in computation (whole other topic I won't get into) not just neurons.
If were just looking at the input output relationship with neurons we can model it pretty well using neuromorphic hardware which is designed to respond like a neuron. There are still gaps but and there are scientists who believe it isn't close enough.
Our current hardware can model neurons closely as well but its less efficient at it. The current AI models use point neurons which takes a lot less computing power but I don't think they're the way to AGI. Personally I'm a fan of the adaptive exponential leaky integrate and fire model or the spike response model
This is really interesting. Do you know of anywhere that I could read more about comparing neuronal function to transformer architecture?
Good question! Unfortunately, I haven't found a good source myself. I've been trying to piece it together based on my theoretical/mathematical neuroscience background and self-study about transformer architectures. The problem is the point neurons in transformers don't map to neurons in cortex. You have to instead look at the bigger functions that are being computed by each system.
The best single source to learn more about theoretical neuroscience is Neuronal Dynamics: From Single Neurons to Networks and Models of cognition. But its super math heavy. If you want to easier to digest videos I recommend Artem Kirsanov on youtube. For transformer architecture 3 Blue 1 Brown has a great series of videos about it on YouTube as well.
I’m not afraid of a little math so I’ll give the book a try, thanks for the recommendation!
I kinda feel like actual AI will not be made by humans, but by a model that is able to reprogram and upgrade itself until it has some sort of agency. That said, my experience in the field is from the 90s so I really know fuck all that's relevant today.
Generally the benchmark for AGI has been "human level intelligence" meaning that from there on AI is also the best AI researcher and human input is no longer needed in developing it. That's where the concept of singularity or intelligence explosion stems from.
The narrative here tends to be that it goes from AGI to ASI (Artificial Super Intelligence) very rapidly but in my view there's not a significant difference between the two. AI that is as intelligent as a human but not more would still process information about a million times faster so effectively it would still be superintelligent.
I disagree on the speed. The "human level intelligence" would be about as fast as a human on tasks. It might do them very differently, so it could be superhumanly fast at some things like reading a whole book.
The thing that I doubt is that this would create the prophecised feedback loop. Sure, the AI could be a bit better than a human at further developing itself but it is probably also really hard to increase the intelligence.
One thing that can't be denied is that AGI could be taught once and then cloned instead of all the work that goes into training humans.
That makes sense. That said, I'm not sure superintelligence is just a matter of processing speed. It's an interesting question - give me a complex problem and say a thousand years to work on it. I kinda feel there's some problems I'm just not equipped to solve, no matter how much processing time I have. But of course, that's impossible to test. And in computer terms, you could copy a whole team of me to work together. Hmm.
Yeah, I concur that the difference between AGI and ASI is likely moot. If an intelligent machine is able to upgrade itself to be more intelligent - well, I think we'd all do that if we could! I'm not sure that more intelligence and faster processing are the same thing, but I'm also not sure that we could distinguish between the two in practical terms.
This is exactly what the mainline frontier labs are trying to do.
Plus I mean what good can an AGI do for me? I'm still not interested in the slightest in paying someone to atrophy my brain.
Oh, AGI isn't for you. Its for the AI cultists to worship once they summon it. The AI cultists think they can control thier god, even thought it will be integrated into everything if it is truly super intelligent. Then they will realize that they didn't learn from all of the dystopia Sci fi warning them about summoning their god.
Either that, or its all marketing bullshit.
Superfly 😎
AGI isn't for us, it's to replace us. AGI is the pipedream they have been selling of actually reducing your workforce by offloading it all onto AGI agents. They either don't reach AGI before their debt obligations come due in 2027 causing the bubble to burst and causing the next great recession or the achieve AGI before that happens and everyone slowly loses their jobs causing the next great recession.
Lucky for us we can provide superior results for 1/1000th the cost lol.
Yeah, I think they forgot about competition or assumed they wouldn't have it by the time it matters.
It only works if they stop training new models and can actually follow their 6 year depreciation schedule on their GPUs. But that would make Jensen Huang cry, and they do have competition right now. You can't put a bunch of effort into creating one AGI model because then a few months later OpenAI or Anthropic releases a slightly better model and all your contracts go away next year.
They're not rich because they're smart, they're rich because they don't have any ethics.
Brain atrophy is only a risk if you don’t reallocate the time saved by applying AGI to an issue you would otherwise have to figure out yourself.
Use it on problems you can’t bear to manage and not on something for which you have a passion.
It could be incredible if society doesn’t collapse.
The problems I can't bear to manage are not manageable by a chat bot, lol.
Plus with the incoming climate apocalypse, I'd rather be more self reliant and not less.
I'm with you but in a professional environment, there is simply no way around AI anymore. It's too good to not use unfortunately.
So I’m a carpenter. AI has zero relevance to my job.
I am in software and I don't think it's that clear cut tbh. There's definitely a right and wrong way to use AI. I think a lot of companies are moving fast and breaking things right now, but a lot of them are going to find out that the thing they broke is the product and they won't be able to slop their way out of a problem they slopped their way into.
At a micro scale, AI is pretty good at writing workable code quickly. This obviously boosts productivity at the individual level, especially in the short term. At a macro scale, good maintainable software requires higher order thinking that transformer based models simply cannot do. So a lot of the stochastic micro decisions an AI makes lead to technical debt that is not trivial to remove, and can't really be resolved by the AI itself, at least not in a way that won't contribute to the problem.
These issues can be curbed by the programer at the point of origin, before it gets out of hand, but obviously that requires more judicious use which cuts into the productivity gains. Because writing code has never been the hard part of software.
TL;DR: fully embracing the slop machine will eventually poison your product in a way that is not really easily recoverable. It is useful, but not "too good to not use."
I'm glad I've never had to work in an office much, I suppose.
What good could hundred thousand of the world’s best scientists do for me if they had decades to study every question I asked and could answer them immediately?
Nothing I guess if you absolutely want to figure it all out by yourself.
Good thing they're working on it regardless of if I pay a subscription to a technofeudalist, lol
AGI “use” has ethical issues of its own in its own right. But if we dismiss those (we shouldn’t) they will be extremely useful.
Imagine a personal assistant that will organize your life and manage all the trivial but tedious tasks in your life. There is a reason why crazy rich people all hire personal assistants.
It's because they're lazy, stupid, and have more money than they know what to do with. I don't have any of those luxuries.
Not going to argue with you here. But your name is quite ironic. Lol.
I have no time or respect for the owner class, sorry if it came across as aggressive towards you, lol.
Rookie numbers. C’mon. Lose it all! Be #1!
yeah that's only the first half... and now people REALLY hate data centers... fuck 'em
I'm not knowledgeable enough for my opinion to have value, but I keep hearing that Chinese models are achieving similar results without the huge infrastructure investment - cos we aren't selling them the chips. If you can achieve similar results without spending hundreds of billions on datacentres - then it's possible that these companies have made an oopsie.
yeah but it's not what you think. you think the chinese are just magically making these models appear from the dirt. the reason they're able to make their models without tremendous grunt powering them is that they distill other people's models - the theft is built into the entire ecosystem, they started by stealing from writers and artists, and are now trickling down to stealing from openai and anthropic.
at some point, someone's got to build the model with inference and that takes a lot of computation to push through.
Every local food truck operator makes more than they do, and for good reason.
Got a local food truck epidemic of mid overpriced food and no food safety standards, in my college town it was great, back home makes me wish they regulated more
Cuz they got churros.
I'm guessing the lost funds are bleeding out to energy concerns. Correct? Asides from brief construction contracts for data centers, I'm unsure how one makes such large sums of money disappear.
Still supposed to be valued 1.500.000 Million dollars. Lol.
Mitch Benn perfectly captured the 2008 financial crash in 1 minute 3 seconds. Looks like it’ll be back in the charts again.
https://m.youtube.com/watch?v=JFtL6REGk08
Brilliant.
Anyone know what this means? True or some more shenanigans?
"adjusted" doing some heavy lifting here.
I mean... If you don't count most of your costs, then sure you can be "profitable (adjusted)".
Presumably they mean they're profitable if you only count operational costs. Which is funny because for an AI company, R&D isn't exactly optional.
Ed can shine some light on it, perhaps: https://youtu.be/CudIfoRiVYQ?t=715
Usually companies will publish an "earnings before interest, taxes, depreciation, and amortization" or EBITDA as a way to show whether their core business is actually profitable (even if shareholders aren't the ones to get the benefit of that business's operating profit). But that's not a standardized metric, and Generally Accepted Accounting Practices (GAAP) won't define earnings in that way.
The big problem is that the more a company has to deal with high cost of borrowing and huge capital expenditures, the more disconnected EBITDA is from actual GAAP earnings. I can build a hotel for a billion dollars and rent out a bunch of rooms, while each room transaction brings in way more money than it costs for me to have a guy at the front desk give them a key and housekeeping to clean their room, I need to actually make a lot more in order to cover the cost of all the furniture I put in the room, the construction costs on the building, and the interest payments to the bank that financed the whole thing. And if the furniture and the building won't actually last long enough to where I made all my money back before I have to reinvest a bunch of money into renovations and new furniture, then I don't have a profitable business. Or if it can be profitable on those numbers but I can't cover the interest payments to the bank, then the bank will eventually take it over and I'll lose the business entirely, so owning stock in that company isn't worth all that much.
So if they're saying they're barely profitable when you take out the very expensive costs of training new models, entering big expensive deals with data center companies to build out infrastructure, the stock they keep issuing to employees and partners, etc., that might keep things going a bit longer, but it's still not actually a path to true GAAP profitability.
It's technically true but not the whole truth. Those calculations leave out certain costs such as the value of the stocks given to employees, the cost of training new models and revenue shared with partners like Amazon. It's not that uncommon practice with technology companies when wanting to show inverstors a cleaner picture of the core business but it does not mean that the company as whole has become profitable.
thats about how much money Meta spent on trying to make the Metaverse work over a 4ish year period. Thats an insane amount of money for one company to spend in half a year. Like a mind-boggling amount of money. Even if you've made $16B in the same time frame, you're still 5 times deeper in the hole
Wonder how they're going to spin 'too big to fail' this time.
"Super intelligence is too important"
Sigh..
"But it hasn't destroyed humanity yet! You can't let it fail right before it's crowning achievement!"
yellow peril
Here it is.
When you want to distract the voters from something you are about to do that is massively unpopular, simply point them to a target. Chinese, communist, emigrants, you choose. You have the media machines to do it. Easy.
They ONLY lost eighty billion dollars??
♫ Sounds like someone is ready for a trillion dollar IPO... ♫
We will observe a substantial rise in pricing at some point; afterward, cost-effectiveness will decline, and companies will scramble to put butts back into seats.
I bet they'd love to do that, but then everyone will start using Deepseek, who are not that far behind.
I thought they became profitable?!
https://aitoolsrecap.com/Blog/anthropic-first-profit-2026-revenue-breakdown
AI companies wouldn't lie, would they?!
That article reports that Anthropic told investors it expected one profitable quarter in 2026. It does not claim the company as a whole had become profitable.
Anthropic has raised large amounts of money by giving investors special instruments that can later turn into stock. When the company’s overall value rises a lot, the value of those instruments also rises. Accounting rules require the company to record that increase as a loss on its books, even though no cash actually went out the door to anyone. This explains about 72 of those 80 billion. It's not actual money spent.
with deep accounting tricks.
If any of these AI companies IPO, you could make a lot of money shorting them!
Is there a way to access this without having to pay for/have an account?
Anthropic logged net loss of over $80 billion in first half of 2026, WSJ reports October 20, 2026 3:41 PM EDT Updated 51 mins ago
Oct 20 (Reuters) - Anthropic reported a net loss of more than $80 billion in the first half of the year and signaled it could spend as much as $750 billion in the coming years to secure more computing capacity, the Wall Street Journal reported on Tuesday.
The loss, up from $13 billion a year earlier, was driven by $72 billion in accounting charges tied to changes in the value of financial instruments Anthropic issued to raise money, the report said, citing a draft prospectus for its IPO.
The Reuters Daily Briefing newsletter provides all the news you need to start your day. Sign up here.
The figures show the hefty costs of the AI race even as investors look harder at whether the money companies are pouring into chips and data centers will pay off.
Anthropic's revenue jumped to $16.3 billion in the six months ended June 30 from $1.1 billion a year ago, while AI infrastructure spending nearly quadrupled to almost $9 billion, the Journal said.
Anthropic did not respond to a request for comment and Reuters could not immediately verify the report.
ANTHROPIC'S IPO VALUATION UNDER SCRUTINY
Reuters reported first in September that Anthropic posted a net loss of $42 billion in 2025, as part of a series of exclusives from its IPO prospectus. The company has benefited from the popularity of Claude Code, its coding assistant that developers use to write software and build apps.
But some current investors and prospective investors who attended Anthropic's investor day last week are weighing a lower valuation for the AI company after its latest financial disclosures, four people familiar with the matter had told Reuters.
The investors who attended the event in San Francisco discussed a valuation of around $1.5 trillion, three of the people had said, a significant step down from the roughly $2 trillion valuation previously expected for the November share sale.
Anthropic told investors at the meeting that its third-quarter revenue jumped 45% to $16.8 billion from the previous quarter, and that it was profitable on an adjusted basis during the quarter.
A cooler reception for Anthropic's IPO could weigh on the wider AI market, whose near-term prospects investors see as hinging largely on the share sale's success, ahead of rival OpenAI's potential listing next year.
Reporting by Aditya Soni and Jaspreet Singh in Bengaluru; Editing by Sahal Muhammed
The full article isn't very long like a handful of paragraphs. TL;DR they lost $72 billion in Q3 despite revenue going up a lot. Same year same quarter was just $13 billion in net loss
Despite? More like because. If compute is subsidized, every extra compute adds to the loss.
72 billion was a paper loss. 8 billion loss is still a lot though!
Can you elaborate on that a bit more, I figured thats what this was trying to say
But I didn't really understand it?
Pre-IPO investors fund the company but it isnt like they give all the money upfront. They do it in pieces and can increase or decrease as targets are reached, failed, or pivoted. This keeps the investors less likely to lose money. As debts and losses are shown the value of the company dropped and how much money is on tap for the future decreases. In this case 72B of funding was "lost". This is paper money because they only had it on paper they didnt have the gold in hand.
Essentially their potential value to their current investors dropped 72B.
The reported net loss is real as they used more money than they made.
Neither factor is going to bankrupt or close the company considering they still have a bunch on hand and their investors still seem to think theres a possibility of profit (just less than before).
That's my non-expert understanding anyway.
This is backwards. It's confusing because the article has two statement related to value.
The value of the company went up in the first half of the year (so the notes held by lenders go up, and company books as loss)
Separately, the article reports on the expected hopeful IPO value from investors, and that went down. That's more forward looking.
It's not 72 billion dollars lost related to the business.
The money didn't exist and wasn't spent.
It's not clear what the instrument was exactly (I would need to look it up myself, not an accountant). But there are a range of charges that aren't actually related to profitability of a business. Value of instruments could be tied to a capital raise earlier. EG if they promised a percent of the company as part of a raise, and Anthropic is worth a lot more now. Their liability increases (the percent they owe is worth more now) and they book as a loss.
The paper loss is still critically important though.
"Oh don't worry about that 64 billion. That's just something or other the accountants dreamed up. You know what those buzz-harshing assholes are like."
It's not as important as you might think. Counter intuitively it's tied to success as a risky start up. Or arguably they should have got better terms during one of their raises - but hindsight and all that.
For the record I wouldn't by them for 1.5 trillion dollars.
Its important to the people deciding whether to give them more money.
As a selling point that they can deliver incredible ROI?
Fair enough
https://periscope.corsfix.com/?https://www.reuters.com/business/anthropic-recorded-net-loss-over-80-billion-first-half-this-year-wsj-reports-2026-10-20/
Thanks for this. The wallhop link posted by another user was a shitty summary.
I tried to paste the original link to archive.ph but got an error.
Can you fill me in/link me to the site you use to get around paywalls?
I mostly use https://periscope.corsfix.com/
Since archive.ph/12ft.io fails quite often for me.
You could try putting the link https://wallhop.io/
(It doesn't work for all sites, but I've had a decent success rate.)
link works for me without paywall or login
huh, I didn't realize it had a paywall because Brave shows it just fine here...
🏴☠️
Yes, look up Ed ZItron on Youtube. He'll be talking about it by Friday at the latest if I know him.
Nice. Don't give them a penny of your money.
Didn't SpaceX just report AI profit? I wonder how creative their accounting is
Grok did the math
Grok is very creative.
Imagine being that deep in a hole and then trying to get an IPO invested. Thing is, it's all about investment. The corporate world has thrown themselves at AI to be the next best thing to cut labor and make THEIR numbers look better, so they're going to hold it up as long as they can now.
🎈📌
Oh well. Anyways....
Anthropic posted a 45% revenue gain between Q2 and Q3, from $11B to $16B. If they can keep growing at that rate, the annual losses are eclipsed in short order.
Only question is whether they can actually do that or they're headed for a soft ceiling. The $80B in loses is a product of their enormous capex budget, which won't be an issue forever.
Currently they're winning so investors don't care about the losses.
It's the theory of the growth investment. Race to first place, monopolize industry, then extract the revenue.
Everyone from
MonopolyMicrosoft to Google to Facebook to Palantir have successfully capitalized on this strategyMonopoly the game?
Yes, exactly
Full S-1 on Q3 losses and training costs needs to get published. Quarterly revenue growth rate was half of Q2's. Not good if it keeps halving, while their committed compute costs rising exponentially.
They went from $4.7B in Q1 to $11.6B in Q2 to $16.8B in Q3
I'm hard pressed to call that bad news, no matter how you want to slice it.
If they only see $3B in Q4? Yeah, I'm sure they'll be crying about it. From $4B to $19B in a year. What's a disastrous business model.
Key figures have yet to be leaked on S-1, which should be released fully soon. Structural losses and cash burn may be much higher than hopes for growth.
That is about what Taiwan just purchased in natural resources.
Here‘s for $80 billion more.
that's a lot of moneys
holy shit.
...that's a shame.
Noice!
But we're worth trillions!
So does Amazon every year, no?
No.... The worst year Amazon ever had was a net loss of $1.4 Billion back in 2000. Most years in the early days they were at a loss of a few dozen million. At the time, they drew a lot of criticism for running red way longer. If Amazon ran at their worst year for 50 years, they still would have lost less money than Anthropic did.
They haven't operated at a loss in over a decade.
Impossible to imagine anything so extremely useful as AI not being profitable in the long run. No sarcasm.
In the long run you'd imagine that GPUs will become more affordable once these massive datacenters have been built. At that point it becomes far more profitable to run your own hardware. Especially if these AI companies need to increase their subscription prices by x10 to finance all this debt they've taken on.
Yeah AI is here to stay but these "founder" companies sure aren't.
Don't worry, they are already talking loudly about how much more "dangerous" those on-premise models are and suggesting that regulation is needed to make sure people only rent it from trusted facilities that can manage it.
They are certainly trying to kill local models and personal computing. I just don't think it's possible.
I see two ways this could happen:
They make it so you cannot run some models yourself without government approval, based on some criteria. For a random citizen, this would be harder than trying to stop piracy, and thus those willing to ignore the technical law may continue however they like, but it would be enough to block businesses from doing so. Kind of like how some software has a license that is a trap for businesses to have to pay if employees end up using it, but is widely popular among consumers that can get and use it freely. So the home market can do whatever if they really want, but the businesses and products those businesses launch can't ignore that sort of regulation so freely.
Another path is if they really get buy in that this is dangerous for mere peons to have: regulate hardware available for consumer purchase. Like no VRAM above 32GB, and nothing with the compute horsepower to go more than say 100 tok/s with a 32B model. Effectively setting a forever cap on home usage to current top-end 'consumer' and putting regulatory restrictions on beefier stuff. Consumer compute becomes just about improving power and cooling and form factor, no more expanding the capabilities beyond current top end.
It will definitely be profitable for the hardware makers. Companies whose entire value proposition depends on the model weights not leaking in an increasingly sloppy vulnerability-filled world of IT? Not so much.
Public libraries are also very useful. Doesn't mean they have to make money.
Bookstores, publishers, writers, etc etc are also very useful and profitable even with the existence of public libraries.
Assuming your assessment of AI utility is correct for sake of discussion, it would not mean it is guaranteed to be profitable for any given company with any given strategy and spending.
Internet shopping and delivery is hugely profitable, but that's of little comfort to pets.com and webvan.
If a company more responsibly managed their spend and commitments and still reached the same utility as Anthropic, then Anthropic would never see profitability.