Zone Media Hell, and welcome to Better Offline. I'm your host ed Zeitron. Subscribe to the newsletter by the merchandise. It's all in the notes. And we're on the second installment of our three part Hater's Guide to the AI bubble and the cracks within the generative AI industry and how they're becoming bigger and scarier and the potential economic meltdown course by a collapse in generative AI spending. Well,
it's not really general if AI spending. It's literally just fucking GPUs, and I think it might be sooner and likelier than many think. Toward the end of the last episode, we talked about one of the inane comparisons we hear between today's nation state size spending on jen ai capital expenditures and the investments that Amazon made when scaling Amazon Web Services, which was literally the foundation of Dow Computing
at scale. I would say someone's going to email and say I'm wrong, not going to read it, and I had to cut things short because we ran out of time.
But I want to.
Continue the conversation because I think it's important to examine this comparison thoroughly, if not just to explain why it doesn't work. It's also I want to stop. I want to stop hearing it. I want when people say it to me, I just want to send them this fucking episode and say leave me alone, buddy boy.
But but bye.
But the first point I want to make in this episode is that generative AI and large language models do not resemble Amazon Web Services or the greater cloud compute boom, and generative AI is not infrastructure. Now, some people compare llms and their associated services to Amazon Web Services or services like Microsoft Zero or Google Cloud, their giant, multi billion dollar operations that basically share their server capacity with companies wanting to run stuff on the Internet or within
their own within their own systems. A very fudgy way of putting. They help make sure that applications work online. These are very very useful services, and by the way, people are wrong to make the comparison between them and
the l lms. As I'll get into now. Amazon Web Services when it launched, comprised of things like and forgive me how much I'm going to dilute this, Amazon's Elastic Compute Cloud EC two, where you rent space in Amazon service to run applications in the cloud, or Amazon Simple Storage S three, which is enterprise level storage for applications and storing things, is not just like a simple hard drive. It's redundancy, it's making sure it's copied in places, so
latency comes down tons of other things. But in simpler terms, if you were providing a cloud based service, you used Amazon to both stored a stuff that the service needed and the cloud actual cloud based processing. So of compute, so like your compute loads and runs applications, but delivered the thousands of millions of people online.
And this is a huge industry.
Amazon Web services are alone brought in WHEB revenues are over one hundred billion dollars in twenty twenty four. And while Microsoft and Google don't break out their cloud revenues, they're similarly large parts of their companies, and Microsoft is used as zero and the past to patch over shoddy growth. These services are also selling infrastructure. You aren't just paying for compute, but the ability to access storage and deliver services with low lateent so users have a snappy experiences
wherever they are in the world. And I know I just said a snappy experiences. I'm not editing it. The subtle magic of the Internet is that it works at all, and a large part of that is the cloud compute
infrastructure and oligopoly of the main cloud providers. Having such a vast data centers, this is much cheaper than doing it yourself, and to a certain point, Jobbox moved away from Amazon Web Services is it at scale, for example, but this also allows someone to take care of the maintenance of the hardware and make sure it actually gets
your stuff to your customers. You also don't have to worry about spikes and usage because these things are usage based, hence the elastic and you can always add more compute to meet demand or just have it in a particular time. There is, of course nuance, security specific features, content specific delivery services, data based services. There's nuance behind these clouds.
You're buying into the infrastructure of the infrastructure provider, and the reason these products are so profitable is that in part you are handing off the problems and responsibility to somebody else. And also, most web applications are not that
demanding of cloud compute. They might be at scale expensive to provide to millions of people, but Facebook was not a super complex, I don't know website depending on thousands or millions of GPUs, and based on the idea, there are multiple product categories you can build on top of something like edblys Because ultimately cloud services are about Amazon,
Microsoft and Google running your infrastructure for you. Large language models and their associated services are completely different, despite these companies attempting to prove otherwise. And it starts with a very very simple problem. Why did any of these companies build these giant data centers and why did they fill them full of GPUs? Amazon Web Services was created out
of necessity. Amazon's infrastructure needs were so great that it effectively had to build out the software and hardware necessary to deliver a store that sold theoretically everything, the theoretically anywhere, handling both the traffic and customers, delivering the software that runs Amazon dot Com quickly and reliably and well, making sure things kept working, making sure they were stable. And it didn't need to come up with a reason for
people to run web applications. They were already running applications client side on their computers. They realized that doing so at scale would be cool, or they were already doing so in a way that was likely not particularly cost effective. And the ways that we're doing so, they were inflexible, and they required specially skills and indeed physical infrastructure personnel. They were quite expensive. So Amazon Web Services took something that people already did and what there was actually a
proven demand for, and made it better and scaled it. Eventually, Google and Microsoft copied done because that's all they can do. And that appears to be the only similarity with generative AI. That due to the ridiculous costs of both data centers and GPUs necessary to provide these services, it's largely impossible for others to enter the market.
You know.
After that, generative AI feels more like a feature of cloud infrastructure rather than the infrastructure itself. AWS and similar medic clouds are versatile, flexible, and multi faceted. Generative AI does what generative AI does well, that's about it. You can run lots of different things in AWS. What are the different things you can run using large language models? What are the different use cases and indeed user requirements
that make this the supposed next big thing. Perhaps the argument is that generator of AI is the next AWS or similar cloud service because you can build the next great companies on the infrastructure of others. The models of say open AI and anthropic and the service of Microsoft. Okay, okay, let's humor this point too. You can build the next great AI startup, and you have to build it on one of the megaclouds because they're the only ones that
can afford to build the infrastructure. One inc wincteny small problem. Companies built on top of large language models don't make much money, and in fact they're almost all deeply unprofitable. But let's establish a few flats to get going. I said, flats, flats, Jesus Christ.
Facts.
Here are the flats I'm establishing. Outside of one exception, mid Journey, which claimed it was profitable in twenty twenty two, which may not still be the case. I've actually reached out to ask them and they didn't get.
Back to me.
Every single LLM model is company is unprofitable, often wildly so. Outside of open ai and oropic, in any sphere which makes the AI coding app cursor, there are no large language model companies either building models or services on top of others models that make more than five hundred million
dollars in annualized revenue meaning month times twelve. Outside mid Journeys two hundred million arr and Iron clouds one hundred and fifty million arr Also fucking perplexity, there are only twelve generative AI powered companies making one hundred million dollars annualized or eighteen point three million dollars a month in revenue. The database then, this is the Information's AI. Generative AI database doesn't have replt, which also announced it hit one
hundred million in analyzed revenue. I've included it in my statement of facts. Of these companies, two of them have been acquired, move Works acquired by service Now in March twenty twenty five after the company shit the Big Big Time, and Windsurf, which was acquired by Google and Cognition in July twenty twenty five and one of the most annoying
deals of all time. But for the sake of simplicity, I've left out companies like Surge, Scale, Cheering, and Together, all of whom run consultancies selling services and training stuff for training models. Otherwise, there are seven companies total that make fifty million dollars or more annual recurring revenue, which is four point one six million dollars a month.
Now, none of this is to say.
That one hundred million dollars isn't a lot of money to you and me.
I just want to be clear.
If you want to give me one hundred million dollars, I'll do anything. I'll wink like a pig for you anyway. But in the world of software as a service or enterprise software, this is jump change HubSpot At revenues are two point six three billion dollars in its twenty twenty four financial year. Three years into this crap, and Generative AI's highest grossing companies outside of open Ai ten billion annualized as of June and Anthropic four billion annualized as July.
Don't like saying that word. Both of them loose billions a year after revenue. There are really three problems here. Businesses powered by Generative AI do not seem to be popular, Those businesses that are remotely popular are deeply unprofitable, and even the less popular generative AI powered businesses are also deeply unprofitable. But I want to start somewhere because I keep hearing about fucking Cursor. Fucking's start with any sphere
and Cursor and their app Cursor. It's an AI powered coding app and they have five hundred million dollars of annualized revenue.
Pretty great, right, ha.
It hit two hundred million dollars in annualized revenue in March and then hit five hundred million in June after raising nine hundred million dollars. That's amazing, ed, ed it's time walk to the garage. ED, it's over for you. Wrong,
it's a mirage. Cursor's growth was the result of an unsustainable business model that it's now had to replace with opaque terms of service, dramatically restricting access to models, and rate limits that effectively stop its users using the product at the price point they were used to go to Arsnash Cursor on red app Take a look. Take a look at how happy everyone is.
I want to know one.
My peers in the media don't seem to have the ability to talk to actual fucking customers. It's ridiculous. This company is circling the drain, and nobody seems to want to talk about it, despite how big a deal that is.
Oh.
Also, Curse is horribly unprofitable, and I believe there are
a sign of things to come in generative AI. A couple of weeks weeks ago, I wrote up the dramatic changes that Cursor made to its service in the middle of June or my premium newsletter and discovered that they timed these changes precisely with Anthropic and open Ai to a lesser extent, adding service tiers and priority processing, which is tech language for pay us extra if you have a lot of customers or face rate limits or service delays.
Asshole.
These price ships have also led to companies like replt having to make significant changes to their pricing model that disfavors users. People are finding in really simple terms that what they used to get for twenty bucks is much much, much, much much smaller curse the users hit rate limits. Replit users are hitting rate limits, and even then when they try and do the same things, they're spending way more
money if they go pay as you go. It's a complete fast But I'm going to repeat some of this stuff from the premium newsletter because there is a time of events that I believe are going to be in
the big short to AI Boogloo all right. In or around May fifth, twenty twenty five, Cursor closed the nine hundred million dollar funding round in a Around May twenty second, twenty twenty five, Anthropic launched Clawed four Opus and new models with Sona and Opus, both of them kind of well known for coding, and on May thirtieth, twenty twenty five, they added service tiers, including priority pricing specifically focused on cash heavy products like Cursor and the cash is when
you put stuff that you're going to be looking at regularly, take a look at it, and you can use it more readily. Cash is the CAC eight G, by the way, is generally something that's for efficiency.
The idea that you.
Would add a toll onto the cash is fucking disgusting and only targeted coding startups. But on May thirtieth, twenty twenty five, Reutter's reported the Anthropics annualized revenue hit three billion dollars, with a key driver being code generation. This translates to around two hundred and fifty million dollars in monthly revenue. June ninth, twenty twenty five, CNBC reported open
Ai'd hit ten billion dollars in annualized revenue. And yeah, when they said ann your recurring revenue, they meant annualized. But the very same day they cut the price of their three model by eighty percent, which competes directly with Clawed four Opus by the way, and This was a direct and aggressive attempt to force Anthropic to kind of like make too ether lower prices or compete. It's just
shtheads fuckinging around with assholes. But on or around June sixteen, twenty twenty five, Cursor changed its pricing, added a new two hundred dollar a month Ultra tier that, in their own words, was made possible by multi year partnerships with open Ai, Anthropic, Google, an Xai, which translates to multi year commitments to spend which can be amortized as monthly amounts.
A day later, on June seventeenth, Cursor dramatically changed its offering to it for its twenty dollars a month subscriptions to usage base, where one got at least the value of their subscription, so a twenty bus a month person would get more than twenty dollars of API course in compute, along with arbitrary rate limits and unlimited access to Cursor's
own slow model that its users hey. Then on June eighteenth, Repler and other vibe coding company that I had previously mentioned announced their effort based pricing increases that were massive.
July first, the Information reported.
The Anthropic hit four billion dollars of annualized revenue making three hundred and thirty million dollars a month, an increase of eighty three million dollars a month. We'll just under twenty five percent in the space of a month.
Hmm.
Where could that money have come from? In simpler terms, Cursor raised nine hundred million dollars and very likely had to hand large amounts of that money over to Open Air and Anthropic to keep doing business with them, then immediately change the terms of service to make them worse for their customers. And as I said at the time, and this is a direct quote from my news there, while some met, no, I can't do the Kevin Ruth's
voice and doing my own stuff, pardon me. While some may believe that Open AI and Anthropic hitting annualized revenue milestones is good news, you have to consider how these milestones were hit. Based on my reporting, I believe that both companies are effectively doing steroids, forcing massive infrastructural costs onto big customers as a means of covering the increasing costs of their own models. There is simply no other
aid to read this situation. By making these changes, Anthropic is intentionally making it harder for its larger costs largest customer to do business. By the way, Cursor is their largest customer, creating the extra revenue by making Cursors product worse by proxy. What's sickening about this particular situation. It doesn't really matter if Curs's customers are happy or sad.
They like open AI's Enterprise Priority Access API Anthropic in this case, require a long term commitment which involves a minimum through put of tokens per second as part of their tiered access program. If Curs's customers drop off, both Anthropic and open Ai still get their cut, and if curses customers somehow out spend those commitments, they'll either still get rate limited or any sphere willkin cur more costs.
Why do you care about this?
Well, Cursor is the largest and most successful genetive AI company by far. In these aggressive and desperate changes to its products suggest that a that its products are deeply unprofitable, and b that its current growth was the result of offering a product that it was not the one it would sell in the long term. Cursor misled its customers, and its current revenue is as a result, highly unlikely
to stay at this level. Worse still, two anthropic engineers left from the the Clawed Code team to go and work at Cursor two weeks ago, and they have already come back. This heavily suggests that whatever they saw over there wasn't compelling enough to make them stay. As I also said, while Cursor may have raised nine hundred million dollars, it was really open Aianthropic XAI and Google that got
that money. At this point, there are no profitable price AI startups, and it's highly unlikely that the new pricing models by both Cursor and Replet are going to help.
These are now the.
New terms of doing business with the big model companies, a shakedown where you pay for priority access or tears, or face indeterminate delays or rate limits. Any start up scaling into an enterprise integration of General AVII, which means in this case anything that requires a level of service uptime has to commit to both a minimum amount of months and the throughput of tokens, which means that the price of starting an AI company that gets any kind
of real market traction just dramatically increased. Well, one could say, oh, perhaps you don't need priority access. The need here is something that can be entirely judged by anthropic and open ai in a totally opaque manner. They can and they will throttle companies that are two demanding on their systems. It's proven by the fact that they've done this to curse them multiple times. But okay, why does curse them
out so much? And it's simple. Generative AI will not get big on selling consumer software without an enterprise SaaS story, they're dead And I realize, I know, okay, folks, it's kind of a little boring hearing about software as a service despite the fact that it's a huge, several hundred billion dollar industry. But this is the only place where generative AI can really make money. Companies buying hundreds of thousands of seats or how industries that rely on compute
grow and without that growth, they're going nowhere. To give you some context, Netflix makes about thirty nine billion dollars a year in subscription new from consumers, and Spotify about eighteen billion.
These are the.
Single most popular consumer software subscriptions in the world, and open ai is fifteen point five million subscribers. Suggest that open ai can't rely on them for the kind of growth that would actually make the company worth three hundred billion.
Dollars or more.
Cuzer, as it stands, is the one example of a company thriving using GENERATIVEA a software company selling software, and it appears its rapid growth was the result of selling a product at a massive loss. As it stands today, Curs's product is significantly worse and it's ready it's full of people furious at the company for the changes. In simpler terms, Curser was the company that people mentioned to prove that startups could make money by building on top
products on top of open AI and Anthropics models. Yet the truth is the only way to do so is to grow, and grow is to burn tons of money. While the tempting argument is to say that Curs's customers are addicted and will keep paying, this is clearly not the case, nor is it an actual business model. Like people that say this, I've never had a drug addiction, but I know people that do it. It's nothing like software. Stop making that comparison. It's insulting to the victims of addiction.
But anyway, this story showed that open A and Anthropics are actually their bigger the biggest threats to their customers and will actively rent seek can punish any of their success stories, looking to loose as much as they can from.
Them before they copy their products.
To put it bluntly, curses growth story was a fucking lie. It reached five hundred million dollars in annulif revenue selling a product it can no longer afford to sell and could not afford to sell long term, suggesting material weakness in its business and any and all coding startups. It's also remarkable, in the shocking failure of journalism that this isn't in every single article about any sphere. I'm doing this part time? Why am I in the asshole here?
Like I'm I don't know, really, though, I do have a question. Where are all the consumer AI starts? I'm genuinely serious.
What have you got for me?
Perplexity Perplexity. Perplexity only has one hundred and fifty million dollars in the annualized revenue, and they spent one hundred and sixty seven percent of their revenue in twenty twenty four, or fifty seven million dollars of spending on revenues of thirty four million dollars on computer services from Anthropic, Open AI and Amazon. They lost sixty eight million dollars and worse still, they still have no path to profitability and
it's not even making anything new. They're a search engine, they have an AI browser. But don't worry. Professional gas bag Alex Heath just did this insane and flumm mixing interview with CEO Aravins Ravinas, who, when asked how it perplexed you would become profitable, appeared to experience what seems to be a stroke like I'm about to read something to you and it's gonna sound strange, but this is exactly what was said. Maybe let me give you another example.
You want to put an ad on meta Instagram, and you want to look at ads done by similar brands, pull that, study that, or look at AdWords pricing of one hundred different keywords and figure out how to price your thing comparatively. These are tasks that could definitely save you hours and hours and maybe even give you up an arbitrage over what you could do yourself, because AI is able to do a lot more and at scale.
If it helps you to make a few million bugs, does it not make sense to spend two thousand dollars for that prompt? It does, right, So I think we're going to be able to monetize in many more interesting ways than chatbots for the browser. I want to be fucking clear about something. Alex seems like a nice guy. If someone said that to me, I'd ask them if they could smell toast. I'd be like, Aravin, Mate, are you okay? How many fingers I'm holding up?
Aravin? You're right?
Did you hit your head on something? The ceilings don't seem that low in here. But mate, you're just spewing utter fucking nonsense. I've read this paragraph multiple times. I do not know what he's getting at. I think he's suggesting something about how you could ask it to tell you what to do with ads.
I don't know. I don't know.
This is the big probably the biggest consumer AI company that isn't open AI, and they speak like they're an insane person or a stupid person. Check out the Business Idiot Trilogy for what I think there. I also mentioned them earlier, but I don't I don't want you to talk to me about AI browsers. Anyone humoring AI browsers is being an imbecile for some reason. They are not a business model. How are people going to make money on the browser. Hm hmm, what do these products actually do?
Oh? They can poorly automate accepting linked invites. Wow. Wow, it's like God himself has personally best my computer. A big fucking deal.
In any case, it doesn't seem like you can really build a consumer AI startup that makes any real money or approach being a real company other than chat GPT, I guess, and that's because the GENERATIVEAI software market is small, with little room for growth and no profits to be seen. Arguably, the biggest sign that things are are troubling in the generative AI spaces that we use the term annualized revenue at all, which, as I've mentioned repeatedly, means multiplying a
month by twelve and saying that's our annualized baby. The problem with this number is that, well, people cancel things. While your June might look great, if ten percent of your subscribers churning a bad month due to a change in your terms of service, for example, that's a huge chunk of your annualized revenue gone and likely gone forever. But the worst sign is that nobody is saying the
monthly figures, mostly because the monthly figures fucking suck. One hundred million dollars of anualized revenue is eight point three
three million dollars a month. To give you some scale, Amazon Web Services hit one hundred and eighty nine million dollars fifteen point seventy five million dollars a month in revenue in two thousand and eight, two years after founding, and while it took until twenty fifteen to hit profitability, it actually hit break even in two thousand and nine, though were invested in cash and growth for a few
years later. And I should be clear them doing that justified so many startups burning cash, so many starts like yeah, look at aws. They were investing in growth, which is a fair thing for companies to do. But I'm being
an asshole. But right now there is not a single generative AI software company that's profitable, and none of them are showing the signs of the kind of hypergrowth that previous big software companies had or Cursor technically is the fastest growing software as a service company of all time. It got there by basically lying. Cursor is never bringing back the product at the twenty dollars price point that
they were selling. They're never doing it. The money they earned was earned it's not fraud because they didn't do it.
I guess it was deceptive, but it's not really to the it's just fucking lying.
It's just lying. And who knows what happens to curser now. But you know what, I'm harping on cursor a bit. What other software startups are there?
Glean, Glean, fucking Glean, Glean, everyone loves to talk about.
Enterprise search company Glean, a company that uses AI to search and generate answers from your company's files and documents. Fun fact, also Salesforce's own Slack has now blocked them from searching Slack. Just arshole on arsehole violence. In December twenty twenty four, Glean raised two hundred and sixty million dollars, broadly stating that it had over five hundred and fifty million dollars in cash with best in class ARR growth.
A few months later, in February twenty twenty five, Glean announced it achieved one hundred million dollars in annual recurring revenue in fourth arter FY twenty five, cementing its position is one of the fastest growing sas startups and reflecting a searching demand for AI powered workplace intelligence. In any case, AR could literally mean anything, as it appears to be based on quarters, meaning it could be an average of
the last three months. I guess anyway. In June twenty twenty five, Glean announced it had raised another funding round, this time raising one hundred and fifty million dollars in It troublingly added that since its last round, it had surpassed one hundred million dollars in AR raw five months into the fucking year. And your revenue is basically the same. That isn't good. That isn't good at all. Also happened to that five hundred and fifty million dollars in cash?
Why did Glean need more? Hey, wait a second, take a look at this. Glean announced their raise on June eighteenth, twenty twenty five, two days after Curses price increase, in the same day that Repler announced the similar price act.
It's almost as if the.
Dramatic pricing increase has affected them due to the introduction of Anthropic Service TRES and Opening Eyes priority processing.
But I'm guessing. I know, I'm guessing.
But it is kind of where that all of these companies raise money and all announced these things around the same time.
Hey, that reminds me, I got another problem.
I got another problem here because I think that there is another reason why the cycles kind of keep repeating. You get a company of that grows, and then they kind of go nowhere, because well, the company doesn't really seem to have a total addressable market much bigger than one hundred million AR and I think it's a little simple.
It's quite simple.
In fact, there really are no unique generative AI companies, and building a moat on top of l elms is near impossible.
If you look a man, am I going to get some emails about this, but bring them on.
If you look at what GENERATIVEAI companies do, now that the following is not a quality barometer, it's probably one of the following things. They're either chatbot one, either you ask questions or talk to This includes customer service bots, searching, summarizing, or comparing documents with increased amounts of complexity of documents or quantity of documents to be compared. This includes being
able to ask questions of documents. Web search deep research, meaning long form web search that generates a document where some parts of it will inevitably be hallucinated or derived from low quality sources, generating text, images, voice, or in some rare cases video, Using AI to generative AII mean to write, edit or maintain code, transcription, translation, or photo
and video editing. Every single generative AI company that is an open Aireanthropic and honestly kind of those two does one or a few of these things, and I mean every one of them. And it's because every single generative AI company uses large language models, which have inherent limits on what they can do. Llms can generate, they can search, they can kind of edit, they can sometimes transcribe accurately, and they can sometimes translate much more well, much less accurately.
I guess within weeks of Curses changed to its services, Amazon and byte Dance release competitors that, for the most part, do.
Exactly the same thing.
Sure, there's a few differences in how they're designed, but design is not a moat, especially in a high cost, negative profit business were your only way of growing is
to offer a product you can't sustain. The only other moat you can build is the services you provide, which, when your services are dependent on a large language model, are dependent on the model developer, who, in the case of open AI and Anthropic, could simply clone your startup, because the only valuable intellectual property is the models, and those models are theirs. You may say, well, nobody else has any ideas either, to which I say, I fully agree.
My rock com bubble thesis suggests that we're all out of hyper growth ideas, and yeah, I think we're out of ideas related to any large language models too. At this point, I think it's fair to ask, are there any good businesses you can build on top of generative AI or large language models. I don't mean ad features related to I mean an AI company that actually sells a product that people buy at scale that isn't called chat,
GPT or claude. In previous tech booms, companies would make their own models, their own infrastructure, or the things that make them distinct from other companies. But the generative AI boom effectively changes that by making everybody build on stuff on top of somebody else's models, because training your own models is both extremely expensive and requires vast amounts of
infrastructure and just pure power. As a result, much of this boom is about a few companies, really too, if we're honest, getting other companies to try and build functional software for them, and these companies Open ai and Anthropic are their customers weak point in a relationship that veers from symbiotic to parasitic at a moment's notice. I cannot stress enough how bad open ai and Anthropic are for
their business customers. Their models are popular, by which I mean their customers customers will expect access to them, meaning the open ai and Anthropic can, as they did to Cursor, arbitrarily change pricing, service availability, and functionality based on how they feel that day or whether they need to pump their annualized revenue for investors.
Don't believe me.
Anthropic cut off access to AI coding platform Windsurf because it looked like they might get acquired by open Ai. They never were. They just harmed that business. They just cut a hole in them. Why because they might touch another business, the most anti competitive shit in the world. And everyone sat there clapping like a fucking seal. Disgusting even by big tech standards. This fucking sucks, and these
companies will do it again. But you know what, Let's talk about the actual uses of generative AI, because the limited number of use cases are because large language models are all really really similar. Because all large language models require more data than anyone who's ever needed, including like four times the amount of data on the Internet. They all basically have to use the same thing, either taken from the Internet or bought from one of the few
companies that scale surge during together or whoever. While they can get customized data or do customized training and reinforcement learning, these models are all transformer based and they all function similarly, and the only way to make them different is by training them, which doesn't make them that much different, just better things they already do. And good lord, is it so is general IFAI is so ungodly expensive and the training is as well. By the way, they have to
pay real humans as well, which they hate doing. And even when they're paying outsourced labor and ken youre at two dollars a pop, they're still losing a ton of money. It's really crazy, actually, how badly built all of this is. And I already mentioned open AI and Anthropics costs as well as perplex The's fifty million dollar bill in a year to Anthropic Amazon and open Ai off of a
easily thirty four dollars million dollars in revenue. These companies cost too much to run and their functionality doesn't make enough money to make them make sense. And the problem isn't just the pricing, but how unpredictable it is. As Matterscheer wrote for cio Dive last year, generative AI makes a lot of companies lives difficult for the massive spikes and costs that from the power users, with few ways
to mitigate those costs. One of the ways that company manages their cloud bills is by having some degree of predictability, which is difficult to do with the constant sleu of new models and demands some new products to go with them, especially when send models can and can and do often cost more with subsequent iterations, not necessarily for much return, except if you're a company like a coding company, your customers are going to actually ask you for the new models.
As a result, it's half AI companies to actually budge in. But ed, What was that? Ed?
What about agents?
Aren't they the thing that will eventually make the insane broken calculus behind generative AI actually work?
What is your accent made? Anyway? Anyway?
Let me tell you about agents. The term agent is one of the most egregious acts of fraud I've seen in my entire career writing about this crap, and that includes the metavers. When you hear the word agent, you were meant to think of an autonomous AI that can go and do stuff without oversight, replacing someone's job in the process. And companies have been pushing the boundaries of
good taste and financial crimes in pursuit of them. Most egregious of them as Salesforce's Agent Force, which leads you deploy AI agents at scale. That's a quote, and brings digital labor to every employee, department and business process. Another quote from Salesforce's website. These are two blatant fucking lies. Agent Force is a goddamn chatbot program. It's a platform
for launching chatbots. They can sometimes plug into APIs that allow them to access other information, but they're neither autonomous nor agents by any reasonable definition. Not only does Salesforce not actually sell agents, its own research shows that the agents and agents in general only achieve around fifty eight percent success rate on single step tasks. And I'm going
to quote the register here. This means tasks that can be completed in the single step without needing follow up actions and more information or multi step tasks, So you know, most tasks they succeed a depressing thirty five percent of the time. Last week, open Ai announced its own chat GPT agent that can allegedly go and do tasks on
a virtual computer. In its own demo, the agent took twenty one minutes or so to spit out a plan for a wedding with destinations of a cander and some suit options, and then showed a pre prepared demo of the agent preparing an itinery of how to visit every major league ballpark. And that's baseball for the non Americans out there. In this example's case, agents took twenty three minutes and produced arguably the most confusing map I've seen in my life. You can see the map in the
newsletter version of this episode. It's hilarious. It missed out every single major ballpark on the East Coast, including Yankee Stadium and Femway Park, which are two of the most well known stadiums in sports, and added a bunch of roundom ones and like one in the middle of the Gulf of Mexico. What team is that, Sammy the deep Water Horizon Devils. Is there a baseball team in North Dakota?
Clammy?
Sammy Samy. I also should be clear this was a pre prepared example. This is the best they had. I want to see the cutting room footage on this, because you best bet that that map looked like straight dogshit, as with every large language model product, And yes, that's what this is, even if open ai won't talk about what model results are. Extremely variable agents are difficult because tasks are if for coal, even if they can be completed by a human being, that the CEO thinks is stupid.
What open ai appears to be doing is using a virtual machine to run scripts that its models trigger regardless of how will it works, and it works very very very very poorly and inconsistently. It's also very likely expensive to run. In any case, every single company you see
using the word agent is trying to mislead you. They're lying gleans ai agents to chatbots with If this, then that functions that trigger events using APIs, which means if an event happens, another thing will be triggered, not taking actual actions, because that is not what lms can do. Service Now's ai agents that allegedly act autonomously and proactively on your behalf are despite claiming they go beyond better chatbots still ultimately better chatbots that use APIs to trigger
different events using if this, then that functions. Sometimes these chatbox can also answer questions that people might have or trigger an event somewhere. Oh right, that's literally the same thing. The closest we have to an agent is any kind of coding agent, which is they can make a list of things that you might do on our software project and go and generate code and push stuff to get
help when you ask them to. And they can do so autonomously in the sense that you can just let them do what a model that doesn't know anything and has no consciousness thinks is right based on its corpus of data and the things you can access to. And it's about as safe as that sounds. When I say ask them to and go, and I mean that these agents are not intelligent at all. They do not have intelligence, and when let run rampant, fuck up everything and create
a bunch of extra work or so. A study found that AI coding tools made engineers nineteen percent slower. Nevertheless, none of these products are autonomous agents, and anybody using the term agent likely means chat bond. And all of this is working because the media keeps repeating everything these companies say. It's a disgrace. We need to stop this. I realize we've taken a kind of a scenic route here, though, but I know he needed to lay the groundwork, because
I really am alarmed. According to a UBS report from the twenty sixth of June, the public companies running AI services are making absolutely pathetic amounts of money from AI. Microsoft, according to the UBS, is making annual revenues of somehow less than the Information report at two point one billion dollars. Service now is making less than two hundred and fifty million, Adobe less than one hundred and twenty five million salesforce less than one hundred million now service now said two
hundred and fifty million dollar ACV annual contract value. This may be one of the more honest explanations of revenue I've seen, putting them in the upper echelons of AI revenue and less. Of course, you think about it for a couple of seconds and think, are these all AI specific contracts or perhaps they're in contracts where you've taped AI on to the side.
It gives a shit. It's also year.
Long agreements that could churn, and according to Gartner, over forty percent of Agenetic AI products will be canceled by end of twenty twenty seven. And really, you gotta laugh at Adobe and Salesforce, both of whom to talk such a goddamn fuck ton about jenerit of Ai. And yeah, I have only made amazed the hundred twenty five million in analyzed revenue from it.
Pathetic crap, dog shit.
These aren't futuristic numbers, they're barely product categories, and none of this seems to include costs.
Oh well, good grief.
Look, a lot of what I've been saying is reminiscent the previous podcasts, and I've gone over this a lot, so I really want to make it clear that the signs are very troubling, and that the things I've warned you about the past couple of years are only getting worse, and the cliff's coming up things are only getting closer.
When we double off of it, things may get really, really bad, And then the next episode we'll talk about how and what that tumble might look like and the noises I'm going to make when it happens.
Thank you for listening to Better Offline, The editor and composer of the Better Offline theme song is Matasowski. You can check out more of his music and audio projects at Matasowski dot com, m a Ttoso w s Ki dot com. You can email me at easy at Better Offline dot com or visit Better Offline dot com to find more podcast links and of course, my newsletter. I also really recommend you go to chat dot where's youreaed dot at to visit the discord, and go to our slash.
Better Offline to check out I'll Reddit. Thank you so much for listening. Better Offline is a production of cool Zone Media.
For more from cool Zone Media, visit our website cool Zonemedia dot com, or check us out on the iHeartRadio app, Apple Podcasts, or wherever you get your podcasts.