| FAO Dan re SaaS etc 08:47 - Feb 13 with 2246 views | nrb1985 |
Genuinely interested from your side of things, as an expert, as to why you think the AI hype is over blown or won’t deliver what the “AI optimists” seem to think it might.
As you will have no doubt seen, utter carnage in the SaaS stocks this week with the various Claude releases and others. Granted, they are fund raising at the moment so it’s in their interests to show off as many new releases as possible - but it’s not just them. There was software for tax planning released this week too from a random startup which will clobber my industry for example.
And we are only in the first innings of this…
How do you see things? [Post edited 13 Feb 8:48]
|  | | |  |
| FAO Dan re SaaS etc on 09:42 - Feb 13 with 1637 views | DanTheMan |
There's loads of technical stuff I could go into but it really all boils down to this.
How are they going to make money off it?
AI is insanely expensive to run, requiring dedicated data centres, power and everything silicon based. The main part, the graphics cards, only last a few years of abuse before they need replacing and they are very, very expensive.
Basically, not cheap to run.
As far as I've seen none of the major companies are actually making money off of AI. The easiest ones to look at are Anthropic and OpenAI who are both haemoraging money.
Now, maybe there is some magic solution to increasing efficiency but as it is right now the sums just do not add up to me. The more people who use it, the more money it loses. In any other business model, that's absolute insane. OpenAI was losing money on people on the $200 a month tier.
As you point out, it's also rather telling how they talk about things.
This is a good example:
Maybe I'm just jaded from another technology hype machine.
First crypto was going to take over, then it was NFTs, then it was the Metaverse (lol), now we're onto AI.
I think AI is more useful than the other three to be clear, it actually has some purpose and depiste my scepticism it can be useful. I use it myself for work.
But until the sums start making sense, I fail to see how this works long term. LLMs aren't even close to AGI, and without massively increasing context windows (which costs more money) or some really insane efficiency savings I don't understand why VC companies would keep pumping money into it expecting zero returns.
If you want a way more in-depth skeptic point of view, I can recommend Edward Zitron. His writing style is... unique and I don't agree with everything he says but he goes into a lot of detail.
https://www.wheresyoured.at/th |  |
| |
| FAO Dan re SaaS etc on 09:48 - Feb 13 with 1595 views | BlueBadger |
| FAO Dan re SaaS etc on 09:42 - Feb 13 by DanTheMan |
There's loads of technical stuff I could go into but it really all boils down to this.
How are they going to make money off it?
AI is insanely expensive to run, requiring dedicated data centres, power and everything silicon based. The main part, the graphics cards, only last a few years of abuse before they need replacing and they are very, very expensive.
Basically, not cheap to run.
As far as I've seen none of the major companies are actually making money off of AI. The easiest ones to look at are Anthropic and OpenAI who are both haemoraging money.
Now, maybe there is some magic solution to increasing efficiency but as it is right now the sums just do not add up to me. The more people who use it, the more money it loses. In any other business model, that's absolute insane. OpenAI was losing money on people on the $200 a month tier.
As you point out, it's also rather telling how they talk about things.
This is a good example:
Maybe I'm just jaded from another technology hype machine.
First crypto was going to take over, then it was NFTs, then it was the Metaverse (lol), now we're onto AI.
I think AI is more useful than the other three to be clear, it actually has some purpose and depiste my scepticism it can be useful. I use it myself for work.
But until the sums start making sense, I fail to see how this works long term. LLMs aren't even close to AGI, and without massively increasing context windows (which costs more money) or some really insane efficiency savings I don't understand why VC companies would keep pumping money into it expecting zero returns.
If you want a way more in-depth skeptic point of view, I can recommend Edward Zitron. His writing style is... unique and I don't agree with everything he says but he goes into a lot of detail.
https://www.wheresyoured.at/th |
I read all that Tweet and my main takeaways from it are :
1. How long can they keep up the con?
2. What the hell kind of name for a dog is 'Bayesian' |  |
| |
| FAO Dan re SaaS etc on 10:01 - Feb 13 with 1536 views | tonybied |
| FAO Dan re SaaS etc on 09:42 - Feb 13 by DanTheMan |
There's loads of technical stuff I could go into but it really all boils down to this.
How are they going to make money off it?
AI is insanely expensive to run, requiring dedicated data centres, power and everything silicon based. The main part, the graphics cards, only last a few years of abuse before they need replacing and they are very, very expensive.
Basically, not cheap to run.
As far as I've seen none of the major companies are actually making money off of AI. The easiest ones to look at are Anthropic and OpenAI who are both haemoraging money.
Now, maybe there is some magic solution to increasing efficiency but as it is right now the sums just do not add up to me. The more people who use it, the more money it loses. In any other business model, that's absolute insane. OpenAI was losing money on people on the $200 a month tier.
As you point out, it's also rather telling how they talk about things.
This is a good example:
Maybe I'm just jaded from another technology hype machine.
First crypto was going to take over, then it was NFTs, then it was the Metaverse (lol), now we're onto AI.
I think AI is more useful than the other three to be clear, it actually has some purpose and depiste my scepticism it can be useful. I use it myself for work.
But until the sums start making sense, I fail to see how this works long term. LLMs aren't even close to AGI, and without massively increasing context windows (which costs more money) or some really insane efficiency savings I don't understand why VC companies would keep pumping money into it expecting zero returns.
If you want a way more in-depth skeptic point of view, I can recommend Edward Zitron. His writing style is... unique and I don't agree with everything he says but he goes into a lot of detail.
https://www.wheresyoured.at/th |
I guess frivolous use of AI doesn't help things! How much bandwidth/processing power is used by stupid trends, like making a cartoon avatar of yourself doing your job? |  | | |
| FAO Dan re SaaS etc on 10:25 - Feb 13 with 1449 views | Kievthegreat |
| FAO Dan re SaaS etc on 10:01 - Feb 13 by tonybied |
I guess frivolous use of AI doesn't help things! How much bandwidth/processing power is used by stupid trends, like making a cartoon avatar of yourself doing your job? |
But it does help them. If people didn't use their models and services frivolously when it's cheap/free, then their user numbers would crash. Number of regular users is a key metric when they pitch themselves to VCs and investors because that's the customer base they will magically monetise and that will definitely pay £200 a month to make silly pictures if you invest £20bn... |  | | |
| FAO Dan re SaaS etc on 10:31 - Feb 13 with 1425 views | tonybied |
| FAO Dan re SaaS etc on 10:25 - Feb 13 by Kievthegreat |
But it does help them. If people didn't use their models and services frivolously when it's cheap/free, then their user numbers would crash. Number of regular users is a key metric when they pitch themselves to VCs and investors because that's the customer base they will magically monetise and that will definitely pay £200 a month to make silly pictures if you invest £20bn... |
It might help them to sell the snake oil, but is it ultimately going to make money? | | | |
| FAO Dan re SaaS etc on 10:42 - Feb 13 with 1395 views | Kievthegreat |
| FAO Dan re SaaS etc on 10:31 - Feb 13 by tonybied |
It might help them to sell the snake oil, but is it ultimately going to make money? |
If they pump the valuation enough, then someone will! Not convinced society will benefit from the shock correction to values though. | | | |
| FAO Dan re SaaS etc on 10:56 - Feb 13 with 1315 views | soupytwist |
| FAO Dan re SaaS etc on 09:48 - Feb 13 by BlueBadger |
I read all that Tweet and my main takeaways from it are :
1. How long can they keep up the con?
2. What the hell kind of name for a dog is 'Bayesian' |
1. Don't know, surely those VCs with all that money can't be that naive can they?
2. Hope it fares better than the boat with the same name. |  | | |
| FAO Dan re SaaS etc on 10:59 - Feb 13 with 1294 views | lowhouseblue |
| FAO Dan re SaaS etc on 09:42 - Feb 13 by DanTheMan |
There's loads of technical stuff I could go into but it really all boils down to this.
How are they going to make money off it?
AI is insanely expensive to run, requiring dedicated data centres, power and everything silicon based. The main part, the graphics cards, only last a few years of abuse before they need replacing and they are very, very expensive.
Basically, not cheap to run.
As far as I've seen none of the major companies are actually making money off of AI. The easiest ones to look at are Anthropic and OpenAI who are both haemoraging money.
Now, maybe there is some magic solution to increasing efficiency but as it is right now the sums just do not add up to me. The more people who use it, the more money it loses. In any other business model, that's absolute insane. OpenAI was losing money on people on the $200 a month tier.
As you point out, it's also rather telling how they talk about things.
This is a good example:
Maybe I'm just jaded from another technology hype machine.
First crypto was going to take over, then it was NFTs, then it was the Metaverse (lol), now we're onto AI.
I think AI is more useful than the other three to be clear, it actually has some purpose and depiste my scepticism it can be useful. I use it myself for work.
But until the sums start making sense, I fail to see how this works long term. LLMs aren't even close to AGI, and without massively increasing context windows (which costs more money) or some really insane efficiency savings I don't understand why VC companies would keep pumping money into it expecting zero returns.
If you want a way more in-depth skeptic point of view, I can recommend Edward Zitron. His writing style is... unique and I don't agree with everything he says but he goes into a lot of detail.
https://www.wheresyoured.at/th |
there was a good interview at the end of radio 4's pm yesterday with cory doctorow who said pretty much the same as you. |  |
| And so as the loose-bowelled pigeon of time swoops low over the unsuspecting tourist of destiny, and the flatulent skunk of fate wanders into the air-conditioning system of eternity, I notice it's the end of the show |
| |
Login to get fewer ads
| FAO Dan re SaaS etc on 11:05 - Feb 13 with 1265 views | Guthrum |
So much of the issue is down to bandwagon-jumping. Reminds me of the dot-com boom of around 2000 ... or the South Sea Bubble of 1720 and Tulip Mania the previous century.
A marvellous new THING of more or less actual money-making potential is seized upon by public opinion and the investment community. Fear of missing out and grandiose claims result in stupid amounts of money being poured into schemes built upon flimsy and ill-thought-out foundations (often negligent, some downright fraudulent). Funds are set up purely to invest in other funds investing in the original idea. Traditional investment channels are drained by the sheer force of flow in the alternative direction.
Eventually the whole thing overheats and crashes when the various entities have burnt through their cash piles and still aren't making any money. A lot of the investors are locked in and cannot escape, or there is simply nothing to claim back. Many institutions (e.g. pension funds, charities, local authorities) suffer greatly, now having huge black holes in their accounts.
A few of the hardiest, most careful and best planned businesses will survive. They may go on to great success, if there was an actual market for them to exploit in the first place (with the South Sea Bubble, there wasn't). The basic principles may go on to become a foundational part of future life, but at a more basic level (e.g. e-commerce).
It's less a technological issue than an investment one. |  |
| |
| FAO Dan re SaaS etc on 11:06 - Feb 13 with 1258 views | USA |
A high quality thread. Wow. Interesting read thanks. | |
| |
| FAO Dan re SaaS etc on 11:10 - Feb 13 with 1243 views | Guthrum |
| FAO Dan re SaaS etc on 10:56 - Feb 13 by soupytwist |
1. Don't know, surely those VCs with all that money can't be that naive can they?
2. Hope it fares better than the boat with the same name. |
Problem is, the venture capital funds are still making fees and commissions from investing what is really other people's money. Trading is what makes their income, not specifically returns (tho that will affect reputation and therefore future earnings). | |
| |
| FAO Dan re SaaS etc on 11:14 - Feb 13 with 1227 views | lowhouseblue |
| FAO Dan re SaaS etc on 11:10 - Feb 13 by Guthrum |
Problem is, the venture capital funds are still making fees and commissions from investing what is really other people's money. Trading is what makes their income, not specifically returns (tho that will affect reputation and therefore future earnings). |
and with vc funds, while valuations rise it looks as if they're doing a good job and they get their fees. when they come to sell their holdings and can't realise the assumed value - which is the issue in the current market - that's the point at which the investors lose their money. the vc managers still keep their fees. | |
| And so as the loose-bowelled pigeon of time swoops low over the unsuspecting tourist of destiny, and the flatulent skunk of fate wanders into the air-conditioning system of eternity, I notice it's the end of the show |
| |
| FAO Dan re SaaS etc on 11:16 - Feb 13 with 1215 views | Kievthegreat |
| FAO Dan re SaaS etc on 11:05 - Feb 13 by Guthrum |
So much of the issue is down to bandwagon-jumping. Reminds me of the dot-com boom of around 2000 ... or the South Sea Bubble of 1720 and Tulip Mania the previous century.
A marvellous new THING of more or less actual money-making potential is seized upon by public opinion and the investment community. Fear of missing out and grandiose claims result in stupid amounts of money being poured into schemes built upon flimsy and ill-thought-out foundations (often negligent, some downright fraudulent). Funds are set up purely to invest in other funds investing in the original idea. Traditional investment channels are drained by the sheer force of flow in the alternative direction.
Eventually the whole thing overheats and crashes when the various entities have burnt through their cash piles and still aren't making any money. A lot of the investors are locked in and cannot escape, or there is simply nothing to claim back. Many institutions (e.g. pension funds, charities, local authorities) suffer greatly, now having huge black holes in their accounts.
A few of the hardiest, most careful and best planned businesses will survive. They may go on to great success, if there was an actual market for them to exploit in the first place (with the South Sea Bubble, there wasn't). The basic principles may go on to become a foundational part of future life, but at a more basic level (e.g. e-commerce).
It's less a technological issue than an investment one. |
I've seen some try and argue that "AI Mania" could go the same way as "Railway Mania". Yes lots of companies go bankrupt, but look at all the wonderful infrastructure left behind that we can utilise. However I'm not convinced nuclear powered data centres with worn out silicon is really what we need or that it has anywhere near the impact of 10,000s of miles of track that are still in use today! [Post edited 13 Feb 11:16]
| | | |
| FAO Dan re SaaS etc on 11:25 - Feb 13 with 1155 views | townblue |
Some SaaS vendors should probably be worried.
At my company we are looking at some new tools and where I would always suggest buying something that is designed for the job and adequately validated, our CTO has decided to try vibe coding something with AI. He has spent hardly any time and has produced something that at least looks like it could do the job.
I have reservations about the approach as who supports, maintains, tests it works correctly, but if it's using this or spending thousands on a SaaS system I think they will choose this. Then think of companies who can have a whole team of people doing this and self building could become a lot more attractive than it has been historically when you needed a team of highly paid engineers to produce bespoke systems. |  | | |
| FAO Dan re SaaS etc on 11:26 - Feb 13 with 1136 views | Guthrum |
| FAO Dan re SaaS etc on 11:16 - Feb 13 by Kievthegreat |
I've seen some try and argue that "AI Mania" could go the same way as "Railway Mania". Yes lots of companies go bankrupt, but look at all the wonderful infrastructure left behind that we can utilise. However I'm not convinced nuclear powered data centres with worn out silicon is really what we need or that it has anywhere near the impact of 10,000s of miles of track that are still in use today! [Post edited 13 Feb 11:16]
|
Tho even with Railway Mania, we were left with a huge amount of route duplication, inefficiency due to poor foresight* and a mass of tiny branch lines which pretty much never broke even. People go on about Beeching, but a lot of what he cut (and there were many closures before and after his time, too) was simply not needed or non-viable.
The financial and material waste was enormous.
Much of the bubble in that instance was cuased by over-promising returns on investment - and sometimes paying returns to early investors from the capital injected later.
* e.g. The tunnels at Welwyn, which have always been a major bottleneck on the main lines heading north and north-west out of London. | |
| |
| FAO Dan re SaaS etc on 11:33 - Feb 13 with 1111 views | Guthrum |
| FAO Dan re SaaS etc on 11:25 - Feb 13 by townblue |
Some SaaS vendors should probably be worried.
At my company we are looking at some new tools and where I would always suggest buying something that is designed for the job and adequately validated, our CTO has decided to try vibe coding something with AI. He has spent hardly any time and has produced something that at least looks like it could do the job.
I have reservations about the approach as who supports, maintains, tests it works correctly, but if it's using this or spending thousands on a SaaS system I think they will choose this. Then think of companies who can have a whole team of people doing this and self building could become a lot more attractive than it has been historically when you needed a team of highly paid engineers to produce bespoke systems. |
Tho it is problematic to self-code all your systems if they need to be compatible with other people's. If everybody is reinventing the wheel themselves every time, we just end up with a lot of wagons which don't roll very well.
Industry standards come to exist for a reason. | |
| |
| FAO Dan re SaaS etc on 11:36 - Feb 13 with 1102 views | DanTheMan |
| FAO Dan re SaaS etc on 11:25 - Feb 13 by townblue |
Some SaaS vendors should probably be worried.
At my company we are looking at some new tools and where I would always suggest buying something that is designed for the job and adequately validated, our CTO has decided to try vibe coding something with AI. He has spent hardly any time and has produced something that at least looks like it could do the job.
I have reservations about the approach as who supports, maintains, tests it works correctly, but if it's using this or spending thousands on a SaaS system I think they will choose this. Then think of companies who can have a whole team of people doing this and self building could become a lot more attractive than it has been historically when you needed a team of highly paid engineers to produce bespoke systems. |
"looks like it could do the job"
And herein lies the danger. Your thinking is right, who is going to support it in years to come.
Sure, if it's something really simple you're probably fine but with software development the issue has never been writing code. That's the easy bit.
Reading code is hard. Getting correct requirements is hard. Refactoring large code bases is hard. Designing something so you don't have to do costly rewrites next year, or in two years, is hard.
It's a bit like the Pareto principle, it takes 20% of the effort to do 80% of the code. But that final 20% is 80% of the effort.
I'm working on a codebase that is now 500k LOC. An AI just cannot work on it to do anything large scale. It cannot take in that much context.
I think you're right in that lots of people will write things that seem to work(ish) instead of buying battle tested software and then a year or two down the line will be left in an utter mess because nobody has a clue how this thing AI produced even works. [Post edited 13 Feb 11:36]
| |
| |
| FAO Dan re SaaS etc on 11:44 - Feb 13 with 1052 views | ghostofescobar |
| FAO Dan re SaaS etc on 09:42 - Feb 13 by DanTheMan |
There's loads of technical stuff I could go into but it really all boils down to this.
How are they going to make money off it?
AI is insanely expensive to run, requiring dedicated data centres, power and everything silicon based. The main part, the graphics cards, only last a few years of abuse before they need replacing and they are very, very expensive.
Basically, not cheap to run.
As far as I've seen none of the major companies are actually making money off of AI. The easiest ones to look at are Anthropic and OpenAI who are both haemoraging money.
Now, maybe there is some magic solution to increasing efficiency but as it is right now the sums just do not add up to me. The more people who use it, the more money it loses. In any other business model, that's absolute insane. OpenAI was losing money on people on the $200 a month tier.
As you point out, it's also rather telling how they talk about things.
This is a good example:
Maybe I'm just jaded from another technology hype machine.
First crypto was going to take over, then it was NFTs, then it was the Metaverse (lol), now we're onto AI.
I think AI is more useful than the other three to be clear, it actually has some purpose and depiste my scepticism it can be useful. I use it myself for work.
But until the sums start making sense, I fail to see how this works long term. LLMs aren't even close to AGI, and without massively increasing context windows (which costs more money) or some really insane efficiency savings I don't understand why VC companies would keep pumping money into it expecting zero returns.
If you want a way more in-depth skeptic point of view, I can recommend Edward Zitron. His writing style is... unique and I don't agree with everything he says but he goes into a lot of detail.
https://www.wheresyoured.at/th |
On the “expensive” part of the argument, server racks for these data centres can be worth up to USD4mill each! I work within the logistics and shipping sector and some of the values for single shipments of multiple racks are eye watering. Don’t want to be too specific for confidentiality reasons, but I am aware of a couple of trailer loads of racks being stolen in the North America in the same incident, valued at over USD20mill per trailer. One theory being investigated is that state actors (N Korea) may have ordered the theft so that they can get their hands on high end, industrial chips, as they can’t make them and sanctions make it very difficult to import them. Funny old world…… |  |
| |
| FAO Dan re SaaS etc on 11:50 - Feb 13 with 1023 views | eireblue |
| FAO Dan re SaaS etc on 11:25 - Feb 13 by townblue |
Some SaaS vendors should probably be worried.
At my company we are looking at some new tools and where I would always suggest buying something that is designed for the job and adequately validated, our CTO has decided to try vibe coding something with AI. He has spent hardly any time and has produced something that at least looks like it could do the job.
I have reservations about the approach as who supports, maintains, tests it works correctly, but if it's using this or spending thousands on a SaaS system I think they will choose this. Then think of companies who can have a whole team of people doing this and self building could become a lot more attractive than it has been historically when you needed a team of highly paid engineers to produce bespoke systems. |
Quite a lot of software and SaaS products are built, in simple terms, in three layers.
Data, Compute, User Interface.
The third layer, a UI that humans do stuff in, is sometimes what is sold, and licensed.
So the value of a traditional SaaS company, is that they sell x number of licenses, to do Y functions.
The data, is usually a companies data. The compute and UI, is specific to the function.
Companies buy lots of SaaS software.
So let’s talk about AI…..it can do compute and present data any way like.
Why do I now need humans using a specific UI and compute that I pay for, per seat, to use my data.
All I really care about is data, and being able to process and present as efficiently as possible.
So why buy multiple SaaS products, when I could have a Data layer and an AI Agent Layer.
As an AI company is now, I would go to a organisation and say, why are you paying for 7 SaaS platforms, at £xxx per license per platform.
I’ll sell you an AI Agentic platform at £xx per license, and you can also reduce headcount.
The AI opportunity, is to simplify the data supply chain.
Just with any company, you have raw material, add value, sell.
Coals to Newcastle, Ice to Eskimos, Tofu to vegans.
As a SaaS company, still need to make it cheap to get/find data, but the add value and sell, is potentially cheaper with AI.
Companies that also make it easier to acquire data, and simply make it available for Agentic AI platforms, should also be successful.
That’s my take.
There maybe a bubble/crash/recovery, as companies start to work out how to use all this, they may start buying too much and not become efficient, and then start to get properly efficient, buy less AI. [Post edited 13 Feb 11:54]
|  | | |
| FAO Dan re SaaS etc on 11:53 - Feb 13 with 997 views | Guthrum |
| FAO Dan re SaaS etc on 11:36 - Feb 13 by DanTheMan |
"looks like it could do the job"
And herein lies the danger. Your thinking is right, who is going to support it in years to come.
Sure, if it's something really simple you're probably fine but with software development the issue has never been writing code. That's the easy bit.
Reading code is hard. Getting correct requirements is hard. Refactoring large code bases is hard. Designing something so you don't have to do costly rewrites next year, or in two years, is hard.
It's a bit like the Pareto principle, it takes 20% of the effort to do 80% of the code. But that final 20% is 80% of the effort.
I'm working on a codebase that is now 500k LOC. An AI just cannot work on it to do anything large scale. It cannot take in that much context.
I think you're right in that lots of people will write things that seem to work(ish) instead of buying battle tested software and then a year or two down the line will be left in an utter mess because nobody has a clue how this thing AI produced even works. [Post edited 13 Feb 11:36]
|
I was having an interesting converstion with someone who said one of the biggest issues with AI is that it has no memory, in the way we humans do. They can re-check their sources with incredible rapidity, but have no recollection of what they did last week. Comes down to the cost of data storage.
So AI knows what its sources say is the optimum choice now, in the moment, but it does not know why a choice was made previously. So, unless it can be fed that information, there is no previous frame of reference.
If a crucial factor falls within a gap in the AI's sources, even if it was informed historically there will be ignorance now. So the output will be flawed.
And, if you're using the same AI to check the work, it will not be spotted. | |
| |
| FAO Dan re SaaS etc on 11:57 - Feb 13 with 943 views | DanTheMan |
If anyone wants a funny read on what open source maintainers are now having to deal with on a daily basis:
https://github.com/sympy/sympy | |
| |
| FAO Dan re SaaS etc on 12:01 - Feb 13 with 897 views | itfcserbia |
| FAO Dan re SaaS etc on 10:01 - Feb 13 by tonybied |
I guess frivolous use of AI doesn't help things! How much bandwidth/processing power is used by stupid trends, like making a cartoon avatar of yourself doing your job? |
Yeah, that one cartoon costs as much as an hour of electricity for a small African village |  |
| |
| FAO Dan re SaaS etc on 12:04 - Feb 13 with 879 views | eireblue |
| FAO Dan re SaaS etc on 11:33 - Feb 13 by Guthrum |
Tho it is problematic to self-code all your systems if they need to be compatible with other people's. If everybody is reinventing the wheel themselves every time, we just end up with a lot of wagons which don't roll very well.
Industry standards come to exist for a reason. |
Not really a problem.
There is a sector today for ETL tools (Extract, Transform, Load)
Sure there are standards, but there are multiple standards.
I don’t need an ETL product.
I just tell AI, see that API over there, pull data and chuck it over there. Ta.
AI agents could disrupt an industry, that is already in place because people use multiple standards etc…today anyway. | | | |
| FAO Dan re SaaS etc on 12:07 - Feb 13 with 858 views | MattinLondon |
| FAO Dan re SaaS etc on 09:48 - Feb 13 by BlueBadger |
I read all that Tweet and my main takeaways from it are :
1. How long can they keep up the con?
2. What the hell kind of name for a dog is 'Bayesian' |
If I had a dog I’ll call him Dave - Dave the Dalmatian.
Or Gary the German Shepherd.
Or Larry the Labrador.
Thinking about it, my dogs name would be heavily influenced by what breed it is. |  | | |
| FAO Dan re SaaS etc on 12:09 - Feb 13 with 848 views | nrb1985 |
| FAO Dan re SaaS etc on 09:42 - Feb 13 by DanTheMan |
There's loads of technical stuff I could go into but it really all boils down to this.
How are they going to make money off it?
AI is insanely expensive to run, requiring dedicated data centres, power and everything silicon based. The main part, the graphics cards, only last a few years of abuse before they need replacing and they are very, very expensive.
Basically, not cheap to run.
As far as I've seen none of the major companies are actually making money off of AI. The easiest ones to look at are Anthropic and OpenAI who are both haemoraging money.
Now, maybe there is some magic solution to increasing efficiency but as it is right now the sums just do not add up to me. The more people who use it, the more money it loses. In any other business model, that's absolute insane. OpenAI was losing money on people on the $200 a month tier.
As you point out, it's also rather telling how they talk about things.
This is a good example:
Maybe I'm just jaded from another technology hype machine.
First crypto was going to take over, then it was NFTs, then it was the Metaverse (lol), now we're onto AI.
I think AI is more useful than the other three to be clear, it actually has some purpose and depiste my scepticism it can be useful. I use it myself for work.
But until the sums start making sense, I fail to see how this works long term. LLMs aren't even close to AGI, and without massively increasing context windows (which costs more money) or some really insane efficiency savings I don't understand why VC companies would keep pumping money into it expecting zero returns.
If you want a way more in-depth skeptic point of view, I can recommend Edward Zitron. His writing style is... unique and I don't agree with everything he says but he goes into a lot of detail.
https://www.wheresyoured.at/th |
Ah ok — apologies, I think I misunderstood your skepticism previously.
I thought you were broadly saying that it couldn’t do basic tasks and that you and your colleagues were often running into brick walls with fairly elementary things — and as such, you doubted any real-world utility?
I agree that monetisation remains an open question, but I think that path is becoming clearer, at least for the major SaaS companies. If you’re Microsoft, for example — or any of the large horizontal enterprise SaaS players — they have multi-year contracts, decades of trust, and crucially, they already hold the customer data. At the same time, they have huge workforces and therefore (sorry to say) may well need fewer engineers if coding can increasingly be handled by AI models.
There might be some SMEs, as TownBlue suggests, who build more in-house — but the chances of large multinationals in highly regulated industries like finance or healthcare moving away from Microsoft or ServiceNow anytime soon seem close to zero.
The other point on monetisation is that people tend to focus on the hyperscalers — but you can already see real-world productivity gains among AI adopters showing up in the US economy.
I absolutely agree, though, that the physical constraints — the finite resources AI requires — are going to be a huge issue. It probably leads, at least initially, to an inflationary impulse as commodity and power prices rise. Whether that’s eventually offset by a deflationary productivity effect remains to be seen. There’s also likely to be fairly significant political pushback if utility bills rise to support technology that may ultimately displace jobs.
Anyway, if you’ll excuse me — I’m off to register for the nearest plumbing course. | | | |
| |
By continuing to use the site, you agree to our use of cookies and to abide by our Terms and Conditions.
