This is happening though. The AI foundries are rapidly expanding into new businesses.
I opened ChatGPT and today it offered me “Finances” which would allow me to connect my financial data with ChatGPT and it could help me manage my subscriptions, etc.
Right now they want to improve their models. That means still getting a lot of data from around the world. Companies wouldn’t give them that data if they thought they might take it and put them out of business.
The smart thing to do would indeed be to pretend to be a service provider until you have enough information that you don’t really need 3rd party data anymore, and then to start putting other companies out of business.
I suspect these companies are less than 1-2 years away from creating a DoorDash/Uber Eats competitor product.
I think AI is more like the fundamental layer kind of changing, that reshape the values, rules and perspectives in modern society. Examples: inventions of paper, gunpowder, train, airplane, internet, and now AI. No body can claim the ownership of this change, and everybody can participate in some ways, selling it, using it etc.
If there were only one firm, then it would make sense to not sell inference. However OpenAI, Anthropic, Google are all in competition with each other to earn market share, developer goodwill, and to get reliable growth, selling inference is the best strategy, rather than making a thousand companies and needing to do the marketing, branding, differentiation, PR, etc for each one seperately.
Also, the AI companies learn a lot from selling their inference. They see what markets are growing fastest, what people are using AI for, information benefitted by the startups using their models putting way more effort into distribution than they are. With enough information, the model companies can then create a generic AI-enabled product (like they have over the past year) that factors in all the learnings the startups have undergone.
AI hype is real, and there is a lot of perverse incentive to keep the hype cycle going. You are right about that.
Rest of your analogy/arguments are just a lot of nonsense too.
Electricity is valuable, and by that logic Siemens Energy should build and own all factories under the sun. TI should stop selling their calculators and compete in Math Olympiad instead!
afaik Siemens Energy never pushed the argument that electricity would be able to do everything humans can within a year or so...
Yes, obviously there's a whole lot of relationship building and effort and judgement which matters at least as much as infrastructure in building businesses, but the whole schtick of Big AI is to insist their product is better than humans at planning and market analysis and building and sales and marketing and especially at effort and concentration, and they're obviously better capitalised than startups using AI to do something in a particular vertical, so if they believe their own hype, why not?
No, but it would mean that they were overhyping how useful electricity was to and it was worth pointing out that if they actually believed what they said, they'd be trying to do everything they said they could do rather than just being a utility company...
Because it turns out capital markets are actually reasonably efficient and throwing more "brains" at them yields steeply diminishing returns in increased allocative efficiency, compared to deploying those same "brains" as capital to produce output commodities?
More like a super expensive version of Factorio for engineers. Everybody I know has a software factory now, and a whole bunch of software nobody wants :)
You still need solid motivation, a good sense of the fundamentals, communication skills, social skills, sales skills, and good taste to build good products with AI.
Also, I think you are conflating two things:
1. The genuine advancement (and challenges) AI brings.
2. The inevitable hordes of less scrupulous people who take advantage of every new trend to make money however they can.
I am fully immersed in building with AI and I barely notice these less scrupulous. I know they exist. They always exist. But why are you paying so much attention to them?
Are shovels just another get rich quick scheme? If they were really all that, they wouldn't be selling shovels to the public. They would be forming corporate partnerships, digging dirt and conquering markets--starting with the mining industry.
So if I am following this correctly, NVIDIA is todays shovel sellsmen?
Others, not miners, also made a lot of money. Personally I recommend investing in AI but not LLMs. I expect John Deere to make a lot of money implementing more AI in farm equipment.
the ai companies have a problem at the moment: the compute required for inference (especially) costs a metric shit-ton of money - so they're crowd-sourcing the funds (unsuccessfully so far, tbh - for example, a $20/m claude sub that i'm on for work, and I regularly see individual conversations where the tokens cost more than that - even up to $350.
It's a tool tho, not a panacea. With the right guidance, for example, you can get some good code results.
Catch 22 --- In order to pay for itself, AI needs to eliminate the very jobs that build the experience and expertise necessary to provide the "right guidance".
At the same time, this will restrict production of quality content that is needed for model training.
In other words, AI creates a self degrading/defeating feed back loop.
Absolutely. At its core, AI is just another algorithm that is subject to the same 'Garbage In-Garbage Out' principle. As more and more of the data used to train the models, is itself generated by AI; the feedback loop you describe gets stronger.
This is happening though. The AI foundries are rapidly expanding into new businesses.
I opened ChatGPT and today it offered me “Finances” which would allow me to connect my financial data with ChatGPT and it could help me manage my subscriptions, etc.
Right now they want to improve their models. That means still getting a lot of data from around the world. Companies wouldn’t give them that data if they thought they might take it and put them out of business.
The smart thing to do would indeed be to pretend to be a service provider until you have enough information that you don’t really need 3rd party data anymore, and then to start putting other companies out of business.
I suspect these companies are less than 1-2 years away from creating a DoorDash/Uber Eats competitor product.
I think AI is more like the fundamental layer kind of changing, that reshape the values, rules and perspectives in modern society. Examples: inventions of paper, gunpowder, train, airplane, internet, and now AI. No body can claim the ownership of this change, and everybody can participate in some ways, selling it, using it etc.
If there were only one firm, then it would make sense to not sell inference. However OpenAI, Anthropic, Google are all in competition with each other to earn market share, developer goodwill, and to get reliable growth, selling inference is the best strategy, rather than making a thousand companies and needing to do the marketing, branding, differentiation, PR, etc for each one seperately.
Also, the AI companies learn a lot from selling their inference. They see what markets are growing fastest, what people are using AI for, information benefitted by the startups using their models putting way more effort into distribution than they are. With enough information, the model companies can then create a generic AI-enabled product (like they have over the past year) that factors in all the learnings the startups have undergone.
AI hype is real, and there is a lot of perverse incentive to keep the hype cycle going. You are right about that.
Rest of your analogy/arguments are just a lot of nonsense too.
Electricity is valuable, and by that logic Siemens Energy should build and own all factories under the sun. TI should stop selling their calculators and compete in Math Olympiad instead!
afaik Siemens Energy never pushed the argument that electricity would be able to do everything humans can within a year or so...
Yes, obviously there's a whole lot of relationship building and effort and judgement which matters at least as much as infrastructure in building businesses, but the whole schtick of Big AI is to insist their product is better than humans at planning and market analysis and building and sales and marketing and especially at effort and concentration, and they're obviously better capitalised than startups using AI to do something in a particular vertical, so if they believe their own hype, why not?
> afaik Siemens Energy never pushed the argument that electricity would be able to do everything humans can within a year or so...
If they had pushed that argument, would it have changed how useful electricity is?
No, but it would mean that they were overhyping how useful electricity was to and it was worth pointing out that if they actually believed what they said, they'd be trying to do everything they said they could do rather than just being a utility company...
TI should stop selling their calculators and compete in Math Olympiad instead!
Can they make more money doing this? I don't think so --- and apparently, they don't either.
Finance and equity markets would be the first place I would look to apply "super intelligence" for huge, quick cconomic returns.
Instead, they peddle "tokens". Wonder why?
Redbull gives you wings.
Because it turns out capital markets are actually reasonably efficient and throwing more "brains" at them yields steeply diminishing returns in increased allocative efficiency, compared to deploying those same "brains" as capital to produce output commodities?
Ever hear of arbitrage and hedge funds?
These companies exist specifically to identify and take advantage of aberrations and inefficiencies in capital markets.
This is something I would expect "super intelligence" to excel at.
https://en.wikipedia.org/wiki/List_of_hedge_funds
More like a super expensive version of Factorio for engineers. Everybody I know has a software factory now, and a whole bunch of software nobody wants :)
The only person getting rich is a Taiwanese businessman in leather jacket.
You still need solid motivation, a good sense of the fundamentals, communication skills, social skills, sales skills, and good taste to build good products with AI.
Also, I think you are conflating two things:
1. The genuine advancement (and challenges) AI brings.
2. The inevitable hordes of less scrupulous people who take advantage of every new trend to make money however they can.
I am fully immersed in building with AI and I barely notice these less scrupulous. I know they exist. They always exist. But why are you paying so much attention to them?
Are shovels just another get rich quick scheme? If they were really all that, they wouldn't be selling shovels to the public. They would be forming corporate partnerships, digging dirt and conquering markets--starting with the mining industry.
I mean… the origin of that analogy is selling shovels to gold miners in the 1849 gold rush. Most of those miners found absolutely nothing.
So if I am following this correctly, NVIDIA is todays shovel sellsmen?
Others, not miners, also made a lot of money. Personally I recommend investing in AI but not LLMs. I expect John Deere to make a lot of money implementing more AI in farm equipment.
NVIDIA is one type of shovel salesman, yes. So are all the big AI companies. And all the other companies we call “hyperscalers”.
The problem is that they have to spend so much time and money inventing new types of mines, for which they need to be the shovel salesmen.
They’re also called “snake oil salesmen”. Or “Robber Barons”.
the ai companies have a problem at the moment: the compute required for inference (especially) costs a metric shit-ton of money - so they're crowd-sourcing the funds (unsuccessfully so far, tbh - for example, a $20/m claude sub that i'm on for work, and I regularly see individual conversations where the tokens cost more than that - even up to $350.
It's a tool tho, not a panacea. With the right guidance, for example, you can get some good code results.
With the right guidance
Catch 22 --- In order to pay for itself, AI needs to eliminate the very jobs that build the experience and expertise necessary to provide the "right guidance".
At the same time, this will restrict production of quality content that is needed for model training.
In other words, AI creates a self degrading/defeating feed back loop.
Absolutely. At its core, AI is just another algorithm that is subject to the same 'Garbage In-Garbage Out' principle. As more and more of the data used to train the models, is itself generated by AI; the feedback loop you describe gets stronger.