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!
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.
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.
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!
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?
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.
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.