There are no correct applications. Even if you accept the dubious claim that its good at coding it would not be good enough to justify the actual non-subsidized costs of using it. Its not just CEOs noting a lack of substantial productivity gains from LLM coding assistants, there are at least a few studies floating around that have examined it.
Unless you want programming to become nothing but rich MBAs paying Anthropic to turn diesel fuel and your local power grid into vibe coded slop, you should be against AI coding on principle alone.
Agree, the current market is warping the value argument.
I am specifically arguing against the use of AI by MBAs. It’s when it is used by the unqualified, when it is at its most problematic.
The same thing happened in accounting when it first started getting computerised. Suddenly a bookkeeper had the power to affect not just the transactions they were processing, but the entire accounting system. It also happened with managers poking around in systems they did not have proper expertise to use. The solution was not to ban the use of computers in accounting. It was introducing systems and controls around them.
I recognize that; what I’m arguing that even when used by qualified professionals (if such a thing could even be said to exist for something so new) the value proposition still doesn’t make sense when you take the real costs into account.
Progress on AI has bottle-necked and the status quo is unsustainable. They can’t make them cheaper without making them worse, and they can’t make them better without making them exponentially more expensive. If AI continues to exist as a coding tool, it will become prohibitively expensive once capital investment starts realizing there’s no other path to profitability. Then, since programmers don’t control the budgets and don’t get paid well enough to afford unsubsidized AI, it will be made to cater to those that do; becoming a tool for them exclusively.
The process has already started: Sam Altman trying to take your RAM, NVIDIA and AMD shifting focus from consumer GPU products. If nobody can run local models, then they have no other option but to rent.
This is why I think If you care at all about the quality of software or programming as a hobby/profession you should be against AI entirely.
If not for the AI inflated cost of hardware. If you wind the clock back on products for 2 years ago that Local hardware can run competent coding and general analysis workflows quite affordably. There are tens of thousands of people and small companies that do this. A setup with 2x 5090s can run a decent model for a small team for $4000 investment. The same hardware today costs $16000, if you can find it.
It’s unsustainable because of the money and hardware black hole that is the Capital markets and their capture of the tech industry. Throw in the Memory Mafia cutting supply again on top of that. They need to die for AI to become sustainable for those who don’t already have hardware capable enough.
I would say however that 1 year ago, that would not be true for local models. I think they reached that point for many people about 6 months ago. When they bypassed Claude and ChatGPT models from about 6 months before. The tooling and harnesses have also improved significantly over the same period.
Ironically, capital markets are burning cash and capital at a massive rate to inflate perceived demand. Ironically producing very little because every point of the supply chain is milking it for maximum exploitation for the least value.
The cost of setting up these tin shack AI data centres is inflated far beyond what proper dataceters cost just a few years ago. The shovel and sack makers are bleeding venture capitalists dry.
If the bubble pops, they will go elsewhere and demand and supply will go back to levels where AI is sustainable, but not with the companies currently running the show. I wish them a quick death, taking as many venture capitalists along with them as possible.
If they’re “productivity enhancers”, why have literally 1000s of CEOs admitted absolutely zero efficiencies or increased revenue due to forced AI usage in the workplace?
Because CEOs are stupid, narcissistic and don’t understand how AI works. Forced use of anything has never resulted in improvements.
Better to ask the highly skilled professionals who use it effectively and for the correct applications.
There are no correct applications. Even if you accept the dubious claim that its good at coding it would not be good enough to justify the actual non-subsidized costs of using it. Its not just CEOs noting a lack of substantial productivity gains from LLM coding assistants, there are at least a few studies floating around that have examined it.
Unless you want programming to become nothing but rich MBAs paying Anthropic to turn diesel fuel and your local power grid into vibe coded slop, you should be against AI coding on principle alone.
Agree, the current market is warping the value argument.
I am specifically arguing against the use of AI by MBAs. It’s when it is used by the unqualified, when it is at its most problematic.
The same thing happened in accounting when it first started getting computerised. Suddenly a bookkeeper had the power to affect not just the transactions they were processing, but the entire accounting system. It also happened with managers poking around in systems they did not have proper expertise to use. The solution was not to ban the use of computers in accounting. It was introducing systems and controls around them.
https://en.wikipedia.org/wiki/Productivity_paradox
I recognize that; what I’m arguing that even when used by qualified professionals (if such a thing could even be said to exist for something so new) the value proposition still doesn’t make sense when you take the real costs into account.
Progress on AI has bottle-necked and the status quo is unsustainable. They can’t make them cheaper without making them worse, and they can’t make them better without making them exponentially more expensive. If AI continues to exist as a coding tool, it will become prohibitively expensive once capital investment starts realizing there’s no other path to profitability. Then, since programmers don’t control the budgets and don’t get paid well enough to afford unsubsidized AI, it will be made to cater to those that do; becoming a tool for them exclusively.
The process has already started: Sam Altman trying to take your RAM, NVIDIA and AMD shifting focus from consumer GPU products. If nobody can run local models, then they have no other option but to rent.
This is why I think If you care at all about the quality of software or programming as a hobby/profession you should be against AI entirely.
If not for the AI inflated cost of hardware. If you wind the clock back on products for 2 years ago that Local hardware can run competent coding and general analysis workflows quite affordably. There are tens of thousands of people and small companies that do this. A setup with 2x 5090s can run a decent model for a small team for $4000 investment. The same hardware today costs $16000, if you can find it.
It’s unsustainable because of the money and hardware black hole that is the Capital markets and their capture of the tech industry. Throw in the Memory Mafia cutting supply again on top of that. They need to die for AI to become sustainable for those who don’t already have hardware capable enough.
I would say however that 1 year ago, that would not be true for local models. I think they reached that point for many people about 6 months ago. When they bypassed Claude and ChatGPT models from about 6 months before. The tooling and harnesses have also improved significantly over the same period.
Ironically, capital markets are burning cash and capital at a massive rate to inflate perceived demand. Ironically producing very little because every point of the supply chain is milking it for maximum exploitation for the least value.
The cost of setting up these tin shack AI data centres is inflated far beyond what proper dataceters cost just a few years ago. The shovel and sack makers are bleeding venture capitalists dry.
If the bubble pops, they will go elsewhere and demand and supply will go back to levels where AI is sustainable, but not with the companies currently running the show. I wish them a quick death, taking as many venture capitalists along with them as possible.