Comparing how traditional tech companies scale to how frontier AI companies scale doesn't seem very fair to me. AI is still relatively new territory, in comparison traditional software companies benefit from decades of investment in cloud infrastructure, networking, and hardware. I'm certain those technologies also went through periods of heavy investment before they became scalable.

The author also talks about advances like smartphones as though they arrived without significant challenges or trade-offs. Every major technological shift has come with some cost, e.g environmental, shortages, investments.

I'd argue most things are difficult to scale in the beginning. Just because today's LLMs are expensive and resource intensive doesn't mean the technology is fundamentally flawed, it may simply mean the current approach isn't the one we'll end up with. But naturally, someone working on another approach would have something to criticise about others.

Finally, I'm not convinced by the claim that we're "stuck" with LLMs just because they're heavily marketed. They create value for many people, which is why they've been adopted so quickly. As the author points out, investors care about economics. I'm sure people would listen if someone developed a cheaper/more sustainable technology that delivered the same value.

I think LLMs, with only a few exceptions, have mostly created the illusion of value. It’s undeniable I can write code, troubleshoot and document much faster than in the pre LLM era, but putting a dollar value to that it’s not particularly easy. One group of people being 100x faster at one task doesn’t mean the organization as a whole runs at 100x the speed, in fact in most cases where the bottle neck is somewhere else, the productivity gains of the 100x group have little to no impact, or even a negative impact in some cases due to token overspending.

Right, if you came out with a technology that worked just as well as LLMs at half the power people/companies would jump at it from the absolutely massive savings in hardware and power. AI companies are looking at every other method they can, and inventing new ones, none of them have worked as well as the transformer so far.

"They create value for many people"

They create value for many people because these people are not paying the cost of the tool;

> They create value for many people, which is why they've been adopted so quickly.

A lot of that adoption is completely useless crap tho. Like ai in vacuum cleaner that ads absolutely nothing useful to anything. AI buttons intentionally at places I randomly click at, so that I am forced to open it. Google search that defaults to ai, so that we have to use it after then nerfed real search.

There are useful usages of LLMs. But huge bulk of the adoption is companies realizing they wont get investors money if they dont add ai button, it does not have to be useful.