Yes, you are the problem. You should know this by now.
The arrival of LLM solutions allowed everyone with a phone to interface with tremendous, extremely complex technology. It was a moment for the ages where decades of research, mathematical genius and terminal nerdiness were finally at the fingertips of anyone, to build anything, from anywhere.
Programming? Creating images? Having a personal, 24/7 guide to anything in human knowledge? Check all those. Spectacularly. The power of LLM-powered Gen AI tools didn't rock the boat, it blasted off into orbit in a titanic-sized rocket ship to take humanity to a universe of ingenuity. It left anyone who still thought boats were a good idea confused, betrayed, and angry.
If it's so good, why am I writing about it? Because this extraordinary power is on course to derail spectacularly and go very, very wrong. And it's your fault.
There are two sides to AI's people problem: cost and human behaviour.
First, LLMs are expensive. I know it runs in your browser and makes adorable pictures of you as a cat, or Studio Ghibli character, or doll in a package, and your auntie sent you one and it was just so precious. The amount of compute time required to turn a single selfie into a bright, anime‑style caricature is monstrous. Auntie Bertha's precious makeover meant a top‑tier GPU was kept busy up to five seconds. That's just about FIFTEEN TRILLION CALCULATIONS.
The one picture just used the same compute budget as analyzing and summarizing about 20,000 words of text. That's The Old Man and the Sea, or the entire Wikipedia corpus on penguins. It's enough compute power to read, catalogue, and reply to 120 average-length work emails.
That's just running the machine that uses an LLM to make cute stuff. Training a frontier‑class large language model such as GPT‑4.5 or xAI’s Grok requires eye‑watering resources. We're talking of a range from mid-hundreds of millions up to HALF A BILLION DOLLARS to house, train, fine-tune, and deploy the same beast you just asked to turn you into a Dragonball character.
The second problem, well it's people. We here, we're all distinguished, professional synergistic technical genius visionary philanthropists of the AI revolution. None of my words apply to present company, of course. But people, they gonna people.
This is industrial-scale compute firepower being thrown at the likes of "how can I get Josh to like me" and "write my homework about China's exports". We're talking about wasting resources that could, I am sure, take a good step towards curing cancer. Engineering crops for hostile environments. Figuring out cold fusion once and for all. You know, stupid stuff like that.
This is the same technology, the same investment it currently takes to keep you entertained.
One of the current limiting factors of frontier AI development isn't really technological, or related to software. We have quite simply reached the absolute limit of our electrical engineering capacity, and cannot figure out how to cram any more computers into a single room without melting the walls. This is real, look it up.
But don't worry, we'll figure it out so you can keep optimizing those Tinder profiles. So you can keep discovering what you'd look like standing next to Taylor Swift.
Things, naturally, are a lot more complicated than sneering at tragicomic reality. In a way, the massive attention dedicated to the messianic arrival of Gen AI and LLMs is an absolutely necessary step towards its acceptance and implementation.
We are in the baby steps phase, the birth of an age of automation and sophisticated tools like we've never even been able to dream of before. And if public acceptance of a world-shaking technological paradigm means melting a couple of data centres down to the core of the planet, so be it.
Maybe, however, it's time to focus on the real problem here. While people out there are being people, and using ChatGPT to cheat on school assignments and make themselves look tanned, what are we, the AI natives, doing?
We're frantically dogpiling on "vibe coding" and freaking out about our precious little words being stolen by evil cybernetic overlords. This is supposed to be our time. We have invested too much time, too many dreams into the coming of AI to let some creepy tech bro turn AI into a sideshow.
It's time we realize that we are people too, and that we completely fumbled the first pass at making this special.
We - the builders, the dreamers - should have known better. We were too busy being clever, or bitter, or distracted to build the foundations of what the world needs. AI-driven tools are powerful enough to interface with us as partners and companions, and could become spectacular force multipliers for all our endeavours.
It's time to step up and make sure AI isn't reduced to an expensive toy. We need to decide what we want it to become. We need to set the tone and present the world we do have with positive examples of its near-endless automation and reasoning possibilities. Then, we need to teach the world how to use it.
Also, look at how cute I am as an EVA.
