introduction
I think it’s funny how the very paper which gave us the modern transformer architecture produced the tool (or collection of tools) most likely to erode in us humans the very thing that its title describes.
This is a blog post about AI, but not in the traditional sense. I’m not interested at all in walking you through what transformers are or how their attention mechanisms work. What I do want to talk about is smaller, and more personal: what happens to your attention once you let a machine hold it for you, and how you can try and keep some of it for yourself.
the purists
When considering the question of “how do i keep my attention”, the purist’s answer is to just never let an LLM (or similar AI tool) anywhere near it. Each and every task gets done manually, and the manual labour is seen as proof of rigour. The problem with this way of thinking is when it gets generalized to everything, including tasks that never required it. There is no pride to be had in writing the same config file 50 times by hand, or in manually drafting content that you’ve already brainstormed extensively beforehand. Refusing to use a tool like AI here on the basis of “not outsourcing thinking” isn’t actually discipline. It’s a purely performative stance dressed up as the righteous and honourable choice.
The cost of this way of thinking is that the purist’s attention never actually gets tested on anything genuinely hard, since it’s already being spent (and wasted) on trivial tasks. Giving every task the same manual effort means you spend a whole lot of time doing busywork, wondering why people who seem to “cheat” their way through everything (or in other words, appropriately leverage an available tool for its intended use case) are always ahead of you.
the outsourcers
The opposite failure is much quieter, but honestly a lot more dangerous. The outsourcer is often blissfully unaware that they are actually an outsourcer. “I’ll just let Claude handle this boilerplate” turns into “I’ll just let Claude handle the auth logic” which turns into “I’ll just let Claude delete the databa–oops.”
The problem is that the outsourcer doesn’t just hand off the grunt work, they turn off their brain completely. They let a model decide which approach is correct, what the tradeoff is, and whether a generated answer actually makes any sense at all. From their point of view, this looks identical to just asking for help with boilerplate.
What makes this approach so dangerous is that it works (initially). If you skip the gym once, nothing really changes. You’re just as strong as you were the day before (and not as sore!), and so the decision to skip looks like the correct one. But over time, this absence of immediate consequences allows the skipped day to turn into a skipped week, month and year. “I’m just too busy”, you tell yourself. “Those gym bros are obsessed and have no life”, you tell yourself. Eventually, all your strength is gone, and you have only yourself to blame.
If you put a frog into boiling water, it’ll jump out immediately. But if you place the same frog in cold water and slowly bring it to a boil, it will be cooked to death.
where the line is
At the end of the day, AI is here to stay. I’m not going to pretend like I don’t have my own grievances with it; the environmental costs are genuinely troubling to hear about when we’re already behind on fighting climate change. There’s also valid moral reasons to be against it, particularly when it comes to more creative fields like writing and art.
But putting your head in the sand and hoping that AI blows over is, quite frankly, stupid. A good rule of thumb to have when deciding whether or not to reach for it is to ask yourself, “Could I explain or defend what the model just produced if someone put me on the spot right now?”
If the answer is no, then you have to understand that what you’re doing is akin to taking on credit card debt. Eventually, the bill is going to come due, and you probably won’t be able to pay it.
If the answer is yes, great! You successfully used a tool for its intended purpose, while keeping your own thinking intact.
Attention, as it turns out, is all you really need.