Thinking Improves Thinking

How did we do it before ChatGPT? How did we write full sentences, connect ideas into a coherent arc, solve problems that had no obvious answer? We thought. That's it. We simply sat with discomfort long enough for something to emerge.

I find this fascinating. You have a problem, so you sit down and think until you find a solution. Sometimes you're not even sitting down. You go for a walk, and your mind quietly wrestles with the idea while your feet carry you nowhere in particular. A solution emerges not because you forced it, but because you thought it through. What happened in that moment is remarkable: new information was created from the collision of existing ideas inside your head. No prompt. No query. Just you.

I remember the hours I used to spend debugging a particularly stubborn problem at work. I would stare at the screen, type a few keystrokes, then delete them. I'd meet with our lead engineer and we would talk in circles. At home, I would lie in bed still turning the problem over. And then one night, somewhere around 3 a.m., I dreamt I was running the compiler, making a small change, watching it build, and suddenly it worked. I woke up knowing the answer before I had even tested it. I had to wait until morning to confirm what my sleeping mind had already solved.

That's the mind doing what it was built to do.

Writers know this feeling too. A sentence that won't cooperate in the afternoon sometimes writes itself during a morning shower. Scientists have described waking up with the solution to a problem they fell asleep wrestling with. Mendeleev wrote in his dairy that he saw the periodic table in a dream. The mind that keeps working when we stop forcing it.

The mind can generate new ideas from its own reflection, something we routinely accuse large language models of being incapable of. LLMs recombine what already exists; the human mind makes unexpected leaps. But increasingly, it feels as though we are outsourcing those leaps before we ever attempt them. Why sit with a half-formed thought when you can just ask? Why let an idea marinate when a tool can hand you something polished in seconds?

if (Math.random() > .4) {
    take_a_leap()
}

I won't say that AI is making us lazy, because it isn't laziness to reach for the calculator to do long divisions. However, we are slowly forgetting what it felt like to think hard. And soon we will stop believing that we are capable of it.

Thinking improves thinking.

The mind is like any muscle. Leave it unstrained and it weakens. Push it and it grows. The best ideas you will ever have are still inside you, waiting for the particular silence that only comes when you stop reaching for your phone.

In the age of AI, the most radical thing you can do might simply be to think.

---