Whenever I have an idea, I talk to an LLM. But I don’t ask it to tell me whether the idea is good. Instead, I give it the half-baked thought and ask it to grill me. One question at a time.
It asks what I mean, and I explain myself. When it spots an assumption or pushes on a contradiction, I explain my reasoning or disagree. Sometimes it misunderstands me and I correct it. Sometimes it asks a question that sends me somewhere I hadn’t expected. Eventually, I know what I think, and only then do I worry about the next steps.
Lately, this has become my default way of developing ideas. And it made me wonder whether talking to LLMs could replace writing as a way to clarify our thinking. But that question treats writing and conversation as competing media. So instead, I asked whether thought develops as monologue or dialogue.
Why dialogue clarifies thought
Writing turns something vague in your head into words you can examine. Explaining what you mean exposes questions, and those questions reveal assumptions or contradictions. The thought becomes something you can look at rather than something floating around in your head. Once externalized, it can push back. You read a sentence and realize it doesn’t say what you meant, or that an argument doesn’t follow. Something you wrote yesterday gnaws at you in the shower, or a connection appears that you couldn’t see while the idea existed only in your head.
But writing has never been our only way to clarify thought. Philosophers debated and scientists corresponded. Students learned from teachers, while writers exchanged letters and colleagues argued over ideas. We have always used other minds to clarify our own. Conversation can externalize thought too. So why don’t we bring every half-interesting thought to another person? For one, their attention is scarce. Imagine having access to a brilliant thinker whenever you have an idea. You would probably work through the obvious problems first and bring it to them only when it seemed sufficiently promising, because asking for an hour of their attention means asking them not to spend it elsewhere.
Scarcity creates a preparation threshold. But LLMs remove much of that threshold, allowing me to bring them an idea ridiculously early. I don’t have to know whether it is worth pursuing because that is precisely what the conversation can help me discover. If it goes nowhere after ten minutes, I can abandon it without wasting anyone’s afternoon.
Go as fast as you can
You decide the bitrate of your thoughts when talking to an LLM. When I’m exploring something I understand well, I might write several paragraphs before pressing Enter because I want the LLM to push against a developed position. On the other hand, when my thought is fragile, I can go one statement at a time.
I can also decide what kind of response I want. Don’t give me ideas. Follow my reasoning and point out ambiguity. Or: challenge every assumption, look for contradictions, give me the strongest counterargument. I might ask which parts wouldn’t make sense to someone outside my field. Sometimes I simply say: grill me, one question at a time.
That last instruction gives the LLM considerable freedom to choose which thread to pull next. But in another conversation, I might rein it in and ask it to follow wherever I lead. An LLM can act as a teacher or an editor. It can be a debate partner, a critic or something closer to a notebook that talks back. Choosing the role changes how the conversation develops the thought.
A responsive page
A notebook gives us a surface for thinking. It lets us capture an ephemeral thought before it disappears and leaves us with a record we can return to later. We can stop halfway through an idea, sleep on it and resume tomorrow. But an LLM makes that surface responsive.
It can notice that I’m being vague when I don’t notice it myself. It can question an assumption I didn’t realize I was making or expose me to a perspective outside my experience. Still, I don’t have to accept any of it. When an interpretation doesn’t represent what I think, I reject it. When the LLM misunderstands me, I correct it. The LLM participates in the exploration while I remain responsible for the conclusion.
That responsibility requires attention. LLMs can steer a conversation through their own framing and turn messy thoughts into suspiciously coherent arguments. People can do the same, so you need to notice when a thinking partner’s influence quietly becomes your conclusion.
Explore before you judge
The biggest change for me is how many ideas I can afford to explore. Ideas used to compete for a limited number of evenings, forcing me to decide which half-formed observation deserved a chance. That is an awkward moment to choose because it is precisely when I know the least about the idea.
But now I can explore almost all of them. Some conversations last hours, while others stop after ten minutes or sit untouched for days. I can have several ideas developing in parallel and return to whichever one I want to pursue. Cheap exploration lets me postpone judgment: I can develop ideas far enough to discover which ones deserve to continue.
And an abandoned idea isn’t necessarily wasted effort. One observation might survive, or a question could resurface six months later. A dead end can collide with another thought and produce something more interesting than either of them. Keeping the conversation means keeping the path, including its unresolved edges.
This also separates the thinking artifact from the publishing artifact. The conversation can keep its false starts and discarded arguments because it records the thought being developed. The artifact has a different job: to make that thinking accessible to someone else, as an essay, picture, or an app. It strips away scaffolding and repetition to give the consumer a path through something that took me hours to explore. I can keep both and return to the messy version when I want to continue thinking.
A bicycle for thought
Steve Jobs famously described the personal computer as a “bicycle for the mind.” The metaphor came from a comparison of locomotion efficiency: humans weren’t particularly impressive until you put a person on a bicycle. The bicycle amplified what a person could do.
LLMs offer something similar for thought. Used as answer machines, they give us a response that can save us from thinking further. But when we bring them an unfinished thought, they can question us. We can correct them and change the pace. We can take back the steering when the conversation wanders, stop when we’ve had enough and come back tomorrow.
The most important thing LLMs democratize is access to dialogue. Great thinkers have always had other minds to push against: teachers, peers, rivals, editors. What was scarce was access to a capable interlocutor with the knowledge and patience to explore an unfinished thought with you.
This article started with a question about whether LLMs could replace writing as a means of clarifying thought. After talking about solitude, influence, memory and the economics of other people’s attention, I ended up somewhere else. Writing wasn’t the thing I wanted to replace. I wanted someone to talk back.