The Intelligence in the Prompt
The Intelligence in the Prompt
One unappreciated benefit of LLMs is that people think more clearly when they write their ideas down.
As a scientist, I have more than 60 notebooks filled with thoughts, sketches, questions, hypotheses, calculations, and half-baked ideas. Those notebooks were crucial to the theoretical development of my discoveries. They weren’t merely records of conclusions I had already reached. They were part of the thinking process itself.
We’ve long known this about notebooks. If you have a vague idea, a problem you’re wrestling with, or a decision you’re trying to make, simply writing it out often helps. The act of turning fuzzy intuitions into sentences forces you to clarify what you actually mean. Half the battle is often just getting the thought out of your head and into words.
As I have said before, LLMs are basically notebooks that talk back.
When you’re describing your thoughts, your theory, your project, or your dilemma to an LLM, you’re also doing exactly what people do when journaling or filling pages in a notebook. You’re writing down your thoughts. You’re organizing them. You’re making assumptions explicit. You’re identifying gaps. You’re discovering what the real question is.
Very often, that process itself is what is helpful.
In many cases, before the LLM has even responded, the user already has a clearer understanding of the issue simply because they had to formulate it into a prompt. The prompt-writing process is itself a thinking process.
This is also one of the profound differences between LLMs and search engines. When we use Google, we typically enter a few keywords. Often not even a sentence. The goal is to locate information that already exists somewhere else.
With an LLM, however, we often find ourselves writing paragraphs. We explain our dilemma, our project, our half-formed theory, our objections, and our intuitions. We provide context. We tell a story.
In doing so, we are engaging in a form of reflection that search engines never demanded of us. The prompt itself becomes part of the cognitive work.
This may be one reason LLMs feel so different from search. They are not merely information-retrieval systems. They are systems that encourage users to externalize their thoughts. And externalized thoughts are often clearer thoughts.
And when the response comes back, it’s frequently not some entirely foreign intelligence delivering profound new insights. Often it’s a cleaned-up, better-organized version of what the user already wrote. The substance may have largely been present in the prompt itself. The LLM helps structure it, summarize it, clarify it, and sometimes extend it.
The entire process is a way of leveraging more thinking power out of the person.
Part of that comes from whatever substantive facts, principles, analogies, or corrections the LLM contributes. But part of it comes from something more basic: the LLM creates a setting in which people are continually externalizing their thoughts.
In this sense, Google and LLMs are doing somewhat different things. Google primarily helps us access humanity’s stored knowledge. LLMs certainly do some of that as well. But they also help us access something else: knowledge that is already present, but only implicitly, within ourselves.
They help turn latent thoughts into explicit thoughts.
That may be one reason people often leave an LLM conversation feeling that they themselves did a surprising amount of the intellectual work. In many cases, they did. The system was not merely answering questions. It was eliciting thought.
A notebook helps you think because writing helps you think.
An LLM helps you think because writing helps you think — and because the notebook talks back.
The result is not merely artificial intelligence augmenting human intelligence. It is a system that often helps people extract more intelligence from themselves than they otherwise would have.