LLMs as a tool for thought
Take a breath with me.
As you breathe in,
think about the air rushing into your lungs.
The oxygen from the air diffuses into your bloodstreams, is used for cellular respiration, and is transformed into carbon dioxide.
As you breathe out,
think about the carbon dioxide rushing out of your lungs.
This cycle of
taking something in,
transforming it, and
sending something out
is pervasive in our lives.
Think about how a painter works.
First, they wait and look and listen until they’re inspired to work. They step up to the canvas and make a few brushstrokes.
Then they step back and look, taking in the painting and planning their next step.
After a moment, they again step towards the canvas and make more brushstrokes.
And repeat. This cycle of
taking something in (observing),
transforming it (planning), and
sending something out (painting).
Think about how conversations work.
As someone is speaking, we’re listening
and planning a response.
After a moment, we respond.
And back and forth we go.
What if we applied this cycle to LLM chatbots? I think most people use chatbots only as a way to steer somewhere. Answer a question, edit some text, write a function, explain a concept, etc. I have, historically. But, more and more, I’m realizing that LLMs can be a great tool for thought. A wonderful brainstorming partner.
Instead of using them as a way to solve X, what happens if we use them the way we use conversation? Getting thoughts out of our heads, listening to their responses, and reacting to them? Like sketching while thinking — the focus isn’t on "creating something". Gaining distance from thoughts as a way to evolve them. A good tool for thought is an extension of your mind.
What might this look like? Let’s examine how to use an LLM chatbot to plan a piece of writing: a collection of concepts, arranged in a structure. Take this piece on Why Chatbots Are Not the Future as an example. There is a main thesis (chatbots aren’t the ultimate UI) that is reinforced by a few main points:
That post started out as a mess of thoughts in my head. I had to figure out what concepts made sense together, decide on a general structure, and organize relevant thoughts into a supporting shape. Writing is hard!
How can we offload some of that work we usually do in our own heads?
Preserving thoughts outside our working memory.
We can only fit so many thoughts
into our heads at once. As a new parent, I have a smaller window than most. Outside of this
window of attention, thoughts getting increasingly fuzzy, losing their shape and dispersing.
But!
When talking through an idea with an LLM chatbot, I can trust that it will remember all of the
context. I’m happy to throw an idea out there and quickly move on to the next one, knowing that it
will be waiting for me.
Practically, this often looks like me saying something like "pin this idea and this other idea for later". And later, I’ll ask the chatbot for a list of ideas I’ve pinned. Sometimes this is even weeks later!
Think about how different forms of dialogue have different cadences. In person, we trade thoughts rapidly, barely a breath between exchanges.
We text at a more measured pace, with minutes or hours between responses.
Emailing stretches even further, with thoughts developing over days.
LLMs offer something unique here: they can match whatever cadence feels natural to our thinking.
Sometimes we need rapid exchanges to maintain a flow of thought. Other times, we need to step
back, reflect, and return hours or days later — and find our thought partner exactly where we left
them, context intact. The conversation picks up seamlessly, whether it’s been five minutes or five
days.
Suggesting thoughts.
A thesis is only as strong as its supporting points. While
ideating, an LLM chatbot can suggest related ideas to evolve a thesis (is this point robust?) and flesh out support for it (what is a better analogy?). My knowledge bank is limited
and focused on a narrow set of topics. It’s lovely to have an increased number and variety of
concepts to pull from.
Practically, this looks like me pausing while writing and asking the chatbot to list related ideas or to list analogies from different domains. Maybe even asking it to evaluate if those analogies strengthen or refute the point I’m trying to make.
Providing perspective and increasing the cycle speed.
Think of the painter: as
they’re close to the canvas, they’re focused on implementing. They’ve lost their view of the whole
picture. Having someone else standing back and giving feedback on the zoomed-out view
supercharges the take something in, transform it, and send something out cycle. In fact, I’ve seen
digital artists keep a zoomed-out view of their canvas open as they paint.
Practically, this looks like me asking about the "flow" of writing as I tweak it. This feels especially valuable, since it’s more laborious to see a "zoomed out" view of writing than a painting (we need to read individual words to get the shape of the whole piece, something I have many thoughts on and think LLMs x UI can improve).
I invite you to try it. Instead of using chatbots to steer toward specific answers, try using them as partners in that natural cycle of taking in, transforming, and sending out that drives creative thought. Not to get somewhere specific, but to participate in the endless exchange of ideas that moves thinking forward.
