The Case for Writing without LLMs

Back to meta-thinking: Why we still need to do the babbling and the pruning ourselves

When I write just for myself, it usually happens in the morning. I often wake up with strong intuitions originating from the dream world that I then have the urge to write down. In doing this, I engage with the challenge of verbalizing thoughts and feelings of non-verbal origin.

This task in itself is categorically impossible. Attempting to simplify such multidimensional inner experiences into mere text — although knowing I’m going to fail to capture them — has taught me something universal.

Writing, whether it’s just for ourselves or for others too, is an imperfect format. To acknowledge this is to also acknowledge what’s left out.

The task of translating my abstract notions into a verbal form is difficult, but I wouldn’t have it any other way. LLMs can’t read my mind— at least not yet — they can merely calculate an estimate of what I might be trying to say. So much gets lost if I outsource this process to next-token-predicting algorithms. Reflecting on which words to use remains essential for self-expression.

1. The Art of Finding the Right Words

People sometimes ask what language I think in. My reaction to this is one of confusion: Why would anyone think with words? That’s so inefficient!

Turns out I have an abstract, barely even visual way of conceptualizing meaning. This makes it particularly hard to verbalize what happens in my imagination. It’s frustrating when I can’t find the words that accurately represent what’s on my mind. As a result, my output is often far from satisfactory.

When we try to put our thoughts into words, sometimes these thoughts don’t have an even remotely verbal nature. This I call abstract-verbal translation.

Sometimes we have learned a concept through verbal communication. It has then been embedded into our thinking and taken a more abstract form — visual or otherwise. I call the process of retrieving these concepts from our memory and rewording them verbal-abstract-verbal translation.

1.1. Abstract-verbal translation

I have a history of experiencing lucid dreams where I communicate telepathically with dream characters. I have often tried to write these dreams down after waking up. It’s been literally impossible to verbalize the messages I’ve received. The feeling, the sensation behind the message still remains strong in my memory. I’ve been able to draw wisdom from my dream characters that I’m conscious of but not able to describe verbally, visually, or in any other way whatsoever.

Finding the words to describe our emotions might be a more common challenge. We can attempt to describe images that come to mind, or physical sensations associated with our feelings. In my experience, and according to people I’ve talked to, words often fall short. This can be frustrating, and understandably so. We might instead try different art forms that embrace multimodality, when one channel of communication isn’t sufficient.

1.2. Verbal-abstract-verbal translation

During my university exchange at the University of Iceland I attended a course by Donata Schoeller on Embodied Critical Thinking and Understanding, or ECTU. We did an exercise during class in which we practiced describing our processes of retrieving words from memory.

Each pair was given a piece of paper with a list of words on it. Our partner would then walk us through roughly the following steps:

1. Memorize the list of words for one minute.

2. Close your eyes.

3. Repeat as many words as you can remember.

4. Describe in as much detail as you can the process of retrieving, or trying to retrieve, these words from your memory. What’s happening in your mind?

I found this transformative. People had associated sensations, emotions, images, sounds — any modality you could think of — into these words. It was surprisingly hard to describe what was happening in our minds. Most of us had never observed our mental processing with such curiosity, or expressed it in such detail. The vastness of nuance, diversity, creativity, and richness of people’s minds was impressive to witness.

To make our thinking more efficient, we associate words with symbols. This covers the majority of how I think as well. The complexity of concepts I can represent this way has increased over time. It’s as if I have an infinitely complicated map which I’m able to zoom in and out of spatially, as well as fast forward and reverse in time. Sometimes it’s barely even visual.

After having easily repeated some words from the top of my head during the exercise, remembering became difficult. My partner asked me to describe the difficulty, and what the act of trying to remember looked like. With my mind’s eye I saw a link, like a line in an Obsidian graph, but there were no letters at the end of that link. Instead, there was an aura of blue color. As I was trying to remember the word, single letters appeared, floating around. Different colors started emerging from the background, along with the letters, shaping form in three dimensions. As I still couldn’t grasp the word, I felt the mental effort as a kind of pain, a physical tension in my head. After becoming aware of it, the colors and letters faded into a gray mass.

After the exercise was over and I was allowed to see the list of words again, I was able to spot the word I had been looking for during this experience. It then made sense why I had associated it with the color of blue, and some letters had come to mind. But I don’t remember the word anymore — I only remember the abstract visual components, and the feelings associated with it.

As I’m writing this story, the words I choose to use are asserting symbols into my imagination, and the original memory starts to fade.

This is one example of how writing — finding the words to accurately represent thoughts — can be a real challenge. And it’s a challenge we need to keep engaging with.

Using LLMs for writing can further widen the distance between our inner lives and the rest of the world. It can do so by pushing us to settle for lazy, simplified, suboptimal versions of representation.

2. How Writing without LLMs Makes Us Better Thinkers

Whenever I have a new idea, it often appears in an abstract form. I’m then faced with the verbal-abstract-verbal translation challenge when attempting to explain myself. And I often fail at this stage.

Unfortunately, LLMs can’t help me. What they can do is generate options for words to describe my idea, among which I then choose the best candidate. But I still need to write the prompt.

In order to formulate a prompt, I need to have some words to represent my thoughts. However, they never entirely succeed in doing that.

When an LLM responds to me, it simulates understanding by connecting words in the prompt to existing text. The result is often disappointing — the output doesn’t match the meaning I have in mind, and something gets lost in between. I then need to return to my abstract space to attempt to draw more from the original thought.

Another problem with using LLMs to help verbalize what I mean is that I need to be able to reverse engineer my own conclusions by carefully examining my associative process. I owe it to myself and others to be epistemically honest and transparent in my reasoning. I can’t possibly maintain these standards while making AI cover the inferential distance. It could make my ideas sound convincing and explain why I’m right — even convince me of reasons why I think what I think. But in this process, I risk losing my original chain of thought.

So instead, I need to keep writing without LLMs to practice and evolve my methods of formulating thoughts into words.

Here are some techniques that have helped me in the process.

2.1. Babbling

• Brainstorming in the face of cluelessness. We’re not going to produce as many novel ideas if before starting to think about a problem, we go and look at what others have said about it. Our original ideas fade into the background in our minds. Writing down and discussing initial, even crazy-seeming ideas can be extremely valuable!

• Practicing open-ended reasoning. Instead of starting from the outcome and reverse engineering it from the “why I’m right” standpoint, start from an open question, explore different paths one could take, and remain humble about where you might be wrong.

• Embodied and visual drafting. Techniques like scribbling handwritten notes, drawing graphs, or going on a creative frenzy in front of a whiteboard, deserve their own chapter. The freedom of painting our mind on a canvas, then stepping back and looking at the piece is a great way of processing thoughts creatively — and privately!

2.2. Pruning

• Writing whole sentences when taking down ideas. One thought is at least one sentence. I often write notes too quickly, using single words or half sentences. This doesn’t capture the whole thought, and I sometimes end up missing the original meaning of my eureka moment.

• Imagining we're speaking to someone. Going through the sometimes boring part of turning our bullet points into full, reader-friendly sentences is part of the process of improving thoughts. If we skip this part, we miss the opportunity to learn how to articulate ideas.

• Asking: “Is this really what I mean? Do I really agree with what I’m saying here?” Does every word truly capture the meaning behind what we’re trying to express? How certain are we about what we’ve written? Is there anything that needs to be added or removed to more accurately describe what we think?

3. Protecting the Organic Thought Process

The very core human process of self-expression through language is incomplete by default. Writing — engaging in the process of attempting to make ourselves understood in the face of this realization is truly uniquely human. It’s a task that’s not just preserved for artists, bloggers, or those who keep a journal — it’s for anybody who engages with the world using language.

Writing remains an important challenge, whatever we’re trying to accomplish with it. When we use LLMs for writing, we miss out on the opportunity to truly engage with the process of expression — but also to understand ourselves and the inner workings of our minds. Let us be aware of the nuance embedded in them, and let our imperfect, human voices be heard.

---

Do you resonate with our mission and would like to engage more with us? We are looking for a social media manager to help us grow our our social presence. This is a junior unpaid position. Check out more here and help us find our dream candidate.