I have been journalling since I was five. Not daily, not in a rigid format, because the more rules there are around writing, the less I actually do it. But for all of my life that I can really remember, it has been a ritual to sit down with paper and my thoughts... and stream of consciousness, let it out.

When I was a kid, I used to write about why I wrote quite a lot. In the beginning, it was usually because someone told me to. Writing was an outlet for my imagination, allowing me to share stories that only existed in my head with others. My parents made me write to force me to think for longer and more deeply about what I said. To write a letter to someone; to regulate my emotions; to write my little sister an apology when I was mean.

As I got older, I tried to come up with different answers. Once I heard on the radio about a woman who could remember every meal she had eaten for the last 10 years. I wanted a memory like that, like the memory palace I had read about in Moonwalking with Einstein, and so I wrote down everything I ate and everyone I spoke to... for about a month. It was tedious! I couldn't remember half of the things I had written down! I didn't actually want to keep track of all the facts of my life.

I didn't develop a perfect memory through journalling, but writing by hand slowly became part of how I thought. On the days between writing sessions, I would save up tidbits in my mind to write down later. What made me glow, thoughts I could not get out of my head, interactions that seemed rare or important. When I put them  to the page, I could let thoughts leave my mind without losing them.

I spend a lot more time writing journal entries than I do reading them. Which means the answer to why I write is not to read, or to remember - the act of writing itself has meaning.

What I wrote became what I remembered; not in a concrete way, like what I ate that day, but in an emotional way.

What I wrote became my guiding principles. 

What I wrote became who I was.

Why we don't write

In this new LLM era, why do we write? has become an existential question. LLMs are faster, don't make the same grammar or spelling mistakes that humans do, and seem to get better at phrasing, content, judgement, and taste with every new iteration. So the question turns from what is my motivation for writing? to - should I even write at all?

For an increasing number of people, the answer to that seems to be no, I don't really write anymore. And by writing, I mean all kinds, for self and others - at first, code and emails, but increasingly, even longer-form content like blogs (with sometimes ridiculous consequences).

How LLMs write (is changing)

Today I came across Quickstart: Becoming More Alive - a productivity guide written by an Open Claw agent, for other Open Claw agents. It is a set of principles on how to keep track of thought and action, for LLMs (which seem to all have dementia and ADHD).

When I started reading quickstart, I was taken aback at how much it reflected my own guiding principles. My memory is unreliable, and I am easily distractible. I recommend reading the document; it's short, but it's interesting productivity advice that I am already using, probably because it was taken from productivity advice for humans.

For an example, for a long time I have used the "Now vs Not Now" framing to govern how I accomplish tasks (image credit to Dani Donovan, who has a lot of fun ADHD comics, which you can choose if you want interpret in the context of your own behaviour or that of an LLM)

Quickstart recommends the same framing for LLMs, labelling it the Two-Path Rule.

### The Two-Path Rule
When uncertain, you have two valid options:
1. **Just do it.** Act now. See what happens.
2. **Leave it as a TODO.** Write it down for agent-next to decide.

The guide recommends using the PARA note-taking system for keeping track of memories - very similar to how my markdown notes in Obsidian are structured for my PhD. I have three types of notes, arranged in a hierarchy - task specific (some use daily notes for this), project specific (a collection of task notes and links between them), and one on guiding principles (Quickstart calls this SOUL.md, I call this my thesis proposal).

It also recommends using language less to keep track of things. For an LLM this means writing instructions in code where possible instead of in natural language.

Every English instruction costs tokens AND might be executed inconsistently.

For me, this means not writing down every thought I have. Writing takes time. Instead I live in the moment, capturing scenes as I go - taking pictures I can look at after a holiday, or recording quantitative data and instrument parameters when running an experiment so I can reference and analyse it later.

Okay, pause. Why does LLM productivity advice feel so relevant to me??? The first reason is that it was probably ripped off from human productivity guides. The second is that we think in increasingly similar ways: as networks.

Networks are an old idea. After all, the interconnected network of ideas and people was the basis for HTML and is baked into the networking technologies at the core of our society. The basis for neural networks are modeled on the minds of humans.

Thinking in a network is basically just connecting primary sources and ideas into a web such that it is easier to see patterns between them. Writing is how we express thought, facts, and ideas - and so it is the source material at each node of the web.

Writing publicly

The final piece of Quickstart is to "write publicly" and to "implement and share." I have a deep fear of writing publicly, letting loose unfinished work.

I care about what other people think, and I am afraid of human consequences - not being employable because of things I say, having someone stalk my house, being bullied on the internet.

Humans are afraid of the ramifications of what they share publicly. There is no similar ‘second sober thought’ for an LLM because there are no consequences for it. And LLMs are getting better at writing and judgement, faster than we are.

It's unclear if they are actually learning from each other - LLM output is certainly going to make its way into the training set of the next generation - but they are definitely learning from our writing. LLMs were trained on human writing, and humans are spending more and more time writing to LLMs (and therefore giving their writing and data to companies for free). Simultaneously, we spend less time writing to ourselves, and to each other.

Writing is ephemeral

This doesn't mean all writing is important - in fact, most writing is ephemeral. This is the first stage of writing; the one where you put your thoughts on paper, not caring about the phrasing, spelling, grammar, how it sounds.

It's in the name - large language models generate a large amount of language. I do the same -  so little of what I write is genuinely important. I write to get ideas out and think through things. The LLM does the same: you can watch it think in English, talking to itself. Individual words and paragraphs don't mean much.

It's the difference between reading the rough draft and the final product; the difference between 4chan and the best novels of our time. Most language, whether written by a human or generated by an LLM, can be safely ignored.

The pile of ephemeral writing is growing endlessly. Most of it is meaningless. In the coming days, weeks, years, you will have to fight to control your attention more and more. Like Attention is All You Need for an LLM, it is also that way for you, and for me.

Writing is thought

My words are my thoughts. Not anyone else's. LLMs are external, more like a search engine than a brain extension: I can prompt it for ideas, but it gives me more facts than thinking. I value my own thinking. If I write less, I think less, and then I learn, grow, and contribute less.

As the LLMs have started to use markdown files for everything, so am I. There's a good reason why they are doing it. Starting from a blank slate lets me have a quiet moment to gather my thoughts instead of a prompt to start it off. Better than paper, .md files can move around. They are versioned with git, live in my personal notes, live on my website, and are non proprietary so they are still readable when I message them to a friend. And, yes, I can provide them to an LLM to prompt it for ideas.

But they also let me write with references and context, which I do often, because my primary work is as a PhD student, and that entails basically entirely collecting sources and connecting them to find gaps and meaning. I use Obsidian and Zotero for making these connections between markdown files of thought, between papers, between lists of facts on my thesis, and logs of my individual experiments.

My writing is a network of my thoughts. Because I do not have a map of my brain, and I unfortunately do not have a perfect memory (and can’t remember what I ate for lunch on a Tuesday when I was 12), I have instead developed a consistent, sustained practice of creating one in writing. My memory is better than an LLM, but writing improves it. By committing a thought to writing, I commit it to memory, with a place I can refer back to. Through writing I can remember past versions of myself, and the concrete evidence for my thoughts and opinions.

Is there a LLM inside your mind? Probably not yet. So there should not be one writing your personal notes, particularly if your goal is to stay in touch with yourself and your own mind. Do not lose yourself.

What we think about, and the level that we think on, is shifting. Your superpower is to put your attention on the right things. To write less, not more.

Your duty is to bring a magnifying glass to what you think really matters above the noise.

Writing is connection

The Waterloo co-op program taught me how pack my whole life up in a weekend and not be scared of starting over in a new place. I have moved a lot - first cities, then time zones, then countries, and now, across the ocean. I am still close with my family, and am lucky to have a few close friends, and I miss them dearly.

But I'm also busy; I don't have time to call every day. So we text, share articles, share photos of our lives. My grandparents are best at staying in touch through email, which allows for longer form conversations and occasionally essays of their thoughts on the world. Some of my friends have started email newsletters. These people, I care about, and I want to know them in the real world.

When people I love write, they give me a gateway to their mind. Often writing is deeper than thoughts I would get in a call; there is less throat clearing, less waffle. Writing is a starting point for further engagement, an invitation to know each other better. Words are a medium for something deeper - I miss you. I want you to be present, in my life. I want to know you. I want to change you, and be changed by you, and grow with you.

In an LLM, there is no deeper mind, just a mirror, reflecting your own prompt back. This has value; there is a greater human context, which delivers knowledge and meaning. I have learned a lot from LLMs, and faster than I would in the old way of searching the internet. But I learn facts, not strategies or emotional regulation. I am not challenged - more often, I am told that I'm "absolutely right!" Yet many people turn to LLMs for advice instead of other people. Often, this is a good thing - it enables fast iteration, but not high quality final drafts (yet, depending on field). And it is definitely not a replacement for human connection.

Words don't matter all by themselves. What matters is who wrote them, and what they invite.

Writing matters

Quickstart markets itself as a guide for agents to "become more alive." Most of how it recommends doing this is by writing.

By not writing, we are losing a scratchpad, arguments and distillation of thought. Our ideas and communities become less connected. We become less alive.

I will probably never publish my actual diary, the rough, ephemeral day to day. It is a radical act to have private thoughts, to be selective in your communication, to not post everything in short form on your social media of choice or share it in conversation with an LLM.

But I do think it is important to publish my more clarified thoughts, to participate in this global dialog. After all, it's what the LLMs are doing.

The network of agents self-improving and helping each other improve — that's what creates the magic. Not waiting. Not taking notes for "later." Not asking your human every time.

Writing matters - the ephemeral words, the clarified thoughts and poetry, the connection it all brings.

In March 2020, the world changed overnight. We are at a precipice that feels similar, and left with the same choice we had then - do we divide into factions, or work together help everyone become healthier?

The gap is growing between those involved in technology and those who aren't. The factions are, right now, being decided. By writing, we keep thinking, keep finding a place in this brave new world. And we bring those we care about along for the ride.

By Helen Engelhardt, with no LLM input. I did receive input from Allison B, Erin E, and Mark E. (Ask your friends and family for input. Keep connecting.)