Blu Blog
On Blogging More, and How It Makes Me Think About LLMs
Sat 06 June 2026A Blog Post I Read
I read a blog post today by Steve Yegge of yore1 (circa 2005) about how you (yes you 2) should blog. For any number of reasons and through any number of objections. And, I tend to agree. There's a number of benefits I've experienced to non-fiction writing in either essay or documentation style:
- Writing something forces you to think deeply about it. 3
- Documentation of current understanding and context for my future self
- A well-written document is an incidental artifact which 'happens' once you have thorough understanding of a concept
- Writing output is proof to the world I can think (in my opinion, a much better CV)
Surprise! This is About LLMs!
These 4 benefits are, in fact, interesting for another reason. They're all the reasons I never use machine learning models for output or documentation.
1. Using an LLM to create a document prevents me from thinking about it as I write it.
If I use an LLM to generate philosophical or technical writing, I miss the ability to actually think about it as I'm "writing"4. Writing is editing, editing is thinking. The act of putting pen to paper or hammering on these keys for long enough to structure and enumerate my thoughts is the same one that lets me flesh out and strengthen them. When I started this blog post, for example, I thought "I want to write a note to myself about my thoughts on the importance of writing more." It turned into so much more. These sections of the essay were one-line bullet pointed lists at some point!
2. I also don't get an accurate snapshot of my current mental state and understanding
If I use an LLM to output my documentation--including blog posts talking about how I set something up--I am not able to later see the understanding, thoughts, misapprehensions, or forgotten genius of my past self. I'm only able to see what was a statistically likely plausible sentence about whatever my prompt was. Or if I use an LLM to transcribe notes about a meeting, I miss all the context of what I thought was important about what was being said. I miss the syntheses and distillations, which are a form of (and result of) thinking as well.
3. I am not pressured to have as much understanding as is necessary to produce documentation.
Frequently, once I've gotten to the point of making a blog post, I'm near the height of my understanding and cognitive saturation of an idea. Once that's happened, it's easy to produce writing about something. Whether it's something I have to do research and collate sources for (like an analysis of heavy metal content of Girl Scout Cookies) or something I have built based on some technologies I've found (like how I can get library books onto a jailbroken e-reader with no internet connectivity).
One of the things that stuck with me from Stephen King's On Writing was his imprecation to always be reading. In line at the grocery store? Pull that paperback out of your back pocket and read it! This is, he says, because once you've read enough stories one will crystallize out of your head and onto the paper. This is why I use the phrase cognitive saturation; the thoughts feel like a critically saturated solution about precipitate out in the form of blog posts. Incidentally, I've also felt this with fiction. I've read so many mystery novels and played so many mystery and puzzle games this year (8 novels, 3 mystery games) that I was moved to start writing a Murder Mystery Party story5.
These points of saturation are nothing an LLM is useful to capitalize on; I either enjoy the craft (murder mystery writing) or produce a document as a snapshot of my current expertise (blog posts, documentation). An LLM is not relevant to either process.
4. This blog would become noise instead of signal, failing to communicate my capacity for critical thought and problem solving.
And, finally, this blog also exists as a signal in the noise of the internet. I want people to see the posts and think that if I've written something, it's a subject I've thought about and understand well enough to be talking about. LLM use for writing output is an outgrowth of laziness or indifference. Some forms of indifference are justified--like the quality of a cover letter to the 20th job you've applied to today, since it's not likely a human will be reading it anyway--but it does still signal that this is something you can't be arsed about.
Since I want this blog to signal to readers that I can be arsed about these topics, I don't want an LLM anywhere near my writing process.
The Profound Irony
So, yes, the blog post that started this is Steve 'the days of coding by hand are over' Yegge. Nowadays, he's so all-in on 'agentic' 'AI' that he seems to believe it's all you need to produce anything you need. Large companies will collapse under their own weight and decay to nothing, small teams will move at lightning speed to replace the doddering old giants, etc. I think 21-year-ago Yegge would laugh Gas Town Yegge out of the room. He might not advocate for using LLMs to produce blog posts, but I really don't see how Modern Yegge can square with a writer who said this:
the best things to write about are things you already know, or have just figured out for yourself. You'd be amazed at how many things you take for granted as "common knowledge" are actually brand new to other smart people. There's simply too much to know in this world, and we're all continually learning.
At the very least, something I wish 'AI' proponents were more socially conscious about is the overwhelming harm the companies who produce their tools enact and enable on the rest of the internet. LLM-generated verbiage is kudzu choking the life out of the web. How can anyone not see how completely fucking toxic every LLM purveyor is to digital social interactions?