I’ve been writing about all of my AI shenanigans a fair amount of late, and I just warned you in another post that the amount of attention I’m paying to editing is going to be diminishing in the coming days. But I spent a couple of hours this afternoon on another change, which you’ll see coming shortly — or rather, immediately — and that is the fruit of a new workflow, which I will describe in its entirety for you, in the interest of transparency and my ritual endless navel-gazing.
When I started this blog, I wrote every post on my laptop, often in Starbucks, and at least initially in a program — the name of which I don’t remember — that had some sort of interaction with WordPress. It was like a WordPress client, an app in which I wrote posts, which then dealt with publishing them in WordPress. Somewhere along the line, I ditched my Windows laptop and moved my life to Chromebooks. Whatever that app was, I could no longer use it. Maybe it was also deprecated by then. I don’t know. But once I was on my Chromebook, my typing all had to be done through the WordPress interface, an interface I never loved. Sometimes I would compose drafts in Google Docs and then paste them into the WordPress interface. Other times I would just work directly in WordPress. It felt like a tax, though. Formatting became harder. Placing images in my posts and making them look the way I liked became harder. Over time, I kind of gave up on that stuff.
Up until about a year ago, that was how things were. I was sort of unhappily writing my posts on my Chromebook, doing my best to edit and format them, but feeling mildly resentful of the whole process, and continually frustrated by the difficulties I faced in making my posts look the way I wanted them to.
A little more than a year ago — really almost two years ago, as I think about it — a couple of things changed in my routine. First, I started biking to and from work and elsewhere. This meant that time I had previously spent writing on my phone on the subway was gone. (Oh yeah, I left that out — I did a lot of writing on my phone, too.) So there was a loss there in terms of writing time. But then there was a gain. I bought myself a little voice recorder that I hang on a necklace I wear around my neck — along with one of my two wedding rings — and I would just dictate to myself: journal entries, blog posts, you name it. And as technology and AI have progressed, I’ve been able to do more and more interesting things with those voice recordings.
As I’ve told you, one of the problems I’ve faced is that I have this profusion of words now. An hour or more a day of dictation. It’s a lot to process, for sure. But in the last couple of weeks, I’ve refined a system that I had been iterating on to the point that I think I’m now in spitting distance of its being exactly what I’d like it to be. And that I’ll describe now.
When I record a post — such as this one — I begin by saying, “This is a draft MDL post. The title is…” and then I start talking. I do that until I’m done, and then I slide the record button on my little recorder off. Next, I’m at my computer, or sitting by my phone. I send the recording via Telegram to a bot that I’ve set up, called TwoBrain — for reasons I don’t even remember. I don’t love the concept of a “second brain,” which is quite popular among tech bros, but somehow I ended up adopting that title for this system. It spans more than just blog posts: it includes journal entries, correspondence, text for other blogs on which I write, brainstorms, and a variety of other types of content.
But in any event, for those that are draft posts for this blog, my system automatically does the following. First, it transcribes the text. Then it formats it, breaking it into paragraphs. And now, it generates an image — based on a prompt that I’ve developed — that it inserts in the post and attaches as what WordPress calls a featured image. It then emails me, and sends me by Telegram, what it’s done, with buttons at the bottom of the message allowing me either to edit, publish, or schedule what it’s given me.
All this reduces to something close to zero the previously prohibitive friction between my thoughts and your eyes.
So just to give you a more granular example of how this all works: I’m recording this post right now as I bike to a friend’s house for dinner. My friend and I, and two others, form a group we call the Gang of Four, which has met for an hour or so every Friday for 12 or 13 years. We’re in the same field and we’re close friends, and two of us — neither of which is me — are a couple, although they weren’t when we started. So I’m biking, a roughly 45-minute journey from my home to my friend’s, on Sunday evening, and it is 6:57 p.m. as I record this. Right now I am on the Manhattan Bridge, just exiting the portion of the bridge on the Brooklyn side, crossing the East River.
I’ve been speaking for a while now, obviously, since just a few moments after I left my home. At some point in the course of the evening — maybe in the bathroom at my friend’s house, maybe on the way home if I’m not able to bike, whether because I’ve had a couple of drinks or because the drizzle in which I’m biking right now becomes a more prohibitive rain — I’ll transfer this recording from the little black recorder dangling from my neck to my phone. I’ll transmit it via Telegram to the system I just described, and I’ll get a Telegram message back, and an email message back, and I’ll eyeball it. I almost certainly won’t edit it, so whatever typos there are, you’ll have to forgive me. And they won’t actually be typos — rather, they’ll be strange ways in which the transcription engine misunderstood my intent. But so be it.
In addition, obviously, there will be an image generated. The image was generated by OpenAI’s image generator — I don’t even know which engine it is, because my developer set it up for me. The excerpt that appears on the front page of the blog will also have been produced by that system. I’ve struggled with these excerpts. They read a little too much like they’re written by AI — which, of course, they are. They’re the one bit of text on this blog that is generated from the start by AI. All the rest is truly my words, just passed through a transcription engine and formatted.
I never use AI to write anything of substance, because while it’s great at many things, boy, is writing not one of them.
I’m going to stop this recording now and call this meta post to an end, as I bike through Chinatown, about to turn left on Allen Street, heading north to my friend who lives on the Upper East Side. I hope you have a delightful evening.
Postscript:
In the event, I got to my friends a little bit early. Not even just on time, but actually early. So I’m recording this postscript to explain the very quick transfer of the post from recorder to phone and from phone to blog. I’m doing it on the street on the Upper East Side after having parked my bike. So between the moment when I first started speaking and the moment when this post and its postscript arrive on the blog, it will have been no more than about 40 minutes.
In general, I stage my posts so they go up in the morning. I’m not sure why I do that. At some point, I imagine I must have read that that was good for search engine optimization or web traffic or whatever the metric was at the time. Nowadays, I don’t know that it matters at all. But I’m still stuck with the habit. So most of my posts do go up in the morning, sometimes scheduled days in advance.
But this is the rare exception. I’m pressing stop on my recorder at 7.21 and whatever time it’s posted on will be an accurate measure of the time it took me to transfer the recordings, do the tiny bit of splicing to get the postscript into the post, and get it up online for you to read.
Header image prompt (openai): Large-format color photograph, dusk, wide environmental frame: the Manhattan Bridge bike lane shot from saddle height, handlebar grips visible at the bottom edge of the frame, the span stretching ahead into a bruised orange-violet sky over the East River. The Brooklyn Tower looms behind, the city grid of Lower Manhattan glitters ahead through a fine veil of drizzle. Clipped to the cyclist’s chest on a thin chain — catching the last hard slant of sodium light — a small matte-black voice recorder, impossibly intimate against the vast industrial geometry of the bridge. Everything is beautiful and everything is slightly wrong: the recorder is too close, too personal, too small against all that steel and water and failing light, and the wetness on every surface makes the whole scene glow like it’s been lacquered.