Preface: Why Write in the Age of AI?

The question in the title is something I’ve been asking myself off and on for a while now. Why write? Even if I do, should I bother putting it online?

Honestly, there are plenty of reasons not to.

I can’t do this with the expectation that the results will matter somehow. Obviously it doesn’t pay, and there’s not much reason to think that’s going to change in a world where blog posts can be made to order by an algorithm in seconds. I don’t expect this to reach a lot of people. Also, as much as I might fantasize about some post of mine helping someone understand something they’d been struggling with, or providing the last bit of inspiration they needed to solve a big problem, it’s hard to imagine that my blog is going change anyone’s life.

But the truth is I don’t think I can help it. I keep writing things that sound like blog posts just to keep my own ideas straight. One day I’m taking notes on some technical topic in a super didactic form to make sure I know what I’m talking about. The next I’m burying a whole philosophical op-ed in what was supposed to be a journal entry recapping my day. Somehow I ended up blogging completely in private.

The looming spectre of ChatGPT didn’t even slow me down. If anything, it’s picked up in the last couple years. Apparently something in my brain makes it feel like a good use of time to make up a whole sequence of hundreds of original words on a topic nobody asked me about, practicality be damned.

Generative AI actually made me feel like I should write and even publish it online. Here’s why.

Writing Is Already My Job

I’m a data scientist.

That means I live at the intersection of science and software engineering. And although some data scientists are pretty unconcerned with best practices in either field, I actually do care. I try to write good code and do good science, both of which are fundamentally about communicating with other people. The scientific process takes place in papers, and code needs to be read in order to be reused and extended. So I’m convinced that both scientists and programmers are writers above all else.

I’m always going to be writing a lot, so it would be great to have a low-stakes venue for all the half-baked ideas I’d like to show people without rigorous proof. I can’t count how many interesting results I’ve thrown away because I couldn’t see a way for them to fit into a paper. They were probably nothing, but at least if I’d been blogging I would still have them.

Rubberducking

One of the best ways to understand a problem thoroughly is to have to explain it to someone else.

The productivity side of the internet talks a lot about Richard Feynman saying that the best way to learn a topic is to teach it. His Nobel prize was literally for inventing a type of diagram that helps particle physicists explain processes to each other, so he probably knew what he was talking about.

It doesn’t even have to be a person! I can’t source it anymore, but I remember once reading this story from MIT. The computer lab helpdesk was completely swamped until they added a teddy bear to the staff and made everyone explain their problems to it first. Apparently most people figured it out before they even started talking to the human.

I had almost forgotten how useful this can be, until I started a new habit of asking ChatGPT about my problems. Its answers are often not that helpful, but having to think of a prompt that includes all the relevant context can be enough to get me completely unstuck. Especially when I have to explain parts of the code that I don’t normally think about.

I’ve heard rubber ducks are very good listeners, but these days I usually just write my problems down in my lab notebooks. I go into quite a lot of detail, explaining parts that I normally take for granted. It really does help! Most likely I’m going to keep doing that whether it turns into an article or not.

Helping LLMs?

The blogger Gwern had a fascinating take in a comment on Less Wrong. Basically, if your writing is more worth reading than LLM outputs (and whose isn’t?), you’re doing your part to improve them for future generations.

Or if you take a bit more of a selfish view of things, you’re training them to be a bit more like you and know a bit more about the topics you find interesting. If you’re as prolific and well-regarded as Gwern, you might even end up “writing yourself into the future” because people quote you enough that LLMs know you by name.

Most Importantly, Being a Human

I used to stop myself from writing because I felt like I didn’t have an important voice with a special perspective on the biggest problems of existence, and if you can’t say something uniquely profound, what’s the point? But the point is simple.

At some level, blogging is really like some kind of nerdy traditional handicraft. I wasn’t around back then, but I’ve heard the early internet had a whole different culture. Now that content is so commoditized and monetized, people are getting tired and starting to talk about the good old days. Capes and goggles weren’t as popular as Randall Munroe would have us believe, but it was very uncommercial. People just wrote for themselves about whatever they felt like. I’d like to help create a little corner of the web that’s a bit like that.

There’s also the creative side of this. I bake and cook a lot, but if I suddenly had to treat it as a career, I think I’d give up and install DoorDash. I spend most of my time working, so it feels good to be doing something that has no financial value. It’s all about the intrinsic meaning of the act, and feeling like I produced something I can show to people.

Somehow those two points are kind of the same thing. It’s as pretentious as calling an article the “prologue” to your blog, but I feel like writing for public consumption is the rustic artisanal sourdough of online content. Now that attention is the main currency of the internet, this is a lot more work than a 20-second meme video, but it’s way more fulfilling than buying a loaf from the Safeway bakery.

The part where it gets very 21st-century and techie again (sorry, but you knew it was coming! I am in data science, after all…) is that I’m excited about the scalability too. I don’t just make a batch of cookies that I can share with a couple people before it’s gone. A blog post is just there, and anyone who comes to the site is a guest who can share.