Hmmm, that question sounds familiar… maybe I wrote about something similar in the past, for instance Would You Use an AI Chatbot to Write Your Blog Post?
That time I was saying that a lot of people were beginning to write their blog posts with the help of AI. And that the result was wonderful grammar (especially for non-native speakers writing in a different languages) but bad for impact.
Fast forward a year and half, and that conclusion is still true. But we’ve also learned a lot about the strengths and limitations of LLM-based AI. You want your writing to have impact, but an LLM is going to write towards the averages. So it pays to try different prompts, check out the various models, and plan in advance for an hour of post-editing.
You need a bit of a critical eye, or you’ll find yourself with an end product even you wouldn’t want to read.
But today instead I want to talk about this topic from the editor’s point of view.
What’s It Like to Be an Editor in an LLM World
In the good old days when editing a blog post I might spend 20 minutes editing “language”: grammar, word choice, and whether the communicative goals were successful for an average reader who knows the field, but doesn’t have the exact expertise that the writer (who is also a programmer, consultant or some other hands-on person) does.
And perhaps another 20 minutes editing the formatting to be coherent across multiple blog posts: capitalizing titles, consistent font sizes and spacing for headers, bullet points that look like the next set of bullet points two paragraphs later, and even looking at the rendered web page to make sure there’s nothing that would interrupt reading flow.
How life has changed in the last two years. Spelling is no longer a problem (even though we had autocorrect before), grammar is much less so (even counting em-dashes, too many colons, and a significant reduction in the variety of grammatical forms), and I’m seeing the most sophisticated and obscure words since that college course I took on Greek philosophy (note to writers: everyday words are just fine!)
So most of my “language” time now is basically making sure the words say what the author wanted to say in the first place (it helps to know the field in this). And LLMs can be a bit hit-or-miss at that. But given the new EU laws on AI content, I need to read and edit every post in full anyway, so I don’t really lose much time fixing the few times that comes up.
One thing I haven’t mentioned so far? The formatting…
Why Can’t Everyone Format the Same Way?
Of course by “formatting”, I don’t just mean type faces, font sizes and punctuation. It also means consistent and coherent styles for bullet points, tables, images, heading sequences and abbreviations, just to name a few.
It would be great if everyone formatted the same way when writing, but you know, these people have jobs, and writing a few blog posts a year is just not as important as making sure clients and bosses are happy.
So when faced with wildly differing approaches to formatting by 50 different people, what are we to do? Create a standard, of course! In the old days we used to call these style guides, and the ones you may remember from school are still around (sorry for the American bias, that’s where I went to school):
- Amerian Psychological Association (APA)
- Modern Language Association (MLA)
- Associated Press Stylebook (AP)
- Chicago Manual of Style
So you can adopt one of those, base your standard off of them, or create your own. And then cry as you see 50 different blog posts by 50 different authors based off the output of 50 different AI models (which our busy authors then want to modify as little as possible) that have 50 different formatting styles.
Fight AI with … AI?
But even when you go with a standard style, you’re still going to be there editing all those blog posts for 20 minutes each, and that adds up over time. And even though that’s pretty much rote time just fixing the same problems over and over again… Hang on…
Did I just say “rote time just fixing the same problems over and over again”? Hey, that sounds like a job for AI…
So here’s the outline of what to do:
- Get the “code” of your blog post (e.g., “Code Editor” in WordPress) – because we don’t want your blog platform to mess things up when you re-insert them – and save it in a file like “original.txt”
- Write out your style guide as an English-language set of bullet points (nested is fine) and save that in a file as well, like “standards.txt”
- Call your favorite AI model with a prompt like “Rewrite the contents of the file ‘original.txt’ using the formatting standards in ‘standards.txt’, and write the output into ‘reformatted.txt’.”
- Paste the contents of “reformatted.txt” into the code editor of your blog platform
- Done! No, wait, double check, AI always makes some mistakes
I’d like to say I’ve gone from 40 minutes per blog post to only 2. But the truth is, I still don’t believe AI models. I still find errors, and it’s apparently on me to fix the prompts. I even get out my old diff tools to check what it changes. I suppose that’s better than training AI models in the first place.
So what I do is first format using my prompt, then double check for any mistakes it makes via diff, and then read the text in rendered mode. I’m down to about 10 minutes total, which is a great improvement, well worth the time it took to try out this approach. It’s also nice that the style guide is in a file that I can change at any point (and which I guess I will have to for the foreseeable future).
So in this case, Yay AI! Formatting is no fun.
This blog post was 100% written by a human.
These Solutions are Engineered by Humans
Did you find this article useful? Using LLM infrastructure to its fullest is still work in progress for most teams, and getting it right early can pay off. We’re always looking for people who enjoy building monitoring solutions at the intersection of AI infrastructure and enterprise operations. Check out our open positions.
Charles Callaway
Author
