hpr-knowledge-base/hpr_transcripts/hpr4601.txt

Episode: 4601
Title: How to be a better writer
Source: https://hub.hackerpublicradio.org/ccdn.php?filename=/eps/hpr4601/hpr4601.mp3
Transcribed: 2026-07-31 16:14:54 (official HPR transcript)

---

This is Hacker Public Radio Episode 4601, for 2026-03-23
Today's show is entitled, "How to be a better writer"
The host is enistello and the duration is 00:09:59
The flag is Clean, and the license is CC-BY-SA
The summary is "By taking apart large language models' writing style, we can learn to be better writers."
Hello there, welcome to Hacker Public Radio. I'm Ann Estelo. If you've ever thought
now you dreamt about creating your own podcast and getting it published, head over to
the Hacker Public Radio website. There are instructions there. It's really easy to get
you started. I'd urge you to do it. So this episode is about improving your writing.
And the way that we're going to be doing this today is by examining the writing style
of large language models, LLMs, which we laughingly call AI. Now, most people can spot
text that's written by a large language model. It's that kind of feeling of soullessness
and impersonal tone, use of the M-capitalized headings. Things like that. But I thought it
might be interesting to examine some text produced by an AI and then maybe try and do something
different. So the first thing to keep an eye out in for is that LLMs are good at spelling
and grammar. And so to be a good writer makes a mistakes. Don't be afraid of making
mistakes. Don't be afraid of misspelling words. Your spell check will probably pick those
up anyway. But don't be afraid of making grammatical errors. Sure, there's probably a grammar
Nazi out there, who might me, who may well pick you up on those and laugh at you behind
your back. But in all seriousness, I spend eight, twelve hours a day reading the internet.
And I would much rather read a few human mistakes than the type of craft that a large
language model will produce. Second thing, verbosity. Large language models are verbose.
They overdo it and they subscribe to the edict that the more words they use, the better.
So as a writer, try and use one word instead of two or three or more. Here's an example
of a kind of thing that I mean. It's an exit from a blog post on the Open AI website,
Open AI, of course, being the company that produce chat GPT. I think it's important to give
Open AI some credibility. And if you like oxygen of publicity, because give it 12 months
or so, and the poor souls will all be out of business. Okay, here's something that chat GPT wrote.
Rapid Consumer Adoption of AI has created a powerful flywheel. Come on, accelerating the
pace at which the technology is being brought into work and professional settings. Close
quote. Okay, let's pair this down and turn it into consumer use of AI has brought it into
the workplace. It's shorter, it's tighter, it means the same thing. Here's another sentence
from chat GPT. Open quote, the history of general purpose technologies, M- dash, from
steam engines to semiconductors, M- dash, shows that significant economic value is created
after firms translate underlying capabilities into scaled use cases. Close quote. Okay, so
let's cut this one down. Technology, comma from steam engines to semiconductors, comma amplifies
firms abilities to create value. Again, shorter means the same thing, not written by an AI.
So, the lesson if you like is, after you've written something, be your own judge. Go through
every word with a red pen or traditionally it'll be a green pen, an experiment with word
removal. Take out a word. Does the meaning remain? You may feel you've lost nuance, but does that
really matter? Perhaps it does. If you're writing creatively or writing poetry, for instance,
it certainly would matter. But is the nuance important in every sentence? Pairing down what
you write incredibly effective. Initially, it's a bit of a pain and it can be more than a bit
soul destroying. Essentially, you'll be coming your own editor and writers on the whole,
hey, editors, but 99% of the time the editor has a point. Eventually, dear listener, the process of
self editing, this process of cutting out your own velocity becomes second nature. And one day,
I guess, I hope you'll realize that what you're writing is sharper and it's more to the point.
When you do wax lyrical, when you do throw in adjectives and adverbs and descriptions and so on
and so forth, you'll feel it's necessary and it's notable and it's notable to your readers and your
readers will thank you. Now, AIC is terrible at this aspect of writing, it adds useless words all
the time, sentences that get there, point over and just a few words typically, chat GPT spins out to
be dozens of words long and the grown-up phrase for this is taughtology, but you can think of it
if you like as laboring the point. Say it once, say it simply. This brings me to a little adjunct
which I guess I could call rule 2A, which is all about syllables. Stop using as many syllables,
sounds really stupid. The tendency comes from new words springing up all the time and they're often
pointless extensions of existing shorter words. Usage is a really common one. Usage is now used
instead of the word use. Here's an example. I made this one up by the way. My usage of the train was
my choice. Sounds okay, but better surely would be my use of the train was my choice or even
I chose to use the train but yeah. Orientate is another one used instead of Orient capability
is now used commonly instead of ability. Why the extra syllable? Don't get me started on
methodology. A methodology by the way listeners is a set of methods not a method. So when was the
last time anyone explained their method or methods? They don't anymore, they explain their methodology,
everyone describes their methodology. Okay, so we may be going a little bit far in stripping out
extraneous syllables but the point remains. There are a few ticks or giveaways that a piece of
text is created by an LLM and so therefore you might want to adapt your own writing so that you don't
exhibit the same ticks or you might like them. I don't know. First one, the rule of three. You see
this all the time and it's become a habit I think now exhibited by human writers. The rule of
three goes like this. Such and such a thing is or it's like x, y, and z. I'm going to quote from the
wikipedia article, link in the show notes, quote reads, the rule of three can take different forms
from adjective, adjective, adjective to short phrase, short phrase and short phrase. LLMs often
use this structure to make superficial analyses appear more comprehensive. It's a superficial analysis
made less superficial by chucking words at it. So this refers back to rule two, pairing it down.
Next tick that you might want to avoid, be aware of comparisons, not only but also, or not just
something, but something. It's not just about something something, something, something,
semi-colon or M-dash. It's something different. A variation on that is it's not only about something,
something, but something else, something else. Wikipedia calls these parallelisms
that explicitly state that a particular item doesn't possess the first characteristic at all.
It's not only about this, but it's about something else. So in summary, how to be a better writer.
First of all, be human, make some spelling mistakes, make some grammatical mistakes. For a reader,
it's good to know that there's a human behind the keyboard, not a piece of silicon in the form of
a large language model. Rule number two, be your own editor, go back through what you've written,
and if you're using tallologies, repetitions, and you're being too verbose, take words out.
Does the sentence have the same meaning? If it does, great. You don't need three words. You
need one. Rule number three, if you sound like an AI, people will think that you are an AI,
and will be less inclined to read you. So you can try and avoid some of those ticks. Like,
the rule of three, for instance, the not only but also not just something, but something.
In conclusion, then there are two calls to action. First of all, go and read the Wikipedia article.
If you're that way inclined, there's obviously a great deal more detail in that than I've
latched on to and spoken about today. It's a good read. For instance, there are lots of
mentions of words and phrases that large language models use now and have used since
2022 and how those words have changed and the second call to action, of course, is
going to record your own episode of Hack a Public Radio. That's all. Thanks for listening.
I'm Emma Stello and this has been Hack a Public Radio.
You have been listening to the Hacker Public Radio podcast, at hackerpublicradio.org.
Today's show was contributed by a HPR listener like yourself.
If you ever thought of recording a podcast, then visit the HPR site to find out how easy it really is.
Hosting for HPR has been kindly provided by anhonesthost.com, the Internet Archive, rsync.net, and the HPR Community Content Delivery Network.
Unless otherwise stated, today's show is released under a Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) license.