From workshops to LLMs–helpful tips for editing GenAI output

How can we apply what we learn about writing in classrooms to the process of editing GenAI output? Which strategies and approaches from creative writing workshops are useful when we’re dealing with prompts and LLMs? And is there ever an appropriate time to use the phrase “I thought to myself”?

In this post, I attempt to steal as many insights as possible from a lifetime spent poring over pages in stuffy rooms and turn them into ideas about how to revise your output when you use GenAI, whether it’s Gemini, ChatGPT, Claude, or whatever text-generating AI tools emerge tomorrow.

Let’s dive in!

How does a writing workshop work?

Attending writing workshops is one way to improve your writing chops. The skills we learn in workshops can be easily applied to the practice of editing generative AI output.

I attended my first writers’ workshop when I was six years old, and was leading graduate creative writing workshops when I was forty-six. Over those forty years, the basic format of writing workshops has remained consistent–a group of people sits at tables in a circle, takes turns sharing their work, and engages in a group discussion led by a moderator, typically a more accomplished writer. Through these discussions, students receive and provide feedback, and discover ways to improve their writing craft. 

When the writers’ workshop format started to become popular in academia in the early twentieth century, it was widely regarded as a weird way to learn how to write. One Harvard linguist, learning that novelist Vladimir Nabokov was up for a position, was heard to remark, “What’s next? Shall we appoint elephants to teach zoology?”

Writing rules of thumb

A lot of the confusion stemmed from the widespread belief that writing was a magical process more akin to shamanism than, say, building a table. Admittedly, there’s a subjective element in expressing one’s self with language, but the fact that language is a shared commodity means that it’s possible to develop best practices around it. The rules of writing aren’t inviolable, they’re more like useful rules of thumb.

Well into this century, the question of whether people can even be taught to become better writers has cropped up from time to time in think pieces and faculty lounge conversations. Now we find ourselves in an era when we’re asking whether machines can be taught to write. The short answer to those questions? Yes, of course. The longer answer, as you can imagine, is more complicated. 

The writer’s toolbox

The first, more superficial thing you learn in a writing workshop regards the tools you throw into your writer’s toolbox. These are a collection of aesthetic preferences, style guidelines, helpful habits, a list of Words Never to be Used, some thoughts on adverbs, and various bits and pieces of useful know-how that you accrue over time. 

For example, here are a few of the items in my own writer’s toolbox that I picked up over the years. Where possible, I’ve named the teachers who provided these particular tools. 

• Never start a sentence with the word “However.” 
• There’s never an instance when “utilize” is better than “use.” 
• Never use the word “relationship” when describing a relationship. (Amy Hempel)
• Structure every story like you’re telling a friend you just witnessed a car accident, leading with the most interesting information right away. (Tony Flynn)
• Commas go on the inside of quote marks if you’re American, outside if you’re British. (Rick Moody)
• There should be no spaces on either side of an em dash–like this. (AP Style)

Adding to your toolbox

The more I studied creative writing, the more I devised tools of my own. Here are just a few that I live by. 

• Adverbs should subvert the verb, not support it, i.e. “Gently caressed” is boring; “Psychotically caressed” is interesting
• Never write “I thought to myself.” Just write “I thought.” Is it possible to think to somebody else? 
• Never write “My life changed forever.” There’s no such thing as someone’s life changing temporarily.
• Never use the word “revolutionary” unless writing about political upheaval on a historical scale.

Police vs. crossing guards of language

You’ll notice that there’s a sort of Ten Commandments vibe going on here. A fair number of the tools in a writer’s toolbox might as well begin with the words “Thou shalt not.” Many of these rules aren’t really about avoiding language that’s incorrect, but being aware that, over time, language gets overused, tired, and loses its power to persuade due to repetition. 

As a writer, you bear some responsibility for noticing when certain words and phrases have overstayed their welcome in human discourse. Language is constantly being replenished by slang, idioms, and technical jargon. An editor isn’t so much a cop as a crossing guard, making sure that cool new words and phrases make it safely across the street while tired, old phrases that are still driving around don’t run anyone over.

Preventing readers from falling asleep

Your task as a writer is to capture your reader’s attention and avoid opportunities to snooze through a paragraph with language that simultaneously surprises and reinforces understanding. 

Bonus exercise: Try coming up with a list of items in your own writing toolbox. A great way to start is to think about what sorts of things that people say that annoy you. Then commit to not using those phrases in your work. 

If you walk out of a writing workshop with more tools in your writer’s toolbox than you had when you walked in, that alone is time well spent. But that’s just the beginning of honing your writing skills. You can go even deeper than that. 

Attention and decisions

The second level of learning in any good writing workshop relates not to how you write, but, more importantly, to how you read. 

Most of us, myself included, were taught how to read by adults who were on board with the importance of reading comprehension. This meant that oftentimes, we’d read under the impression that our task was to “get the meaning” in a particular book, story, poem, or whatever. When we “didn’t get the meaning,” we’d feel ashamed, and that’s where many of us decided that books were scary and designed to humiliate us.

When I think about the books that made the biggest impact on me, I can say that it was precisely the books that confused and confounded me most that taught me to become a better reader and, by extension, a better writer. 

In eighth grade, I stumbled on a copy of Dante’s Inferno, and found myself completely lost in this weird epic poem written by some Italian guy in the fourteenth century. Later, as a grad student, I was totally flummoxed by Samuel Beckett’s trilogy, which concludes in a riveting word salad of a volume called The Unnamable. Making me come up to the edge of my ability to understand what I was reading turned out to be a blessing. 

The best way to confront these sorts of reading experiences is to stop and ask yourself why the work is confusing. This gives some shape to your own ignorance. Understanding why you don’t understand something is, after all, a form of understanding. 

In the case of Dante, I was mostly unaware of the historical, religious, and cultural context in which the book was written, so I sought out information on who this Florentine heretic actually was. As for Beckett, I think I mostly clutched just my head and prayed for this thrill ride of babble to end. 

Reading incomprehension

The more you read, the more you’re able to discern when your incomprehension is the product of flagging attention. We’ve all had the experience of looking up from a book and realizing that four pages just flew by while we were otherwise daydreaming. 

There’s no shame in drifting off into random thoughts while you read. Just go back, find where your attention began to fade, and start again from that point. Training your attention involves paying attention to when you’re not paying attention. You can even say to yourself, in your head, this is where I stopped paying attention. Is there something in the work itself that caused your attention to wane, or do you just need a snack? Get up, walk around, come back to the text. Read the thing that’s confounding you multiple times, and go slow. Chances are, whatever you’re reading will begin to open up.

I want to emphasize this point because I intend to come back to it when we’re looking more intently at generative AI output: the more we train our attention when we read, the better our writing becomes. 

In a solid writing workshop, the group often focuses intently on a sentence or paragraph. It’s kind of amazing how many hours you can spend on a single page of text, how deeply you can consider the ways a particular writer makes an argument or tells a story. Flaws, mistakes, typos, and weak writing start to become more obvious. Soon they’ll start jumping out at you. 

Elevating your attention level when reading tends to bring up questions of how certain combinations of words produce certain emotional or intellectual effects in a reader. You begin to uncover the sheer density of decisions that go into something that’s exceptionally written. 

The more decisions a writer made during the writing process–even if they decided to change the word “that” to “which” and then back to “that” again–the better the work tends to be. 

If I’m working on a revision of something and the first paragraph is exactly the same as it was in the first draft, I know that it is inherently not good enough. The tension of a deadline comes not from rushing to get down as many words as possible before I have to turn something in, but from maximizing the number of decisions I can apply to each sentence, paragraph, and section, before the thing is pried out of my hands by an editor. 

So how does this apply to revising GenAI output?

An attitude we might adopt if we’re going to use GenAI for our jobs and not the other way around is that the output is never going to be good enough. Mind you, I’ve been as stunned as anyone by the growing capabilities of GenAI to draft blog posts, white papers, poems, essays, and rap lyrics. I want to be clear that I’m not taking the position of a knee jerk, liberal arts, NeoLuddite, AI-hater. 

Seriously, I’m not. 

I’m more interested in finding the right path through the deluge of content that the LLMs are beginning to deliver. Just as is the case in a writing workshop populated by human beings who are offering one another helpful feedback, working with AI-generated copy requires you to develop your toolbox, focus your attention, and make decisions about the output you’re working with. 

A case against saving time

Generative AI has begun changing how we communicate, and one of its often-cited advantages is that it saves time and helps us become more productive. Technologies that liberate us from drudgery surround us and make our lives better in myriad ways. I, for one, am grateful that I don’t have to use a washboard for my laundry or clean all my dishes by hand. Innovators who’ve saved us time and tediousness–we salute you!

Knowledge work that relies on creativity is different. When your job involves working with content in some way, whether you’re designing web pages, laying out white papers, crafting email copy, coding, giving presentations in front of groups of people, etc., you improve over time because you’ve paid attention, conducted research, made decisions, and curated your personal tool box of rules of thumb and strategies that work.

As Gen AI evolves from a set of tools that help us write, design, and code things into systems that take responsibility for complex activities that involve planning, assessing, and executing multi-step projects, our co-existence is going to depend on the degree to which we continue to pay attention and make decisions in concert with it, rather than simply pasting the output of a prompt onto a web page and calling it a day. 

3 common ailments of GenAI output to look out for

In my work at SADA, I get to learn from brilliant people who understand technology in ways that I struggle to wrap my head around. I’m constantly absorbing insights from SADA experts about cloud security, location intelligence, change management, and a myriad of other topics. Most of these experts come to writing from STEM backgrounds; I meet them coming from the opposite direction to help clarify and communicate their ideas. 

When I started my career working in tech and began editing content produced by smart people who don’t consider themselves writers, I noticed that when their writing needed help, it needed help in unique ways. 

What’s interesting now is that the work that’s coming my way that’s generated with AI seems to need help in similar ways. There’s a uniformity to the ways in which GenAI writing needs work, and I’m coming to develop a taxonomy that may be useful in spotting it. When reviewing GenAI output, here are three ailments to look out for.

1. Vague Pollyanna Syndrome. Copy that is non-specifically positive. Look out for verbiage that includes some variation of “should” or “it’s important to.” Yes, it is important to do things that are good as opposed to bad, and people should do the things that are productive instead of non-productive. This syndrome often stems from the limitations of an LLM to tap into what’s proprietary or truly unique about a company, product, or service. Look for instances of vague happy talk as opportunities to provide concrete examples, metrics, or more specific language that only you as a human can provide.
2. Movie Trailer Voice. Also known as “In a world…” voice. Starting a blog post with “In today’s technological world…” or “In the modern security landscape of today…” makes you sound like a fourth grader padding out the word count in a book report. We see this all over GenAI output. It’s a way to provide a false sense of context. Oftentimes, the real opening sentence of a piece of writing is actually the second one. See if the meaning of your opening paragraph suffers if you get rid of the “In a world” voice. Stick this one in your writing toolbox, it’ll come in handy.
3. Formatting Frenzy. Maybe there’s a more effective way of communicating your message than a thirty-point bulleted list, with each bullet point beginning with a capitalized term, followed by a colon, followed by essentially the same thing the capitalized term just said? Overly uniform formatting runs the risk of exhausting a reader. Give your reader plenty of places to rest, with bullet points, page breaks, headings, and numbered lists.

In conclusion

Improving your writing skills, whether for your own creative work or the work you generate using GenAI, involves cultivating a lot of opinions about what makes writing effective. These opinions can and should be up for lively debate.

We can come to a consensus around our opinions about language over time, as long as we maintain a healthy culture of respectfully communicating about language in addition to with it. 

Language grows, mutates, and evolves over time. Applying focus to the little details of language as you’re using generative AI is time well spent, improving both your writing skills and the corpus of content upon which GenAI models depend.

One of the things that makes GenAI so fascinating is that it, too, operates within the ecosystem of language. The outputs we receive are based on predictions that arise from centuries of humans toiling away in the fuzzy gray areas of communicating with one another, trying–sometimes successfully–to spread understanding. 

A few words about SADA AI

I hope you’ve enjoyed getting your hands dirty with these observations on the craft of writing and how it pertains to editing GenAI output. I wanted to step aside for a moment and tell you a bit about why and how SADA is involved in this whole GenAI explosion in the first place. 

Think of SADA as a collective of technology helpers. You know that one person in your family who you can trust to understand the new remote control that goes with the TV you just bought? The one you ask for help when that weird thing is happening with your phone? 

Imagine that person, but on a more massive scale. Instead of just making sure your router is properly connected, this expert makes innovative Google Cloud technologies work for entire companies. SADA is full of these experts. And one of the areas of intense expertise at SADA is laser focused on generative AI. 

Within SADA, there are multiple group chats firing away at all hours of the day with messages relating to the latest developments in Gemini 1.5 Pro, Vertex AI, and all the other Google AI advances that are coming in hot. They’re using and taking copious notes on every GenAI product they can get their hands on. SADA engineers and experts on every aspect of the cloud are diving deep into the AI weeds in real time, converting their growing awareness of what these technologies can achieve into easy-to-manage solutions for customers of any size and any industry. 

SADA’s AI experts are hungry to discover ideas that may seem half-formed or overly ambitious. Every consultation with a customer represents an opportunity to explore a novel new use case, land upon a cost-saving strategy, or deepen the understanding of these powerful new tools and situate them in the wheelhouse of your organization’s vision. Reach out and schedule a complimentary conversation with SADA’s AI team to get started on what’s next for this powerful technology.