On Writing Really Very Goodly and Stuff, Vol. 1: Making AI Sound Not Like AI (Part 1)

AI seems inescapable in the modern era, and probably for good reason.

With regard to writing in particular, it can transform what would have been hours of work into something that simply takes a few minutes.

But there’s a problem with all of this; a weakness that we all probably should have seen coming, but something that most of us seem to have missed.

Everything is starting to sound the same.

And by sound the same, I mean, everything is starting to sound like it was written by AI.

And so in this first volume of “On Writing Really Very Goodly and Stuff,” what will very surely no doubt eventually go on to become a thousand part series, I would like to focus on a few key points: what “AI slop” is and how to avoid making it, effective ways to use AI in your writing without leaving everything to AI and the single piece of punctuation that can make even a document that was wholly crafted lovingly by human beings seem as though it were AI generated.

From AI slop to AI not slop

Slop is a word that’s come up a lot recently with regard to AI.

Originally a word for the mish-mash of random table scraps and other edible items given to pigs in particular as their feed, it’s taken on a bit of a life of its own in the modern era.

If you’ve ever met an oinker, you’ll come to find out very quickly that, while they are actually quite clean and intelligent, they aren’t exactly the pickiest of eaters. As such, farmers can throw a mish-mash of random household food scraps into a bucket, the contents of this bucket being the “slop” in question, to give to the pigs as feed. This slop often consists not only of leftovers, etc., but also of scraps and trimmings that humans might normally not think to eat.

As time passed, slop became a pejorative ladled at food, particularly that which appeared to be any unappetizing or unintelligible combination of food. More recently, it’s frequently applied to food which, while perfectly appetizing and tasty enough, is cobbled together from low-quality ingredients that are lacking in nutrients, like a lot of processed food made in factories with minimal human contact comes to mind. The sort of thing that looks, smells and tastes fine, good even, but leaves you feeling kinda empty after consumption.

One might start to picture how this has become something that is applied to a lot of AI-generated content as of late.

The reason so much AI-generated content is hit with the “slop” moniker is because, much like that processed food, it generally feels mass produced and as though it were devoid of any human touch (in part because it often is). Even if the disparate parts come together to form something that is outwardly fine, consumption often leaves one wanting.

And while that term has largely been lobbed at AI-generated images and videos, it’s slowly becoming something that gets applied to anything that is clearly AI generated, regardless of the quality of said generation. Sometimes it’s even applied to things that aren’t AI generated but that have certain specific tells (we’ll cover this later).

But what’s a good, concrete example? Well, let’s say you were to, for example, ask an AI engine to generate a blog about how to write with AI while avoiding sounding like you did; one that included common pitfalls that make it obvious that one is using AI to avoid and also on ways that one may not normally think of to use AI. And then you took that blog that the AI wrote for you and posted it as-was on your company website as though you had written it; well, people might call that slop.

So, how then might one go about using AI to make their writing workflow more efficient without having it devolve into slop?

Moving from content generated by AI to content generated by a guy (or girl)

There are a number of ways to take a more tool-oriented approach to AI utilization; ways that keep your own creativity and ingenuity intact while leveraging the substantial power of AI to streamline workflows.

For example, imagine you’re tasked with writing a paper on something that you’re not deeply familiar with. As someone who has dealt with this in the past, one of the areas I struggled with was figuring out where to get started, or what specifically to focus on, especially for topics that are broader in scope and scale. While getting started might normally require a significant time investment related to research alone, AI can expedite this for you by giving you a list of sub-topics related to that broader topic, which can in turn help you find an entry point and get to the task of actually writing quicker.

And on the subject of research, using AI as a quick way to aggregate research related to a specific topic, particularly with links provided, saves significant amounts of time. If you’ve used many search engines lately, you’ve likely noticed the dip in quality they’ve experienced. It seems like, for a variety of reasons, that the results you actually want are getting buried deeper and deeper. AI can not only pull out results relevant to what you’re looking for, it can also provide short summaries for each one, helping you find what you need faster.

There is, however, one caveat to all of this: whether using the search engine’s own built-in AI or a separate AI to pull results from a variety of sources, it is essential that you verify the sources themselves rather than simply relying on what the AI tells you that the sources say in its summaries. This is as, remarkable as the AI can sometimes feel, it is prone to occasionally making errors, which can lead to errors in summaries, or worse, hallucinating, in which it will craft information from whole cloth in a misguided attempt to help.

In short, even when using AI to aggregate research, only ever rely on information that you’ve verified yourself via provided links.

Another great use case is having the AI check what you’re writing for any logical fallacies. Logical fallacies can be an easy trap to fall into, especially if you’re writing something longer and you’re writing it alone, so having the AI double check to see if there are any holes in your logic can be an easy way to ensure that you’re heading in the right direction with whatever it is you’re writing.

The one pitfall here, and it’s something to be aware of especially if you’re writing for professional purposes, is that anything you upload to the AI is going to be stored as training data. Easy ways around this are to anonymize the piece that you’re writing as much as possible, or, even better, to summarize your basic arguments and feed those to the AI instead of the piece you are writing itself. This ensures that the AI isn’t getting any more information than it needs from you.

None of these are handing the keys over to AI; to run with the car analogy, these methods are less autopilot and more GPS; a tool-oriented approach that makes it easier to reach your destination in a timely manner.

But what if you’re completely stuck, or in a rush, and you need something now now now or with a bare minimum time investment? Well, this is when you would lean more into the generative side of AI. The easiest way to do this would be to have the AI generate portions of your piece; three to four sentences each is a good starting point, and you can then expound upon those sentences while editing the text generated by the AI to ensure that your voice is still able to shine through and to give it that human touch that purely AI generated content is lacking.

But this last method has one significant pitfall; something that you need to avoid at all costs...

Quick aside that will at first seem irrelevant but will soon become horrifyingly relevant: I used AI to help write this blog. I went to a commonly used engine and asked it for a list of sub-topics related to making AI-generated text sound not like AI. The AI generated for me a list of 10 different items, and while I didn’t end up using any of those items, something stuck out to me. This was the first line of the AI’s reply:

“Ohhh, great question — this is the eternal struggle”

There’s something in this sentence that sticks out like a sore thumb; something I don’t remember having encountered all too often before people started using AI like crazy; something that I see all the time now.

The em dash. (Continues to Part 2!)

Kawamura International’s services

With XMAT or our other platforms, you can utilize machine translation and generative AI in a secure environment without worrying about the transparency of the cost. At Kawamura International, we do more than just provide machine translation or generative AI services. We offer a range of support to help make your translation operations run more smoothly. If you are facing translation challenges at the moment or have any questions about our language services, please feel free to contact us.