Article summary
When I woke up this morning, one of the first things on my mind was that my next Atomic Spin blog post was due in less than 3 days, and I hadn’t started anything.
As an Atom, I write for Atomic Spin every 60 days. Writing for Spin is my adult-life equivalent of writing papers in college-I usually find the process reasonably satisfying and I’m generally happy with the result, but I always put it off until the last minute and try to get it over with as quickly as possible.
Recently, OpenAI released a groundbreaking chatbot called ChatGPT, and the news media has been absorbing the implications. So this morning, I had my Spin post on my mind when I browsed to this article in The Atlantic, provocatively titled, “The College Essay is Dead”,. And then, my wheels really started turning.
Could I delegate my blog post duties to GPT-3? I decided to try it out.
Using GPT-3 to write for Spin
A quick google search for ‘How to use GPT-3’ led me to the Open AI Playground, and from there I was off to the races. Now I needed to come up with a prompt for GPT-3 to write about. For a bland test case, I tried the following (a topic I’ve written about before):
Write an article about how to decide whether a business should build custom software or buy off-the-shelf software.
In under a minute, GPT-3 wrote me an article. It was a good article. I probably could have called it quits at that point, but now I was intrigued. How would it do with a more complex topic, maybe something involving a management challenge? I settled on the following:
Write a 1000-word article about the importance of effective change management when deploying a new piece of software into an organization. Include recommendations about how to help employees adapt and react positively towards new business processes and new habits that they will need to adopt.
I felt gleeful as GPT-3 spat out paragraph after paragraph of pretty decent prose that hung together as a cohesive, and actually quite useful, article. However, there were a few problems to fix. GPT-3 came up with a good set of core ideas, but it repeated those ideas without adding new information. And while it was readable and held some useful ideas, other ideas were incomplete. The writing also lacked personality or persuasiveness. So I patched it up, removed the repetitive parts, and added a little of my own flair. I could have gone a lot further, but for the integrity of this experiment, I decided to stick to a minimalist approach.
The Results
This was the Spin post that I published as a result: Deploying New Software into Your Business? Don’t Neglect Change Management.
As an article, it’s not the best, most innovative thing I’ve ever published. But the ideas are solid and helpful. It’s not terribly different in form, structure, or tone than many of my other articles. I stand by it as a useful piece of content. And it was embarrassingly quick to write. More interestingly, it was a fun exercise for exploring how much I could get done with AI (and how quickly), and where the robots fall short. And it raises interesting questions about how internet search and content, and the ethics surrounding them, will change and evolve in the not-so-distant future.
Does it matter that this time I let an AI do the heavy lifting on an article, as long as I can validate from my own lived experiences that the information is true and worthwhile?
How will SEO change due to AIs being able to churn out content that’s more or less indistinguishable from that written by humans?
Will I use GPT-3 to write my next article? No. But I can see it, and tools like it, as being extremely useful for the creative process — helpful for busting through writer’s block, as a springboard for something else, for exploring and interrogating and combining ideas, or… even to get out of homework.