How the Greylock Glass uses AI in our creative workflow

Close-up cartoon of a metal robot with glowing eyes with a typewriter for a body.

The author generated or modified this text in part with GPT-3.5 pr GPT-4, OpenAI’s large-scale language-generation models. Upon generating draft language, The Greylock Glass reviewed, edited, and revised the language to their our liking and takes ultimate responsibility for the content of this publication.
AI-assisted article

When you see this little robot badge at the beginning of an article, it might imply a number of things related to our use of Artificial Intelligence in the newsroom. It never signifies, however, that the article was generated entirely by ChatGPT or any other large language model system. Before explaining how we use AI, let me first clarify why we employ it as a journalism tool.

The Greylock Glass is a compact newsroom, usually staffed by one individual — me, the Editor-in-Chief, Jason Velázquez. I also happen to be the design editor, photographer, copyeditor, proofreader, web administrator, sales rep, customer service rep, and publisher. The revenue that “we” generate barely covers the cost of merely keeping the website online and available for all to enjoy regardless of ability to pay.

The decision to onboard the incredibly powerful capabilities of AI stems from the desperate need to reduce costs. Other publications have been utilizing AI in various ways for a while now, but have integrated the technology in ways I feel are less than transparent and designed (whether they acknowledge it or not) to eventually replace staff positions. Since, currently, our staff consists of just me, the hope is that by slashing the time it takes to perform certain tasks, I can allocate more time to fundraising to HIRE some assistance. From the outset, I determined that The Greylock Glass would adhere to ethical standards our readers would likely understand and approve of.

The first pillar of this ethical framework is Transparency. Every article that incorporates AI in its creation will be accompanied by the badge at the top of the page, which links to this page. Editorial Humanity is the center pillar — the principle that no human in The Greylock Glass editorial department will ever lose their job due to our use of Artificial Intelligence. Homo sapiens sapiens have to look out for each other, right? Thirdly, the pillar of Human Responsibility lays down the law that the buck stops with flesh and blood editors. All work created with the help of AI will be reviewed by a human, and the final responsibility for the accuracy, veracity, and overall quality rests on the shoulders of the Editor-in-Chief and Publisher.

So, what kind of work is AI doing for us? The range of applications seems to be expanding by the week, but let’s review some of the tasks we’ve used it for in the last month or so since its inauguration:

Assistance with outlining ideas for a story

We have asked AI models that have access to real-time online information to compile the main points on both sides of contentious issues. From there, we can conduct our own research and figure out where to go for more information, and then develop the shape of the story that best serves the reader. A major benefit of this approach is that, by requesting AI to provide both (or multiple) perspectives on an issue, I am compelled to examine viewpoints I might have ignored because they didn’t align with my own biases. Any tool that helps improve the balance in our reporting is much appreciated.

Distilling vast quantities of text

I dreaded going through the June 2023 United States Supreme Court decisions pertaining to Affirmative Action programs at colleges and universities. The Court’s Opinion, along with concurring arguments, runs about 140 pages. Dissenting arguments add another 100 pages. I wondered what would happen if I dumped the entire Opinion into AI’s maw and asked it to generate an abstract. Seconds later, I had my answer. I went through the bullet points and searched the document to ensure that the summary was accurate. Not only did I now have a fair breakdown of the majority opinion, but was astounded to find that what AI composed almost mirrored what I’d read in national news outlets (giving me cause to suspect I wasn’t the first to come up with this idea).

Then I did the very human work of picking up the phone and turning to email to set up local interviews and prepare to write up the local impact of the SCOTUS decision — a task AI isn’t quite ready to take on.

Data extraction

I know I’m giving away all my best secrets here, but I had a spreadsheet that went on for dayyyyyyys, and I just needed a very narrow set of data transferred to a smaller and more manageable simple table. I was staring at a day filled with eye-strain and copy/pasting tedium. Or was I? Drafting the prompt that would explain to AI what I needed so that it could produce my list took about 15 to 20 minutes of trial and error. But once I hit the right phrasing, I felt like I’d hit the jackpot. What would have taken me hours of tedious data entry was done in seconds. Even the prompts that didn’t quite work offered insights into other potential task solutions.

  • Pro Tip: Did you know that if you have a spreadsheet that’s been turned into a PDF, you can open it in Adobe Acrobat (Pro, not Reader), and then export it BACK into a spreadsheet format? From there, you can manipulate columns and rows like the worksheet was your own.

Copyediting and Proofreading

This area has always been a weak spot for The Greylock Glass, despite the fact that I enjoy proofreading and am pretty good at it. It’s just that by the time I’ve worked on an article for a day or two or more, I’m ready to be done with it when the piece is complete, (but not necessarily free of typos, punctuation errors, etc.) Enter the humble AI proofreader! When I wrote for a daily, copyeditors and proofreaders with far superior grammar skills and sharper hawk eyes than mine always ensured that the mechanics of my writing were top-notch. I can’t afford a proofreader at the moment. I may never bother to hire one.

Copyediting is a whole nuther animal. I have tried to use AI for higher-level fine-tuning of what I write, but it always attempts to “fix” my language, bringing it into alignment with conventional journalistic prose. It might be a while before a model exists that can learn how I write and leave my idiosyncrasies alone. This article, for example, was run through AI for proofreading — I could imagine the pinched Jr. High School English Teacher face it wore as it changed “bad-ass” to “formidable.”

So there you have it. Dirty secrets exposed. Proof that I’m aligning myself with our AI overlords in hopes that they’ll spare me come Robot Independence Day, a.k.a. The Singularity. Or at the very least, you now know I’m suddenly rolling wheelbarrows of money to the bank from all the money I’m saving running this membership-driven newsroom with AI assistance.

Seriously, though, The Greylock Glass is committed to our mission of becoming a sleeker, smarter, more bad-ass watchdog news outlet each year, looking out for the interests and well-being of Berkshire residents, especially those who have less ability to speak for themselves or access information about decisions shaping their lives. We are adopting AI because we NEED to be more efficient in covering Massachusetts’ largest county (with the third-lowest population density). We recognize that standards of ethics and usage need to be in place, however, and have developed this framework to keep ourselves honest and to ensure that you, the reader/listener, always understand how the journalistic sausage is made.

We absolutely welcome your comments, suggestions, and questions about our AI policy. Please feel free to contact us at [email protected].

Best regards,

Jason Velázquez

Editor-in-Chief, The Greylock Glass