Recent digging on Dartmouth’s institutional Claude platform led me to a “team” project section where a seemingly random group of people in the Dartmouth community, intentionally or not, shared customized versions of Claude AI for anyone within the Dartmouth email ecosystem to view and use. My natural instinct was to snoop, and although there were a handful of interesting things there, three different Claude models with meticulous instructions particularly caught my eye.
They were all created by a member of the Dartmouth Office of Communications digital engagement team. They are respectively labeled “Dartmouth Social Caption Writer,” “Dartmouth Social Caption Editor” and “Dartmouth Social Carousel Brainstorm.” Each of them was loaded with a long, elaborate list of instructions.” The social caption writer has a set of instructions that is nearly 3,000 words long, while the editor has a set of instructions that’s almost 4,000. The instructions include platitudes like “You’re like the best editor writers have ever worked with,” and telling Claude that it is an “expert storyteller.”
You can read more about what our reporters found here, including internal strategy documents from the Office of Communications.
Am I surprised that Dartmouth uses artificial intelligence to write its social media captions? Not really. A quick peruse of Dartmouth’s official Instagram captions tell you most of what you need to know –– generic sentences punctuated by an alarming, AI-slop number of green hearts.
Reading through the instructions given to the models was the most interesting part of this whole thing. Someone spent a shocking amount of time coming up with extremely precise documents meant to perfectly calibrate Claude to be the very best social media manager, and to know Dartmouth’s brand flawlessly. Ever since The Dartmouth was made aware of these models, I’ve amused myself estimating one thing: How many social media captions could have been written in the time it took to write the instructions for a machine meant to generate social media captions?
The strongest recurring pattern in the instructions is a near insatiable desire for humanity. An obtained document named “Dartmouth Social Strategy” seems to outline an overarching social media strategy for Dartmouth. The first line under the “Brand Voice” section is: “Our main goal is to always sound human.”
Another document plugged into the model called “Social Caption Writer” is titled “The dead giveaways of AI-generated writing.” It is a nearly 3,000 word document that includes the word “human” 19 times, encouraging it to sound as human as possible. The text includes a list of scores of words that, according to the list’s creator, immediately tip readers off to the fact that something is AI-generated, and includes a more broad commentary about how structure and the absence of genuine human experience makes AI writing “simultaneously competent and hollow.” The “fundamental tell” of AI writing, the article uploaded to the instructions section of one of the models, is “absence rather than presence.”
Is this true? For my own sake, I hope so. Some estimates now posit that more of the internet is AI-generated than human-generated, and that this influx of content is causing a sort of AI slop oil spill where low quality content floods in the internet and is used to train AI models with diminishing returns. To me, it’s unsurprising to learn that we have joined the collective dump of slop that continues to bury the work of real human beings. I’m hesitant to say that the solution to our problem is to abandon the internet altogether, but this increasingly feels like what must be done in order to preserve human artistry.
As an aspiring writer, I have no illusions about the precarity of my future. It might very well be true that 20 years from now, once AI has taken over the world, columns like this one will be looked back at as the snobbish last stand of the human writer attempting to pull a “gotcha” on innovators working to make our world more efficient. I just find this moment in time fascinating –– that there seem to be people out in the world who are wholly convinced that with enough instruction and prompting, we can find some kind of humanity hiding inside our computers.
I also wonder about the interns and copywriters who work at the Dartmouth Office of Communications. Do they get to write anything at all anymore? Will any entry level people in these jobs know how to write, or is this the primordial generation of a never-ending flood of AI ooze as we progressively outsource everything?
Shortly after I began work on this column, the models were taken down from the main “team” page and are presumably now private. I wish before this happened that I had the chance to ask the models a couple of questions. What is your main goal? What do you think of your job? Even I can’t resist the urge to anthropomorphize, scared I’ll find something just like me on the other side.
Opinion articles represent the views of their author(s), which are not necessarily those of The Dartmouth.
Eli Moyse ’27 is an opinion editor and columnist for The Dartmouth. He studies government and creative writing. He publishes various personal work under a pen name on Substack (https://substack.com/@wesmercer), and you can find his other work in various publications.
