College Applications: Cultivating Textures, Layers, Voice, and Personality
Should Students Use Gen AI to Assist in the College Essay Writing Process?
College Application Season Like No Other:
College application season is upon us once again, but this year, students have large language models at their disposal when preparing their application essays. How are students, teachers, colleges responding to this significant alteration to the admissions landscape? At this point, it is too early to tell, but it is safe to say that it will be an application season like no other in the past.
Shift to the Ethical Question:
Since the advent of ChatGPT in November of 2022, writers and critics have debated large language models’ abilities to write a good college application essay. Sometime between March and May of 2023, an initial consensus emerged: ChatGPT’s application essays were predictable, lacked a personal voice, and oftentimes strayed from the prompt. On the basis of this evidence, most new outlets at that time concluded that ChatGPT could not write a good college essay. But since then, we humans have gotten a lot better at using LLMs like GPT, Poe, Bard, and Claude. In other words, we now bring a wider array of prompt engineering skills to various LLMs and, using those skills are getting much better, are creating much more authentic-sounding results. And so, once again, news outlets, most recently Forbes and the New York Times, are asking the question again, but this time focusing on the ethical inflection point: Should high school students use LLMs to write [or assist with the composition of] their college essays?
Higher Ed’s Overall Silence about Gen AI Use:
In many ways, this is a much more interesting question–one that takes us to the heart of several questions central to the project of Educating AI: What skills do students need to develop in order to become competent writers? What role should generative AI play in the development of these skill sets? And how do teachers evaluate these skill sets when they are being demonstrated through the networked interaction between human and machine? Unfortunately, this year’s seniors take up the work of writing their college essays in the context of questions like these and many others, and without direct instruction or significant guidance from their home institutions or, more crucially, from higher education. At the time of the composition of this piece, only a handful of colleges and universities have articulated clear case-use policies for generative AI, usually banning it outright, leaving many seniors to guess about whether their prospective use of generative AI will be detected and counted against them in the application.
Most Seniors Play It Safe:
As a result, many seniors are deciding to play it safe, writing their Common App essay largely without the help of generative AI, and only turning to generative AI for assistance with the increasing number of supplementary essays that selective schools demand from their applicants. When a student is applying to 10-12 schools and has to write something like 5000 words in a short span of time, reliance to generative AI to brainstorm or edit seems like a natural response to larger system that is putting unrealistic demands on students who are at the same time usually taking their most demanding courses in high school. Any yet, higher education’s near total silence makes such potential use a gamble, inadvertently contributing to the stress, uncertainties, and inequities that have haunted the college application process for decades, if not centuries.
Just How “Good” Is LLM Writing Anyway?
But the question does nag: Just how good is the output from generative AI? Here, the concept of genre might offer us some guidance. (“Good” is such a slippery and subjective term–particularly in the absence of solid criteria.) When students write college essays, they are operating inside a very peculiar genre, one in which personality, authenticity, creativity, purposiveness, self-reflection, and self-promotion form the essential substratum. Admittedly, our best LLMs (Poe, Claude, GPT) can be prompt-coded to produce those generic features. And yet, a LLM fundamentally needs something personal, authentic, creative, purposive, self-reflective, and self-promotional to work with in order to have a positive outcome.
LLMs Can Help Senior Reduce Chaos and Find Purpose in Draft Material:
If you as the writer do not feed substantial experience into the LLM, the LLMs will simply generate a text-book-eseque flattened-out college essay that no one will want to read and certainly will not inspire a selective school to give you a full scholarship. Here is a little secret for those who care to listen: Good writing has textures, layers, and irregularities that hinge on voice and personality.
Textures arise from the specific vocabulary that you deploy in idiosyncratic ways. An LLM can imitate this if fed a large amount of your spoken or written text, but an LLM can never actually determine the next thing you are going to say… In that gap between the LLM’s prediction and your idiosyncratic way of speaking, lies you, the person these colleges want to hear speak and write.
Layers arise from the patterns of thought that unfold deep inside your mind and heart. The same goes for layers. LLMs can use your words to predict patterns in your conceptual maps, but colleges want to hear from the person in the gap.
And so, my recommendation, particularly in the case of the Common App, is to allow yourself the time to explore the irregularities of your own voice and personality.
Student / LLM Networks: A Collaborative Process:
The following is an experimental process for LLM use during the college essay composition process that should take place after the student completes a first draft largely unassisted. The premise of this process is that most first drafts for the Common App essay are chaotic, too long, and lack a clear through-line or central message or purpose. Use your favorite LLM to decrease chaos and determine purpose. But buyer beware: Only use this process if you are unconcerned with the possibility that colleges will use AI detectors to grade essays and possibility disqualify some essays through the evaluation process.
If you find yourself with too long of a first draft, use an LLM to streamline your material. Ask it to “maintain voice and style, but cut it down to 625 words.”
Usually the LLM won’t maintain your voice or style, but that may be besides the point for the exercise. If you want to guide the LLM back to your voice and style, try this prompt: “Add back specifics, but keep page length the same.” I have had some luck with it.
Then tell the LLM to regenerate the content a few times to see some different options.
Copy and paste these AI brainstorms into a document, and consider them as working propositions for a way towards a final draft.
After you have run these experiments, do the most daring thing you can possibly do at this point. Put away all your draft material (your own and the LLM’s), and write the essay again. Just sit down and write it through. Don’t worry about getting in all the details. Focus on clarity of vision and smoothness of transitions.
Then with this second draft in place, take out the other early draft material. Start to build in the missing pieces.
Leave out anything that detracts from the central message. Let the purpose or through-line that the LLM draft helped you to isolate guide you through your final edits.
Note how collaborative this whole process is. AI is providing a vision, a vantage point on your work. And your job as the writer is to seize on that vision and to balance it out with your textures, layers, and irregularities of voice.
Gen AI and the Future of Education
So what about the big questions about the role of generative AI and the future of education?
Whether we like it or not, higher ed through its overall silence about use of generative AI in the college application process has set an implicit curricular agenda for K-12. At the pinnacle of a student’s writing process, the collaboration between human-machine is now not only permissible but a highly valuable skill. What I am exploring in the paragraph above is a collaborative process at the very ground floor, a veiled shadow of possibilities to come. I know of authors on Substack who have turned over almost all of their writing practice to a collaborative modality between human and machine. Indeed, the necessity to publish continuously in order to build your audience appears to quietly mandate such a move for many online writers. Perhaps one reason so many authors are moving to Substack is the platform’s relative openness in comparison to Medium, for instance, to human-machine collaborations.
Gen AI and Core Competencies, Skill Sets, and Literacies
But I digress… I think it is important to recognize and highlight that higher ed’s seeming inaction is in fact a very significant action, one that will have a domino effect on K-12 curriculum in the years to come. As an educator, I am in part anxious and in part excited by the changes to come. I am fearful that writers will prematurely collapse the chaos of the first draft in search of a thorough-line or a marketable purpose. I am excited that writers will find ways to more efficiently maintain textures, layers, and irregularities in their voice and written expression. At the same time, higher ed’s silence puts increasing pressure on primary and secondary ed’s piecemeal efforts to maintain unassisted spaces within schools to develop core competencies, skill sets, and literacies upon which higher-level literacy and writing interaction with LLMs depend to some extent. At the moment, primary and secondary ed are primarily relying on detection tools to do this work. In my humble opinion, we need better tools and rationales. More on that in upcoming newsletters.
In the meanwhile, seniors, write your first drafts. Lean into your human resources and editors. But don’t underestimate the value of the new tools that are now at your disposal.
This is Nick Potkalitsky from Educating AI.
Disclosure and Warning:
I personally do not think that colleges and universities are going to spend time, money, and energy scanning students’ essays for AI use. We already know that many colleges and universities will be using AI to read and evaluate student essays and applications. The irony of those same schools taking students to task for using the same products to write their essays seems too great. That said, I would only recommend using LLMs on your college essays if you are feeling lucky!!! This is the first year when LLMs is a serious option for most students and thus is a year like no other. During next year’s application season, seniors will be on firmer ground when it comes to LLM use having seen how colleges respond to actual instances of use. Caveat emptor!!!
A Sample of Articles on LLMs and College Essays:
That sounds like a reasonable process. The rewriting, I personally feel, is much easier than trying to graft in personal touches. Recommendation letter writing is a particular genre than mirrors college essay writing in its essential attributes. I think we can maintain integrity through the process as long as we continue to foreground and control our voices in the final stages of writing and editing. Good luck with your recs, Roberto! Always a pleasure corresponding!
Hi Nick. At this time of year I am assaulted by letters of recommendation for college. Between all other duties it can be exhausting. I have been using GPT by inserting my students’ qualities and interests in the prompt for a letter of recommendation. From there I rewrite the letter in my voice. I need support with that initial step or I might not get started.
Thanks so much for your writing.