NYT tests AI generated short essay responses

Above should be a free link. A few of the responses were okay, I thought. Ofcourse, some were weird doozies. Do you think that a reader would have caught all of these the formulaic, AI generated responses?

Honestly they are better than I would have expected. My take away is the more information you can give the bot, the better the output.


Nope. Just talking about the small subset of schools that even cares about the essays.

I also haven’t heard of any schools checking essays for AI, and some have encouraged using AI to help with essay brainstorming. As we know the AI detectors are not all that accurate. Some counselors I know put past years’ essays into a detector and they came back as AI generated, when they weren’t.

It’s time to scrap college essays.


Yes, even before AI, those with resources were submitting essays that were not their own work. Better to have an ungraded essay portion on the SAT or ACT, so that colleges can see how the person writes, and bring back a mandatory standardized test.


I struggle to see any path towards most schools going back to test required especially in light of the recent SCOTUS race decision. The SAT essay was also being gamed because it was so formulaic.


As i said, no grading on essay, just admissions office reads it.


There seems to be little support among AOs for adding an essay to the SAT/ACT. At many schools the majority of applicants are test optional anyway. Nor does anyone want to further enrich CB or ACT.

1 Like

Very true. So i guess now all essays can be ghost written, not just those who can hire a writer.


I am more a fan of the graded paper. Responding to a prompt under time pressure is not real writing. There are plenty of excellent writers who need to flesh out an idea over a few drafts. I agree with the tests though.


But AI can help with those subsequent drafts

At some level you have to trust the system. Just like you have to trust that recommendation letters are honest. I said a graded paper - Princeton, for example, asks for teacher comments to be included. At our school it would be considered a break in the honor code to use AI, though peer reviewing is available.

It’s possible that AOs will start to run application essays through AI detectors, which, as many have pointed out, are not perfect. I recommend that students run their own essays through AI detectors (easy to find online), and if they come up with a score that suggests AI-generated writing, revise to give the essay a more distinctive voice. Some people do write in a style that seems more generic, and a detector could pick up on that as AI-generated. Sometimes this happens to people who do their own writing but then use Grammarly or a similar app to clean up grammar and syntax (Grammarly likes bland, by-the-book, and predictable writing). So the best way to prevent a false positive might be to keep the rough edges that enhance an individual writing voice.

1 Like

I have a kid like that. Very smart but STEM oriented and very reserved. His writing is technically flawless, but dry and generic. He writes excellent research papers but this kind of self-reflective writing is extremely difficult for him.

1 Like

My recommendation is that he saves his notes and uses the “track changes” function on Word or google docs or whatever he uses to write papers so he can show his work if he’s ever accused.

I don’t think an AO will accuse. They will simply move to the reject pile.

At school it is not an issue because they are familiar with his writing style. In fact, the couple of kids who have gotten got at our school did so because the teacher recognized that the writing was not consistent. Kids get in trouble for using grammerly too because they will blindly plug in words that are more sophisticated than their usual vocabulary.

Right, I was thinking more of teachers and professors.

Sure, AI can do a passable job of writing essays of all sorts.

The schools that care about essays are going to be underwhelmed by AI generated essays because AI lacks personality. Supplements are another way of understanding personality.

I’ve seen a lot of AI written stuff and while yes, it is well written, it’s dry. So once AI learns to generate character, motivation, temperament, moral standards, etc…, I still think humans are smart enough to spot a real person when the see it, as opposed to a computer. Maybe I’m an optimist.

This is all a very interesting new frontier. I wouldn’t be one bit surprised if some of the most selective colleges come up with a supplement that AI can’t create. I have no idea how that will happen, but if we are smart enough to create AI, we are smart enough to outwit it.