LLMs and other stories

Recently, I went on a semi-permanent social media hiatus. Yes, I deactivated all my social media accounts for a much needed break, possibly a permanent one, we'll see. In any case, before I took off into the sunset to touch all the grass, one of the TikTok videos I saw was of a university student who had to defend something she'd written because it flagged as 100% written by AI. She contested the assertion because it was not just untrue, but by her own admission, some of the worst work she'd ever produced. Despite that, she had 10% of her final score deducted because of excessive AI usage. My favourite part of all this is that the alleged AI used was Grammarly, a grammar correction tool given to the students by the university itself! I'm not sure whether to laugh or cry. 

First of all, the fact that we're dependent on AI to detect itself accurately is fucking insane and senseless but that's not what I'm here to rant about. My biggest concern here is the fact that we've gotten to the point as a society where we deem anything too grammatically correct, properly structured and punctuated as AI generated, to the extent that when a UNIVERSITY student submitted an article that was properly written, it was flagged. When did this begin and who is championing it? Is the expectation here that people write as sloppily as they do on social media? Because, in case you people have forgotten, some people know how to write and present grammatically correct arguments. Some people know the purpose of punctuation marks (and their correct names mind you), some people know what an essay is and how it's supposed to read, some people know that there are different types of essays etc., and within that set of people, if there is a subset I would expect to know how to use proper grammar and punctuation, it would be university students. 

"Oh but Noe, these kids use ChatGPT to write everything these days!"

That may be so, however as someone who has worked with AI tools, I can tell you for free that I have never seen an LLM (Language Learning Model, blanket term for AI tools like ChatGPT and DeepSeek) structure a factually correct, persuasive and grammatically sound essay without endless prompts from the user. Any decent writer who has used ChatGPT knows that the first thing it spits out is usually lifeless, all over the place and overall robotic - as it should be, a digital robot wrote it. You know that to get anything worth reading out of ChatGPT and its ilk, a user has to spend a considerable amount of time adjusting the output using prompts. And I don't mean to be controversial here but if a user is putting in that many prompts - reading and editing and fine tuning - I would argue that the LLM they're using is nothing more than a dictation tool at that point. It is no different than having a person type while you tell them what to say. If you insist that the student should type their own work otherwise it's not the same, I must ask why. Is your issue here with the student's output and learning or how easy the student has it?

"Both!", you emphatically proclaim, "these children have it so easy, back in my day-"

Girl, back in my day I had to write 2000-word essays by hand on lined paper during timed exams that constituted 50% of my final grade. BY HAND. I am currently typing this blog post on a MacBook and nothing on earth would make me insist that students handwrite essays because I had to. Why would I want someone to suffer just because I did? Suffering didn't make me learn any faster, it made me resent learning and my teachers.

"Okay, but they're not learning!"

Correction, they're not learning the way that YOU did. They are learning, just not in the way you'd like them to. Children in the information age don't have to learn the way you did because information is not as difficult to access as it was in your day. In your day, in my day, we had to keep the information in our heads for quick retrieval because getting to a library wasn't quick. Even if you had libraries at home like my family did, you still had to find the book, flip through the pages and source the information which took time. Compare that to now, when any student can ask the AI assistant on their phone the answer to any question and that answer will be provided in nanoseconds. They do not need to keep information in their heads anymore because we all have at our fingertips, access to all the information that has ever existed. We are in the information age, we are in the utopian future you dreamed of when you had to handwrite essays in sweltering hot classrooms, so why are you insisting on remaining in the past? Pray, tell.

My final point on this is that we cannot depend on AI to detect itself, for the simple reason that we do not depend on companies to audit themselves. I understand the reasoning behind using AI to detect AI is the age old belief that it takes a thief to catch a thief, but you're wrong there and here's why: every piece of popular word editing software in existence now has AI embedded in it - Microsoft has Copilot, Google has Gemini, Canva has Magic Write etc. AI detecting tools aren't reading the work, they're looking for AI markers in the code and if they detect any, they will flag the work as AI generated.

If you truly want to detect AI in the work of a writer or student, you will have to read it yourself, yes the old fashioned way of poring over someone's work to determine if it's worth the time spent. Oh, does that sound daunting? Oh, you don't want to do it, that's why you're getting AI to look it over for you, because you don't have the time? Well, would you look at that! Turns out AI is good for the goose after all. Before you complain about how you have multiple submissions to read and your students have only one course, I enjoin you to remember what university was like. Remember that the perpetual tiredness you feel today is because you never got any sleep during undergrad, juggling all the courses you had to do while finding time to live your life was exhausting and you barely made it. Remember how you hated the professors who behaved just the way you are right now with your tyrannical self and can it. AI is here, please get on board.

Last but by no means least, please know that everything is not AI. The most popular "tell" that people perpetuate is that use of em dashes indicate AI generated text and I need you people to please, for the love of all that is right and good, shut the fuck up. Editors, journalists and any English student worth their salt hates you as much as I do in this moment because what the fuck? Punctuation marks exist for a reason, come on now. Don't even get me started on those of us who grew up in a certain era of British colonies and had to learn English the hard way. Please stop.

In conclusion, my argument is that we need to find the balance between policing plagiarism and simply being resistant to, or worse, demonizing the unfamiliar. As I mentioned in the beginning of this essay, in case you happen to have forgotten, some people are actually brilliant and can write. You'll just have to read the work submitted by your students to figure that out because it is something you catch when you read the work AI produces. If you don't want to do the hard work of reading your submissions, then what can I say? Live and let AI, babe.