LLMs & Academic Essays
Use them like the Homo Sapiens you are
Thirty-nine 3,000-word portfolios and three 12-hour shifts later I am ready to beg you, my students, to start using Large Language Models (LLMs).
The bells of doom are ringing in Higher Education. “The End is Nigh!” “They all cheat!” “All the essays are written by ChatGPT / Claude / Bard / Gemini!”
No they aren’t. You, all my BA students, have clearly spent long, restless, frustrating hours producing lots of poorly formulated sentences as if you have never heard of LLMs.
You write extremely long paragraphs with no punctuation to speak of. You’re doing your best to replace straightforward phrases with convoluted, multi-syllable words preferably with impressive French, Latin or Greek roots. Your texts are a universe where |your|, |you’re|, |theres| and |their’s| roam free. A majority of you even write up to the 3,300-word limit, seemingly unable to stop, rephrase, shorten the flow of unnecessary words.
Written language is not your friend. Perhaps you don’t read enough? Perhaps you grew up expressing thoughts in 140 characters or with 3 emojis? Perhaps you no longer or just rarely can command the attention required for expressing thoughts with written language? Not judging, just asking.
Welcome all to 2024, the age of LLMs.
An LLM won’t write your portfolios. Not yet. Not well enough. But soon! However, by the time they are able to (merely a question of ‘when’ not ‘if’) any self-respecting educational institution will stop asking you to complete tasks that can be written by an LLM.
An assessment task that can be written by an LLM should not be an assessment task. Just like a lecturer who reads Powerpoint slides should be replaced by a Zoom recording. Both are a waste of time and space and an affront to everything that constitutes learning and teaching.
Use LLMs where they are strong.
Use them to check your grammar, to clean up your sentences, to shorten your verbose paragraphs. Use them to get rid of those 200-300 extra words that express exactly nothing beyond your inability to express something. Use them to make your language just a bit more formal.
Use them as tools that allow you to examine what you’ve written and will then summarise it in as many words as you ask for. ‘Claude, give me a 150-word summary of the following text.’ You’ll get a concise version of what you’ve written — you’ll see what your language amounts to. You will see what you have written about, as opposed to what you think you have written about.
Use them for what they are good at.
Don’t use them for what you want to become good at.
The argument is not that AI or an LLM cannot write your paper. It’s that you don’t want it to. It’s all so obvious: you can’t get good at anything by letting others do it. I won’t get any slimmer by watching you fast.
The underlying issue is that the use of LLMs threaten the certification function of Higher Education Institutions (HEIs). If the value of a degree is not the gain in your knowledge and ability (and everything that goes with it from an increase in productivity to personal happiness) but merely the act of certification, then LLMs necessarily become WMEDs (Weapons of Mass Education Destruction) in the hands of the certificate-buyer (the student). In the arms race the follows, the certificate-seller (the HEI) starts investing ever more in its defense budget (AI detectors and staff spending long hours AI detecting). Certificates from HEIs with less potent armies of detection depreciate in value.
There is an obvious way out: a de-massification of education. Students want to learn, teachers want to guide them. HEIs bring supply and demand together and certify. All that’s needed is certification requiring the completion of tasks that in fact measure students’ knowledge, ability, creativity.
The only downside: individualised education is time-consuming. So what? Universities expect their students to spend lots of time preparing their projects, researching, executing and evaluating them. Just allow teachers to do the same.
In the meantime: can you all please start using LLMs for fine-tuning the formal aspects of your academic submissions?
You’re not wrong at all
I feel the creators of this technology have really distorted the perception through how they talk about it. It’s call generative not refining AI.
It also doesn’t help that this tech is so underbaked at the moment.
What I want is a gammerly or an auto correct System built into word that show sentence and paragraph suggestions in context allowing to choose by sentence or word.
Right now it feels like if I wanted to use AI to simplify my at the best quality I’d need to copy chunks, too small to have appropriate context, into chat GPT4 (paid). Getting a text with parts I do and don’t like and then having to poorly stitch the original and the simplification together. This will change but right now specialised tools are based off older or smaller models and the best AI models that have a large contextual buffer are too general and unspecialised to be used in this way.
Even I find a way around this because it’s not hard to do. There is still a level of uncertainty right now. We’re not at a point of acceptance yet so if I use it.
it’s detected will I get in trouble? Probably not but we’re not at the point where the answer is just no.
And more personally I still feel most AI isn’t morale stealing the work of artists without credit, payment or permission. While we exist in an era where we can choose not to use them, I will. Much like sharing all our data and personal information with tech companies isn’t an option anyone because they’re too big and integrated into society; there will be a time where I’ll have no choice.
I couldn’t agree more with this post. The utter lack of teaching how to explore these new tools is sad. It leaves those curious enough to experiment with LLMs to maybe have a better chance in the future because it’s a TOOL. Not a replacement for a human. When I hear people disregard AI use or experimentation with it, I think of people who denied using the internet, computers and typewriters. Just look at the ones who embraced those things first...
The way people work is evolving, and if you don't adapt, you'll be left behind. As someone in your cohort, I hope you noticed with 1 of the 39 essays that there weren’t many your|, |you’re|, |theres| and |their’s|. Maybe because of some LLM use, who knows?