Dear Staples students and faculty,
I’m Tom Zhang from the Staples class of ‘23 (which makes me an old-head) and I currently study Computer Science (CS) on the Machine Learning (ML) + Artificial Intelligence (AI) track at Caltech, the same university that Sheldon Cooper, Bruce Banner, and Walter White attended.
While at SHS, I often volunteered to help with some math and science classes in my free periods which gave me a sense for what it’s like to be a teacher.
As an aside, you’ll realize in college that the teachers at Staples are way better educators than even Nobel Prize-winning scientists.
So, I wanted to provide some background on principles behind large language models (LLMs) like ChatGPT, dangers of using AI as a student, and how AI can better serve teachers and, in turn, students.
The Guessing Machine
AI is a very new field of science so a lot of the terms are B.S. If you go back in time, AI would be called Machine Learning, and before that, Statistics.
If you haven’t taken a statistics course before, it’s essentially a way to learn from data: creating visualizations/summaries, describing behaviors/populations, and predicting trends/outcomes.
Of these abilities, predictive modeling is the most recent common ancestor of what is today considered AI. In essence, ChatGPT and all the other LLMs are just next-word guessing machines that work as follows:
- Read a block of text as context
- Crank the numbers to guess the next word
- Repeat above steps until answer finished
The key part is cranking the numbers and guessing the most likely next word in the chain. That’s the only thing LLMs are trained to do, the rest of the behavior emerges from the statistical wizardry.
AI is not intelligent, it just knows which next word is most likely based on the previous words. In math terms, ChatGPT finds the output that maximizes P(output | context), the probability of the output word given the context words.
AI is not dumb. It’s really good at exams and other tasks, but it’s smart in the same way that your smartphone is smart—it’s a smart tool, but it’s only a tool.
The Honor Code
“No member of the Caltech community shall take unfair advantage of any other member of the community” is the Caltech Honor Code, it’s the only rule we follow here.
But it’s still quite weak, because “any other member” should include the student themselves; it’s just as wrong to take advantage of yourself as it is to take advantage of someone else: treat yourself the way you would want to be treated if you were someone else.
As a student, using LLMs is a terrible idea not only because you will have an unfair advantage over other students, but because you will not learn anything. It’s not okay to not learn because that makes your school a worse place; your classmates learn from you and vice versa.
Exercise Zero
For example, if you have access to ChatGPT in Intro to Programming, the AI will be doing all the writing for you. In computer science, you won’t know how to write a loop. In chemistry, you won’t know how to do a titration. And in physics, you won’t know when the apple drops on your head.
Just ask the Staples Lifting Club: if you want to gain muscle to lift heavier weights, is the optimal workout going to be three sets of having the robot squat the weights for you? Or are you the one that’s gotta lift those weights?
Archimedes invented the lever not so that people could become more buff, but so that they can lift heavier weights without doing as much: AI is not the one unlocking its potential through understanding problem-solving and persuasion. It’s you. The human.
The Sub?
Some students have expressed concerns about teachers using AI and AI replacing teachers, and these concerns are certainly understandable because it would be detrimental for AI to dictate the content covered in classes or completely take over the role of teachers in the classroom.
However, these concerns are quite far from realization because a virtual teacher cannot replace a real teacher. As someone who remembers what it was like to go to virtual Zoom classes, I remember that half the kids were playing chess and Valorant, half the kids were scrolling on TikTok and Amazon, and another half put up an animation of themselves paying attention while they went out!
By induction, using the fact that “kids these days aren’t as smart as they used to be,” we can conclude that teenagers will never reach the level of self-discipline where they will choose to learn through some AI lesson plan over playing video games and scrolling reels.
But that might just be me!
Unorthodox Toolbox
Teachers don’t really have a toolbox of technologies like most other professions.
In modern professions, restaurants like Frank Pepe’s have UberEats (it’s delicious and the only good thing about Yale), and Kobe’s A5 Wagyu Beef farmers have JBL speakers to play Mozart to the cows.
Recently, I was in a very rural part of Kobe, Japan. They didn’t have paved roads so taxi drivers just pulled carriages instead of driving taxis.
Similarly, all Staples teachers get is a whiteboard with markers made of toxic chemicals (even though they love sniffing them) and a so-called SmartBoard that’s always broken. The most noble profession also seems to be the one with the crappiest toolbox!
Also, most people with jobs would love to get replaced! Specifically, just the boring part of their jobs.
Exercise Infinity
Every year, the great Staples migration of tests from upper- to underclassmen means some students already know what the test answers will be. As a teacher, there’s a constant need to tweak the test a little bit so that students can’t just memorize the answers.
Imagine if a certain clever teacher wrote a program to generate new problems for every single test so that there’s no longer a need to write new tests every year and change the questions manually…
Well, this has already happened at Staples, I won’t name this shy teacher but I’m sure news will spread through the grapevine. And one might think that it’s witchcraft or AI (same thing), but this is effectively replacing part of a teacher’s job!
Is it bad? No, it’s amazing! Every teacher would probably volunteer a sizable fraction of their salary just to get one of these programs written so that they don’t have to rewrite tests anymore.
Too bad no one will make any money off of this because any LLM these days will write that program for free at the price of the test questions becoming part of the AI company’s dataset.
Teachers can cut out so many menial tasks from their lives just by incorporating AI into their routines; here are just a few ideas I would try if I were a teacher:
- Problem generation/grading
- Generate new lab/project ideas
- Summarize written student feedback and suggest improvements
- Plan out class/topic schedules
Are We Cooked?
At Caltech, when we see fire, we like to think about it as the gift of fire from Prometheus, except for when the LA wildfires happened, in which case we were slightly less optimistic.
Every revolution, including the AI Revolution, is a trial by fire. Although the current state of AI in education might look like an episode of Kitchen Nightmares, it’s more likely that education will be reborn with a higher capability of serving learners and educators.
Ultimately, as a school with some of the greatest teachers in the world (not joking), it’s the role of Staples to lead the way in the future of AI-assisted education. Also, if you’re an administrator, then you might get a huge promotion for doing something like this instead of some worthless schedule overhaul.
A Small Note
To end this article, I would like to thank all the Staples teachers I’ve had—even if you gave me a B—because it was very useful! As the admissions officer reading my transcript wrote in my file, “This kid has a few B’s, which is quite refreshing to see.” So teachers, keep refreshing these kids’ transcripts!
And as a piece of advice to Staples students: it doesn’t matter where you go to college. Going to a better college does not improve you in any way. To move forward, you must take the steps to improve yourself and no name-brand will help with that.
If you enjoyed this article or have any feedback, feel free to reach out by email at [email protected]!
