The end of the high school essay

When I was in school in Ontario, about 30 years ago, homework was for practice. It was evaluated usually, but not a major part of the final grade; grades for homework were mostly for simply (trying to) complete it. Some students will get their parents or friends to proofread it. Some don't have parents who can read. Some students will cheat; there's software out there that'll make doing homework easier than it should be. (Very guilty of that one myself even back then.)

We were expected to write essays in class, during class time. Most of our grades were based on that. In my later years of high school, my teachers stopped even checking my homework for math etc, unless I wanted them to. Grades for my final year courses were almost entirely from in-class exams and projects. From what I've heard it's still similar here, even with remote schooling. (Though that certainly creates its own issues!) Is it different elsewhere?

Essays might not be the best way to assess reading and writing skill, but this article's reasoning leaves a lot to be desired. The author seems to value an essay insofar as it demonstrates reading comprehension, and not as a vehicle for teaching writing skill. Students need to understand what they are reading to write an essay, but the writing itself is another skill entirely that the author here doesn't seem to address. I read a lot of my peers' works in college and came to the conclusion we need to encourage more writing skill, not less.

> Students get more points for good typing and clear sentence structure than they did for actually thinking deeply, questioning the status quo or changing their minds.

I hear this all the time from engineering types. Writing well means finding a clear, compelling way to communicate your ideas. Sure, we should try to avoid hiding bullshit behind a façade of flowery prose, but a great idea communicated poorly is useless.

> If you grew up in a household with verbally agile family members, you probably did way better on essays than your peers, but not due to much effort on your own.

And? If you grew up in a household with engineers, you probably did better at math. I don't get how this affects any one given assignment type over another.

I spent my school years from grade 4 or so to grade 10 writing the most awful bog-standard school-style essays for assignments. The 3-paragraph structure, the rigid sentences, the fact regurgitation... I got good grades, but I'm sure the teachers hated reading them as much as I hated writing them.

In a similar timeframe (starting from grade 7 or 8), I started writing online. I mostly participated in online debate forums because I couldn't find anyone with (in my highschool opinion) an open mind to argue with in my small town. I'd happily write detailed 12-paragraph posts on the topic dejour in a natural voice with independently researched and well-cited sources.

I distinctly remember being in 11th grade when a switch flipped. I was writing another dreadful essay on the meaning behind The Scarlet Letter for an English class when it occurred to me: what if I wrote this like I was posting on onlinedebate.net? Instead of writing a "persuasive essay" as half a decade of school had taught me, I wrote A Post arguing a point like the internet had taught me. I got an A and the teacher loved it. I wrote everything from there on out like how I actually wrote, not how I thought HS essays were supposed to be written.

I don't know that I have a strong lesson from this. I learned nothing about how to write from school and a lot from the internet. No idea if that's generalizable, though. All I can say with confidence is that I hated (and still hate) The Highschool Way of writing essays.

ChatGPT is also (superficially) good at grading high school essays. I find it slightly amusing that students may be submitting essays written by AI, that then some instructors may be grading it with the help of an AI. BTW ChatGPT does not seem to have a very high opinion of its own writing.

"The answer is simple but difficult: Switch to the Sal Khan model. Lectures at home, classes are for homework."

I think this could actually work!

I don't know, did Wolfram Alpha end homework problems for Calc 1? Not really, because it's useful practice for the test you will probably take. Pointless essays will be diminished but for stuff like AP English exams, where you have to sit down, read a passage then write an essay analyzing it, essay-writing is still the most useful homework. If anything, I think it just enables a larger gulf between those who know and those who don't. I knew plenty of people who in my HS Calc class used Wolfram to do all their homework and then flubbed the tests. I also knew kids who did the homework the way the teacher intended and then I knew some of us who used Wolfram as a way to get unblocked. I see that as more where these language models fit in. Not that never again will a high school student have to write a dumb essay, but never again will a smart kid writing a dumb essay be stuck with writer's block staring at a blank page. Run GPT, generate 1000 words, read, start stealing and rephrasing and reworking until you get back in the flow.

A take I haven't heard yet that isn't just "chat robots bad":

It's not like the output of a teenager means anything, the whole point is that they activate their brain to critical thinking and learning. The exercise of making ChatGPT output look convincing, and familiarizing yourself with that output to defend it if questioned, is still a learning exercise. It looks different, but it's honestly just as impressive as them writing about what Hamlet was thinking when his dad died or whatever the fuck that was about.

What may have even more impact on HN readers is the possibility that we are seeing "The end of the bloviating blog." I say this as someone who—under the pen name raganwald—built much of my public reputation writing blog posts. But looking back, many of my blog posts were low-effort and superficial. The ones I'm really proud of in hind sight are few and far between, and they were the ones where I had something to say, and put some work into both researching and saying it. The rest could easily have been written by AI, if not now, soon.

And just as the author says is the case with high school essays, the correlation between blog posts I'm proud of today and the payoff at the time (marks for high school essays, online popularity for blog posts) is tenuous at best. Popularity is driven by a bunch of factors, and a decade ago, a lot of those factors involved cosplaying an insufferably overconfident techno-priest.

(I am not saying I wrote great stuff that nobody liked. The ones I'm proud of also touched people in some way. But looking back, social media was not set up to reward putting a lot of work and thought into writing, not even HN.)

I'm kind of happy that AI is able to automate and industrialize relatively shallow and low-effort writing. The sooner we recognize that it is not a signal of either hard work or deep thought, the better IMO. That is... If we as a species recognize it. If not, we are on our way to a really boring dystopia where public sentiment is driven entirely by deep-pockets gaming every form of social media we create.

Until we understand the limitations of ChatGPT (see Stephen Wolfram's post), it is entirely premature to immediately begin identifying the ways in which learning should change.

Seth is posting a rant against essay writing. As a millenial (who saw the advent of the internet change school curriculum), I know that our workload became more intense after internet became more pervasive. Before it would take a trip to the library to write a paper. Now, we were doing 4 times as many papers and consuming 4 times as much information, making us X times(!) smarter than our predecessors. And we changed the rules of the workplace dramatically. (We were usually just as competent, if not more competent, as the more senior people in the office. & We were usually underestimated.) AI is going to do the same to education. It is not going to make children less capable of writing, but far more capable of writing. Children are literally going to go over AI generated information and make it better.

I should start fearing the next gen of young adults. I am going to be just as outdated as the boomers are today.

Props to them for shouting out Danny Dunn and the Homework Machine.

Danny Dunn was the kid star of a series of youth-oriented novels from the 1950s that were early exemplars of the "kid and scientist adventure" genre that's persisted in media to this day in the form of Jonny Quest, WarGames (original concept), Back to the Future, and Rick and Morty. His eccentric neighbor Professor Bullfinch would invent sci-fi gizmos or involve himself in scientific malarkey that Danny would get sucked into and cause all sorts of trouble.

In this case, the gizmo was Miniac, a powerful but miniaturized (fits in a room), clanking 1950s computer. It had a microphone that you could ask questions into and it would type out cogent answers (much like GOT today!). Danny Dunn, placed in care of Miniac while the professor is out if town, instantly realizes the homework automation potential of the machine, but the mean kid breaks it causing it to output wrong answers and he and his friend Irene must actually know their stuff in order to see that the answers are wrong and fix it. Finally their teacher basically tricks them into using it as a research tool and actually expanding their minds. I'll leave the implications for today's GPT-equipped kids as an exercise.

It's a great book and a great series. I wish more kids were exposed to its like today.

"We’ll never again need to hire someone to write a pretty good press release, a pretty good medical report or a pretty good investor deck. "

See also: glass houses, throwing stones.

ChatGPT has some interesting thoughts on the subject:

The argument that the end of the high school essay is a good thing is misguided. While there may be some challenges associated with the traditional essay format, such as difficulty in giving useful feedback and a lack of emphasis on deep thinking, banning the use of GPT-3 and other technology is not the solution. Instead, we should use this opportunity to find a better way forward.

First, we should acknowledge that the traditional essay format does have limitations, such as the difficulty in providing useful feedback and the emphasis on superficial elements such as typing and sentence structure over deep thinking. However, these limitations do not negate the value of the essay as a tool for developing critical thinking and writing skills.

Instead of banning GPT-3 and other technology, we should explore ways to use them to enhance the essay-writing process. For example, GPT-3 could be used as a tool for providing feedback and helping students improve their writing. Additionally, we should focus on finding ways to provide more meaningful and effective feedback to students, rather than relying on superficial metrics.

Furthermore, we should also consider the reasons why essays are assigned in the first place, and ensure that the assignments are aligned with these goals. Essays are not assigned to simply prove that students can read and write, but to encourage deep thinking, questioning the status quo, and the ability to develop and express persuasive arguments.

In conclusion, the end of the high school essay is not a good thing, but rather an opportunity to find a better way forward. We should not ban GPT-3 and other technology, but rather use them to enhance the essay-writing process and focus on providing more meaningful and effective feedback to students, while also aligning essay assignments with the goals of developing critical thinking and writing skills.

I took German in high school. My senior year German teacher was chill as all hell and there were only like six or seven people in the class. Our final exam was oral and we were supposed to answer a set of five prompts with a minimum of 2-3 minutes per prompt.

I was really dreading the exam because I’m a fairly shy person, but ended up loving it. After we got through the first prompt she ditched the exam and we ended up just having a 15-minute conversation in German. (My German was significantly more developed than it is now, after years of not practicing…) Was much more engaging and a better judge of my knowledge than filling in multiple choice bubbles.

I wonder if (for places that can afford it) we will go back to a model where students are required to be on-prem and speaking aloud. Significantly harder to cheat or fake answers. Extremely time consuming and also vulnerable to biased teachers, though.

Given that these engines know how words are associated but not their meanings, some amount of human editing is required to avoid bogus text. Things like a missing not or pro-/anti- switches or removing any biases in the training set. But an editor can look over a whole lot of generated documents.

I like the Sal Khan model personally, but given the difficulties with online school w/ regards to adherence and attention I'm not sure it universally scales. Add onto the fact that students might have questions and will have to save them for the day after, leading to being behind overall -- I'm not sure it extends that well.

Essays aren't fun to write and are obviously trivial w/ LLMs, but the in class solution could definitely work. In high school they're obviously shallow and formulaic, but idk if there's no benefit at all. Even writing shallow comments on internet forums benefit from practice writing coherent sentences lol

All this will do imo is push the needle. People will get better at writing and better at seeing that it’s an auto generated text blob. Also you can still quiz the kid on the writing and the understanding of the concepts: thinking that having to write about them (and forcing them to pad their thoughts to a certain number of words) is the only way to ensure understanding is Byzantine and conservative for conservations sake.

Perhaps there is a role for those writing tools which record everything you write as you go including edits, revisions etc. Faking this reliably will be pretty tedious. It'll be like showing how you arrived at the answer in maths. You submit the finished work and your recorded writing process. Not sure why this isn't already used, since it would stop people paying someone to write the essay for them as well.

Read...

Autocomplete: Coming to terms with our new textual culture

https://hedgehogreview.com/web-features/thr/posts/autocomple...

and

https://blog.ayjay.org/the-media-ecology-of-college-writing/

My current best use-case for chatGPT is to shorten emails. I’m the kind of person that will add a lot of prose to emails - I plug much of my writing into chatGPT saying “please shorten this email” and it works well. Not perfect (sometimes cuts out important info) but well! This is to respect my teammates times.

This is probably 10th+ article on the topic on HN. And still good part of the comments are focused on the essays problem.

Considering the rate of the progress, grading high students in the next 5 years is not going to be the most important thing to worry about.

My inclination would be to actively require students to use AI in drafting. It seems plausible to me that successful work in the future will require AI collaboration. The skill we are looking for is apparently output based (like much of the work we turn in for pay later in life - we get paid for output, not process) - getting to the final product with AI shouldn't just be rewarded, it should a skill that is actively taught. It's a tool kids need to learn how to be good with, akin to "googling", it has its limitations that kids should learn first hand.

Edit: typos... because this wasn't written by AI...

> There’s not a lot of evidence that getting good at writing book reports or regurgitated essays under typical high school conditions leads people to success or happiness later in life.

Possibly true, but most people never get good at writing. Or at reading, for that matter. Including college grads. Perhaps, for the ones who do, it does lead to success or happiness later in life!

The story with language skills isn't all that different than math skills: most folks top out around the middle grades, or maybe early high school, and never get much better. People just treat them very differently. They'll volunteer that they suck at math, while lots of poor readers and writers take any suggestion that they're not good at those activities as a grave insult. "Bad at math" is regarded as normal, while "bad at reading" is considered pretty close to "moron".

> The first challenge of the essay was the asymmetrical difficulty in giving useful feedback. 30 essays, 5 minutes each, do the math. It doesn’t scale, and five minutes isn’t even close to enough time to honor the two hours you asked a student to put into the work.

This is true. Grading in writing classes is incredibly time-intensive, if you want to do a good job of it. It scales very poorly, as some schools and teachers discovered when they tried to put higher student loads on remote language arts teachers during Covid.

> Students get more points for good typing and clear sentence structure than they did for actually thinking deeply, questioning the status quo or changing their minds. If you grew up in a household with verbally agile family members, you probably did way better on essays than your peers, but not due to much effort on your own.

Well, sure, but also grading standards can only be as rigorous as admin and parents allow them to be. Start demanding that kids get actually good at writing to receive good grades and you're likely to have a bad time. Grading must be tuned so that practically all kids who show up (and many who rarely do) will pass, and a very high proportion will receive A and B grades. Rocking that boat is not usually a great career move. You'll have admin and parents up your ass in a hurry.

> So, now that a simple chat interface can write a better-than-mediocre essay on just about any topic for just about any high school student, what should be done?

> The answer is simple but difficult: Switch to the Sal Khan model. Lectures at home, classes are for homework.

Mmm. Given the level of compliance I saw in college, when it came to reading assigned material, I'm not convinced this is a good solution to widely deploy. Or, the experience of many with remote school during Covid. It's worth remembering that tons of strategies that work great for ideal students fall apart when applied to average or poor students. I have a feeling this would just turn into the teacher trying to cram the lesson into the first 10 minutes of class, so their classes' test scores don't fall down to the Earth's core.

> In a school that’s privileged enough to have decent class sizes and devices in the classroom, challenge the students to actually discuss what they’ve read or learned.

That's... just expensive prep schools. Especially the decent-class-size part. Vanishingly few public school teachers have an average class size under 15—mainly special ed, or high school teachers who are lucky enough to mostly teach "smart kid" electives. Most are more like 18-23, and quite a few are a lot higher. Ask any teacher and they'll tell you class runs way smoother, discussions are better and the students more engaged, and they get more material covered when a large chunk of a class is out because the flu's raging through. Ignoring all the rest of this and just cutting the average class size in half would be a huge improvement, but it's also not going to happen. Meanwhile, yes, four- to ten-student discussion-based courses that expensive prep schools often use absolutely are a very good model—if you have the money.

> At first, this is harder work for the teacher, but in fact, it’s what teachers actually signed up to do when they become teachers.

> This is far less cohesive and controllable than the industrial model of straight rows and boring lectures.

This seems to assume at least better teachers, on average, and possibly more of them. Either of those means far more spending. It's not happening.

> If we want to train people to take initiative, to question the arguments of others, to do the reading and to create, perhaps the best way to do that is to have them do that.

This is what elite prep schools already do. We know that model's better—it didn't suddenly become true when ChatGPT came around. Public schools just don't have the money (or freedom to kick out highly-disruptive students) that is required to do that.

> We’ll never again need to hire someone to write a pretty good press release, a pretty good medical report or a pretty good investor deck. Those are instant, free and the base level of mediocre. The opportunity going forward remains the same: Bringing insight and guts to interesting problems.

I do wonder what's going to happen when poor readers and writers not only start judging AI writing, but their feedback (via which passages they choose to keep, if nothing else) is incorporated into the model to encourage it to produce more writing like that. I've known business owners who couldn't write coherent instructions for making a peanut butter sandwich—and I mean for a normal person to follow, not the excessively-precise kind that's sometimes used to demonstrate how computer programs work—and degree-holding managers and bureaucrats whose writing could be outdone, in terms of how well it communicates a message, by a somewhat-bright fifth grader. What kind of writing are they going to select, when they tell ChatGPT 8.0 to generate, say, marketing copy? Will it be "pretty good"?

For English speakers, of course, for now.