What a lovely essay. Reminds me of the way I loved the liberal arts growing up. I missed having classes like that in college (AP'd and ACT'd my way out of most requirements).
English teachers seem especially prone to that friendly and sporting demeanor the author has. Professors from the engineering schools are far more prescriptive, probably due to the nature of the material.
This essay resonated with me because it highlighted the similarity between AI-written texts, describing the result as a word salad. And this also reminded me about some words from my teacher of Russian Literature: that the "bright future" themed novels of the pre-WW2 Soviet writers — works produced under strict political control — read like one big novel without a beginning and an end, and not as separate works.
And this grayness and sameness is what happens when people are forced to "think" as a chorus, either by the authorities or their censorship, or voluntarily by using the same AI's help.
Altman’s analogy didn’t hold up. Calculators were uncontroversial
Calculators are uncontroversial now. But when they first became cheap and widely available, they were not allowed in math classes. Then only four function calculators, then graphing calculators. But still today, programmable calculators are prohibited in many academic contexts.
Turns out education done right is vaguely a speed-run of how the knowledge was developed. Adding calculating tools makes sense as you advance the the corresponding point in the process. Honestly, I think there should be a chunk of precal and calc where they use slide rules only, then calculators of increasing complexity (or just increasingly complex features of one calculator).
"When will I use this in real life" is a declaration that you have no expectations of learning the next lesson that builds upon this one.
The concepts of adding machines and calculators were also slowly phased in over the span of a century. The first commercially successfully adding machines hit the market in the 1890's, and pocket calculators took off in the 1980's. AI went from theory to answering hand written math homework questions from a photograph in a few years.
As a teen in the late 80's I had an HP calculator that I programmed to compute molecular weights given an input string like "H2SO4". It felt like having a secret superpower, especially when I participated in competitive exams. I was a very straightlaced kid and would not have used the program if it such things were explicitly forbidden, but as far as I could tell, they never were.
>> Altman’s analogy didn’t hold up. Calculators were uncontroversial
> Calculators are uncontroversial now. But when they first became cheap and widely available, they were not allowed in math classes.
The author of TFA means specifically for his cohort of students, not in general. He polled his students, and the result was that they thought calculators weren't seen as unethical but they were more skeptical/uncertain about AI. By his current students, now, not in general.
It's also an extremely misleading comparison. Basic calculator functions do not in the slightest replace anything taught in a maths class. Using ChatGPT not just to write entire paragraphs (replacing composition), or even providing the writer with ideas (replacing the creative aspects of writing) isn't comparable to adding two large numbers together.
The equivalent in maths would be if you handed students a theorem prover or have Wolfram Alpha give you step-by-step solutions and obviously nobody to this day allows this, because like ChatGPT for writing it'd defeat the point, that students think.
When I was in uni we were allowed basic but not programmable calculators during exams and a lot of CS classes even were pen&paper, if the prof was a bit hardcore
Totally correct. In the 90's as a kid in school using a calculator was highly debated amongst teachers and the ability to bring one out on your desk depended on the teacher.
In grade 2 i had a teacher who would say "I don't believe in erasers", you know, the things that "undo" pencil. As a ~6 yr old i actually didn't understand this phrase: "Well I have one, they're real!"
The point that you're (and everyone is) glossing over here is relative positions on the skill gradient.
A first grader probably would be prohibited from using any kind of calculator on arithmetic tests, 4-fn or not. But 8th graders are usually permitted scientific (non-programmable) calculators.
As you go up in grade level, you "get access to" calculators capable of functionality at the level below you. Because the point is that when we're educating students we want them to actually learn the subject matter, but once we've deemed them to have understood it and we have them move onto the next goal, we give them the tools to make that prior goal easier. We lessen the burden of the little mechanical concepts they already know so that they have an easier time becoming familiar with the next more advanced concepts.
AI systems are so much more advanced than what's capable on a TI programmable calculator. It's hard to draw clean boundaries around the tiers and enforce them by telling the model "help the user with tasks of tier 1-4 but not 5+". That's the issue, that it's really infeasible to strictly use them strictly as learning tools. You can almost do it with a lot of self-discipline and self-reflection to analyze your own workflow, but it's not generalizable across domains.
>>> imagine how radically math class must have changed when calculators became widely affordable
It didn't.
I was in math class when calculators were introduced. At least for high school level and beyond, the curricula were designed to make problems solvable without calculators, and they weren't of much use. This was still the case when I taught an undergrad college math class in 1997. Graphing calculators were allowed, and the kids who tried to use them just screwed themselves up.
I would have gladly changed the curriculum to use calculators and computers from the very beginning. As tools, and not just to administer the same old exercises and quizzes. Give them Jupyter Notebook. Math education has never been a success story.
Education faces a dilemma, which is that it has always used heuristics to guide study and assess performance. Exercises such as the "three paragraph essay" had no use in the real world, even long before AI could generate them on demand. When one of those heuristics is broken, another one has to be found. Even word processing forced teachers to grade papers on content, rather than mechanics.
> most students are overcommitted; college is expensive, so they need good grades for a good return on their investment; and AI is everywhere, including the post-college workforce.
Yeah. Overcommitted to partying and skipping class.
Has this author ever been to an average American university?
> There are valid reasons why college students in particular might prefer that AI do their writing for them: most students are overcommitted;
Tangentially: I've helped out some college students with mentoring and advice from time to time. One common theme I've noticed is that their class load virtually doesn't matter. They find ways to run out of time no matter how much free time they start with.
We all like to imagine the poor, overburdened college student working 2 jobs and attending classes to make ends meet when reading statements like that. But to be completely honest, the students like that usually have their time management on point. The hardest ones to coach were the students who had no real responsibilities outside of classes, yet who found their free time slipping through their fingers no matter what they did.
Among all of the other problems with easy AI cheating, I wonder how much the availability of these tools will encourage even more procrastination. Feeling like you always have the fallback option of having ChatGPT write the homework for you leaves the door open to procrastinating even longer
> I asked my students to complete a baseline survey registering their agreement with several statements, including “It is unethical to use a calculator in a math class”
Unless there was more to this survey, this wording seems misleading. In a college-level math class, using a calculator is a common expectation depending on the type of class and the problem. The students would probably think of their TI-89, not a magical AI calculator that could solve every freeform problem for them.
Here are some of the essay topics I had them read aloud:
[...]
I expected them to laugh, but they sat in silence. When they did finally speak, I am happy to say that it bothered them. They didn’t like hearing how their AI-generated submissions, in which they’d clearly felt some personal stake, amounted to a big bowl of bland, flavorless word salad.
But that's what it has been trained on - almost all academic writing is bland flavorless word salad, and this is extremely noticeable in title fads. I have a nearly decade-long game running with my friend where me make up absolutely bullshit concepts that could nevertheless be plausibly published in a journal, and the process has been going on long before that.
'Verbing the noun: towards a genericization theory of expressivity in high-entropy counter-heterogeneity' describes an ongoing problem in academic writing where novelty is deprioritized in favor of acceptability by an evermore tightly circumscribed set of peer professionals whose socioeconomic interests favor the establishment of intellectual stasis that maximally conserves positionality in a quais-Simmelian network space parameterized by income, tenure proximity, and citation count.
Or put more clearly, the more academics write to impress each other instead of to reach the public, the more generic their titles and language will be. Being able to parse and regurgitate wordy titles and abstracts constitutes table stakes in academia, so the incentives tilt toward burying the lede any original proposals as deeply as possible so as the minimize the career-damaging possibility of rejection on technical/syntactical grounds.
I understand why much of the discussion about AI and university education has focused on first-year writing classes in the U.S. Some of my own first experiments with ChatGPT in December 2022 were having it write school-like essays [1].
Over the past few months, I cotaught a university class in which we also had first-year students use and reflect on their use of AI in their classwork. But the context was different: the class was a seminar on science communication (how to make science engaging to children and the general public), and most of the time was spent with the students doing group projects. Also, the class was at a university in Tokyo, and we taught it in Japanese.
We have just started analyzing the feedback from the students, but my impression is that they were less conflicted about the use of AI for their group work than they might have been if they had been doing their projects as individuals.
Meanwhile, as the semester progressed, agentic frameworks started to mature. I spent a lot of time on my own experimenting with Claude Code and Gemini CLI. While none of the students in that class seemed to use them, it became clear to me that such higher-level cognitive tools will pose an even greater challenge to higher education than essay-writing chatbots do now.
AIs grading AI-generated essays looks like a recipe for model collapse. That's why we certainly need people who go after the "diminishing returns" of improving their writing skills beyond the "good enough" that AI delivers.
Should education systems aim for that for all of their students? Certainly, because AI alone is not sufficient to raise the bar. As impressive an AI is when it seems to invent a new molecule, it is still only possible because of the original works of many people.
Really enjoyed that. Shows the "messy middle" that people are caught in during this current wave of AI tech. I think one undeniable positive outcome has been the like collective introspection AI has sparked among so many people. With a tool challenging what it means to be "human" or "creative" (more so than prior technological advances), I've been seeing a lot of wonderful discussions, articles, videos, etc. with people wrestling with those questions and also just affirming their own singular voice and unique creative essence. It's been cool to see.
A lot of the purposes in education for which the use of AI would be considered "cheating" involve writing assignments of one sort or another, so I don't know why most of these education scenarios don't simply redirect the incentive.
For example, in an English class with a lot of essay-writing assignments, the assignments could simply be worth 0% of the final mark. There would still be deadlines as usual, and they would be marked as usual, but the students would be free to do them or not as they pleased. The catch would be that the *proctored, for-credit* exams would demand that they write similar essays, which would then be graded based on the knowledge/skills the students would have been expected to gain if they'd done the assignments.
Advantages:
- No more issues with cheating.
- Students get to manage (or learn to manage) their own time and priorities, as is expected of adults, without being whipped as much with the cane of class grades.
- The advanced students who can already write clearly, concisely and convincingly (or whatever the objectives are of the writing exercises) don't have to waste time with unneeded assignments.
- If students skip the assignments, learn to write on their own time using ChatGPT and friends, and can demonstrate their skills in exam conditions, then it's a win-win.
This all requires that whoever is in charge of the class have clear and testable learning goals in mind -- which, alas, they all-too-often do not.
Hand written essays in class. Short and longer form discussions, questions and answers in class. End of unit, term, semester, year. Interview on topic in spoken form.
> several statements, including “It is unethical to use a calculator in a math class”
That's too broad to be of any use, if the math class is teaching you to calculate in your head, then using any calculator is cheating. If the math class is teaching some algebra equation solving skills, then using a programmable calculator that auto solves them is cheating.
That's the similar issue with such experiments - they unfortunately aren't rigorous to provide any insight into education
It occurs to me that writing by those in STEM fields and those in the humanities is entirely different and each group dislikes the other’s writing. When I was in college, my professors in technical classes had no problems with my writing. After I graduated and I wrote some technical articles, my writing received praise from readers. However, when I took two semesters of mandatory English writing classes in college, my professors hated my writing and nothing I did made them happy with it.
When LLMs became widely usable, I was one of the people who really liked much of the writing that they did. I found it was relatively close to my writing style, which I consider to be good, despite the disagreement from those in the humanities. It was close enough to my own writing that I have even had people on Discord accuse me of using LLMs to write my messages for me, when I had not.
The linked article was clearly written by an English teacher. He criticizes AI-generated texts as “a big bowl of bland, flavorless word salad“. Now, there are many cases where LLMs output nonsense, but in cases where the writing logically flows, does not self-contradict in any way and avoids unnecessary repetition, “bland” and “flavorless” are good. The goal of writing is to convey information across space-time; writing that is “bland” and “flavorless” is the best way of conveying information.
I can see a number of things he did in his writing to avoid being “bland” and “flavorless”, and I consider them to be examples of poor writing:
He used dozens of idioms that make the text difficult for non-native speakers and unpleasant for native readers. He used a number of colloquialisms, including some that are inappropriate in professional contexts (although I will not repeat them since I refuse to write them). He used a word whose only definition is provided by Urban Dictionary and therefore is not even an official word:
He also brought politics into an apolitical topic. The injection of politics is a great way to derail any form of productive dialogue and should be avoided.
He used a story format of the kind that has infected journalism. It is very rare that the process by which something was learned is useful to readers and presenting it for dramatics wastes their time. That is with the exception of stories on the topic of security, where hearing the process is often genuinely informative. My aversion to this writing style is so severe that I have a standing policy to stop reading a news article the moment that I see that it uses this style for a topic that is not security-related. After reading his article, I will extend my policy to apply to essays by academics in the humanities too.
He made numerous attempts to evoke emotional responses to elicit agreement, rather than to make clear arguments based on facts. This is great for propaganda, but not so great for making points. Every one of these appeals to emotions is poor writing.
Beyond those things, he also did not properly cite sources in multiple places. To name a few, the quotes from Sam Altman and Annalee Newitz are uncited. As an academic, he should know better.
Some of these things might actually have places in certain types of writing. They certainly have places in propaganda. They also have places in fictional literature. However, they do not have places in attempts to argue a point.
I imagine if he corrected all of my criticisms, he would find the result to be “bland” and “flavorless”. That is how an attempt to argue a point should be.
> imagine how radically math class must have changed when calculators became widely affordable
I was there. We had been given slide rules, and a decent chunk of my 11-yr-old maths classes were devoted to teaching us how to use them. Calculators were banned because they "didn't teach us anything" (but somehow slide rules did? It didn't make any sense at the time either).
Over the course of the next few years, calculators became more acceptable, and by my 18-yr-old Maths A-level class, we were being advised on which scientific calculators to buy.
Thinking back, I have had some woderfull professors that had expertise beyond just published work and knew how to convey it.
Then I have had many bland curriculum reciters matching a semi decent youtube lecture, and some bad communicators where you were better of just reading the book they tried to teach.
There was also a few that couldn't really teach, but whose class was more like a standup comedy performance.
I'd say with the exception of the first and last group, ChatGPT would probably be a good if not better replacement for day to day teaching and mentoring.
> Reflecting on the fact that 3 credits at UVA costs me $5000+ and 2100+ minutes,” Drew wrote, “I do not believe I grew enough through this course for it to be worth it.” Having noticed only “incremental improvements in [his] writing and thinking,” he concluded that “I would rather have spent this large sum of money and time on a course that interests me and teaches me about my career aspirations, like the finances of real estate. If I need to learn to write, I believe AI can serve me well for MY purpose at a fraction of the cost
I find this a strange criticism. If you look at all of human written stories, the number of stories is quite limited too. There's a lot of retelling of "classic" stories (except in classical times they were probably retelling them too). This even applies to religious stories:
The kings/men who fight over a beauty (which can be a land/crown, or a woman) overcoming a monster/evil opponent or a series of opponents (The Illiad/Troy, Beowulf, King Arthur, ...)
The orphan/abandoned kid/extremely poor that conquers the kingdom, either through marriage or by conquering a (series of) challenges (Moses, Aladdin, Oedipus, Theseus, Heracles, Cinderella, the foundation story of islam/mohamed, ...)
The vizier/prime minister who decides he'll be king and becomes incredibly evil to achieve it. (robin hood, paradise lost (ie. the story of lucifer/the devil), queen esther, aladdin, MacBeth, ...)
The not-evil-but-quite-evil mother who sees her king/husband make a child, either with another woman or sees/fears she or her child will be put aside because of the other child and ... (Hera, Snow White, Medea, biblical story of Abraham, ...)
And then there's stories like Game of Thrones that are in large part a combination and integration of a lot of such stories: Circe and her children. John Snow, after being rejected, climbing up and up and up. Bran becoming king. Arya living through a sort of Herculanean heroic epic. And the kings fighting constantly.
Humans are clearly still a big step up even from state-of-the-art AI, but we are not infinitely creative like we like to think we are. It's a difference in scale, not a fundamental difference.
The calculator comparison is so tired and misleading. It's embarrassing to think that rote computation like calculators serve in most math classes is comparable to the "formulate an insightful answer to a prompt and express it clearly through writing" role an LLM fills in (especially early) humanities coursework.
I recently did a short stint teaching English to university students. This is English as a second language (possibly unnecessary context) but the level is fairly high. The course was hastily rearranged to prevent the students using llms. They had to hand write their essays under supervision, which made them a real pain to grade btw.
Then I was asked to review the questions for the final exam and I noticed some pretty weird constructions in the sample texts and sentences. Nothing completely grammatically wrong just unusual or semantically off. The example I recall was "... disproportionately affects men and women" but there were other worse examples. You can guess where this is going. But I knew that another native speaker had written the listening texts so I didn't want to directly criticise these, and I thought they might have been pulled from magazines or some corpus. But, of course, the course leader had generated them with chatgpt.
I guess nobody tried to adjust the temperature via the API then...
You can make something really unbland then!
Eg. "Write a 3 sentence story" Temperature=1
The old cat, Bartholomew, stretched languidly in a patch of afternoon sun, his purr a rumbling motor. A mischievous bluebird, emboldened by the cat's sleepy demeanor, swooped down to steal a whisker for its nest. Bartholomew's eyes snapped open, a silent promise that their game was far from over.
Temperature=2:
The old clockmaker found a gear he didn't recognize, its teeth shimmering with an otherworldly light. He fitted it into the grand clock tower, and as the hands struck midnight, the town square was bathed in a soft, silver glow. From that day on, no one in the town ever seemed to age.
Assignments should in total be worth ten percent of the total mark. Not too much to cheat over and waste your chance of valuable feedback. Not too little to ignore, so giving some short term incentive to actually work on them.
What a great application of AI in college course. The processor made them critically think about how AI may impact their writing, the value of it vs human writing. That's exactly what college should be doing, producing critical thinkers to navigate tomorrow's world.
49 comments
[ 2.6 ms ] story [ 77.0 ms ] threadEnglish teachers seem especially prone to that friendly and sporting demeanor the author has. Professors from the engineering schools are far more prescriptive, probably due to the nature of the material.
And this grayness and sameness is what happens when people are forced to "think" as a chorus, either by the authorities or their censorship, or voluntarily by using the same AI's help.
Calculators are uncontroversial now. But when they first became cheap and widely available, they were not allowed in math classes. Then only four function calculators, then graphing calculators. But still today, programmable calculators are prohibited in many academic contexts.
"When will I use this in real life" is a declaration that you have no expectations of learning the next lesson that builds upon this one.
> Calculators are uncontroversial now. But when they first became cheap and widely available, they were not allowed in math classes.
The author of TFA means specifically for his cohort of students, not in general. He polled his students, and the result was that they thought calculators weren't seen as unethical but they were more skeptical/uncertain about AI. By his current students, now, not in general.
The equivalent in maths would be if you handed students a theorem prover or have Wolfram Alpha give you step-by-step solutions and obviously nobody to this day allows this, because like ChatGPT for writing it'd defeat the point, that students think.
When I was in uni we were allowed basic but not programmable calculators during exams and a lot of CS classes even were pen&paper, if the prof was a bit hardcore
In grade 2 i had a teacher who would say "I don't believe in erasers", you know, the things that "undo" pencil. As a ~6 yr old i actually didn't understand this phrase: "Well I have one, they're real!"
A first grader probably would be prohibited from using any kind of calculator on arithmetic tests, 4-fn or not. But 8th graders are usually permitted scientific (non-programmable) calculators.
As you go up in grade level, you "get access to" calculators capable of functionality at the level below you. Because the point is that when we're educating students we want them to actually learn the subject matter, but once we've deemed them to have understood it and we have them move onto the next goal, we give them the tools to make that prior goal easier. We lessen the burden of the little mechanical concepts they already know so that they have an easier time becoming familiar with the next more advanced concepts.
AI systems are so much more advanced than what's capable on a TI programmable calculator. It's hard to draw clean boundaries around the tiers and enforce them by telling the model "help the user with tasks of tier 1-4 but not 5+". That's the issue, that it's really infeasible to strictly use them strictly as learning tools. You can almost do it with a lot of self-discipline and self-reflection to analyze your own workflow, but it's not generalizable across domains.
Yes, they are uncontroversially bad. Schools that don't use them have higher scores.
Unfortunately, even SAT/ACT have calculator slop now.
It didn't.
I was in math class when calculators were introduced. At least for high school level and beyond, the curricula were designed to make problems solvable without calculators, and they weren't of much use. This was still the case when I taught an undergrad college math class in 1997. Graphing calculators were allowed, and the kids who tried to use them just screwed themselves up.
I would have gladly changed the curriculum to use calculators and computers from the very beginning. As tools, and not just to administer the same old exercises and quizzes. Give them Jupyter Notebook. Math education has never been a success story.
Education faces a dilemma, which is that it has always used heuristics to guide study and assess performance. Exercises such as the "three paragraph essay" had no use in the real world, even long before AI could generate them on demand. When one of those heuristics is broken, another one has to be found. Even word processing forced teachers to grade papers on content, rather than mechanics.
Yeah. Overcommitted to partying and skipping class.
Has this author ever been to an average American university?
Tangentially: I've helped out some college students with mentoring and advice from time to time. One common theme I've noticed is that their class load virtually doesn't matter. They find ways to run out of time no matter how much free time they start with.
We all like to imagine the poor, overburdened college student working 2 jobs and attending classes to make ends meet when reading statements like that. But to be completely honest, the students like that usually have their time management on point. The hardest ones to coach were the students who had no real responsibilities outside of classes, yet who found their free time slipping through their fingers no matter what they did.
Among all of the other problems with easy AI cheating, I wonder how much the availability of these tools will encourage even more procrastination. Feeling like you always have the fallback option of having ChatGPT write the homework for you leaves the door open to procrastinating even longer
> I asked my students to complete a baseline survey registering their agreement with several statements, including “It is unethical to use a calculator in a math class”
Unless there was more to this survey, this wording seems misleading. In a college-level math class, using a calculator is a common expectation depending on the type of class and the problem. The students would probably think of their TI-89, not a magical AI calculator that could solve every freeform problem for them.
[...] I expected them to laugh, but they sat in silence. When they did finally speak, I am happy to say that it bothered them. They didn’t like hearing how their AI-generated submissions, in which they’d clearly felt some personal stake, amounted to a big bowl of bland, flavorless word salad.
But that's what it has been trained on - almost all academic writing is bland flavorless word salad, and this is extremely noticeable in title fads. I have a nearly decade-long game running with my friend where me make up absolutely bullshit concepts that could nevertheless be plausibly published in a journal, and the process has been going on long before that.
'Verbing the noun: towards a genericization theory of expressivity in high-entropy counter-heterogeneity' describes an ongoing problem in academic writing where novelty is deprioritized in favor of acceptability by an evermore tightly circumscribed set of peer professionals whose socioeconomic interests favor the establishment of intellectual stasis that maximally conserves positionality in a quais-Simmelian network space parameterized by income, tenure proximity, and citation count.
Or put more clearly, the more academics write to impress each other instead of to reach the public, the more generic their titles and language will be. Being able to parse and regurgitate wordy titles and abstracts constitutes table stakes in academia, so the incentives tilt toward burying the lede any original proposals as deeply as possible so as the minimize the career-damaging possibility of rejection on technical/syntactical grounds.
I understand why much of the discussion about AI and university education has focused on first-year writing classes in the U.S. Some of my own first experiments with ChatGPT in December 2022 were having it write school-like essays [1].
Over the past few months, I cotaught a university class in which we also had first-year students use and reflect on their use of AI in their classwork. But the context was different: the class was a seminar on science communication (how to make science engaging to children and the general public), and most of the time was spent with the students doing group projects. Also, the class was at a university in Tokyo, and we taught it in Japanese.
We have just started analyzing the feedback from the students, but my impression is that they were less conflicted about the use of AI for their group work than they might have been if they had been doing their projects as individuals.
Meanwhile, as the semester progressed, agentic frameworks started to mature. I spent a lot of time on my own experimenting with Claude Code and Gemini CLI. While none of the students in that class seemed to use them, it became clear to me that such higher-level cognitive tools will pose an even greater challenge to higher education than essay-writing chatbots do now.
[1] https://www.gally.net/temp/202212chatgpt/defaultessay.html
Should education systems aim for that for all of their students? Certainly, because AI alone is not sufficient to raise the bar. As impressive an AI is when it seems to invent a new molecule, it is still only possible because of the original works of many people.
For example, in an English class with a lot of essay-writing assignments, the assignments could simply be worth 0% of the final mark. There would still be deadlines as usual, and they would be marked as usual, but the students would be free to do them or not as they pleased. The catch would be that the *proctored, for-credit* exams would demand that they write similar essays, which would then be graded based on the knowledge/skills the students would have been expected to gain if they'd done the assignments.
Advantages:
- No more issues with cheating.
- Students get to manage (or learn to manage) their own time and priorities, as is expected of adults, without being whipped as much with the cane of class grades.
- The advanced students who can already write clearly, concisely and convincingly (or whatever the objectives are of the writing exercises) don't have to waste time with unneeded assignments.
- If students skip the assignments, learn to write on their own time using ChatGPT and friends, and can demonstrate their skills in exam conditions, then it's a win-win.
This all requires that whoever is in charge of the class have clear and testable learning goals in mind -- which, alas, they all-too-often do not.
AI can do the full assignment and do it faster and better
That's too broad to be of any use, if the math class is teaching you to calculate in your head, then using any calculator is cheating. If the math class is teaching some algebra equation solving skills, then using a programmable calculator that auto solves them is cheating.
That's the similar issue with such experiments - they unfortunately aren't rigorous to provide any insight into education
When LLMs became widely usable, I was one of the people who really liked much of the writing that they did. I found it was relatively close to my writing style, which I consider to be good, despite the disagreement from those in the humanities. It was close enough to my own writing that I have even had people on Discord accuse me of using LLMs to write my messages for me, when I had not.
The linked article was clearly written by an English teacher. He criticizes AI-generated texts as “a big bowl of bland, flavorless word salad“. Now, there are many cases where LLMs output nonsense, but in cases where the writing logically flows, does not self-contradict in any way and avoids unnecessary repetition, “bland” and “flavorless” are good. The goal of writing is to convey information across space-time; writing that is “bland” and “flavorless” is the best way of conveying information.
I can see a number of things he did in his writing to avoid being “bland” and “flavorless”, and I consider them to be examples of poor writing:
He used dozens of idioms that make the text difficult for non-native speakers and unpleasant for native readers. He used a number of colloquialisms, including some that are inappropriate in professional contexts (although I will not repeat them since I refuse to write them). He used a word whose only definition is provided by Urban Dictionary and therefore is not even an official word:
https://www.urbandictionary.com/define.php?term=cyborging
He also brought politics into an apolitical topic. The injection of politics is a great way to derail any form of productive dialogue and should be avoided.
He used a story format of the kind that has infected journalism. It is very rare that the process by which something was learned is useful to readers and presenting it for dramatics wastes their time. That is with the exception of stories on the topic of security, where hearing the process is often genuinely informative. My aversion to this writing style is so severe that I have a standing policy to stop reading a news article the moment that I see that it uses this style for a topic that is not security-related. After reading his article, I will extend my policy to apply to essays by academics in the humanities too.
He made numerous attempts to evoke emotional responses to elicit agreement, rather than to make clear arguments based on facts. This is great for propaganda, but not so great for making points. Every one of these appeals to emotions is poor writing.
Beyond those things, he also did not properly cite sources in multiple places. To name a few, the quotes from Sam Altman and Annalee Newitz are uncited. As an academic, he should know better.
Some of these things might actually have places in certain types of writing. They certainly have places in propaganda. They also have places in fictional literature. However, they do not have places in attempts to argue a point.
I imagine if he corrected all of my criticisms, he would find the result to be “bland” and “flavorless”. That is how an attempt to argue a point should be.
I was there. We had been given slide rules, and a decent chunk of my 11-yr-old maths classes were devoted to teaching us how to use them. Calculators were banned because they "didn't teach us anything" (but somehow slide rules did? It didn't make any sense at the time either).
Over the course of the next few years, calculators became more acceptable, and by my 18-yr-old Maths A-level class, we were being advised on which scientific calculators to buy.
It's an interesting analogy, as TFA says.
Then I have had many bland curriculum reciters matching a semi decent youtube lecture, and some bad communicators where you were better of just reading the book they tried to teach.
There was also a few that couldn't really teach, but whose class was more like a standup comedy performance.
I'd say with the exception of the first and last group, ChatGPT would probably be a good if not better replacement for day to day teaching and mentoring.
Somehow this hits hard
The kings/men who fight over a beauty (which can be a land/crown, or a woman) overcoming a monster/evil opponent or a series of opponents (The Illiad/Troy, Beowulf, King Arthur, ...)
The orphan/abandoned kid/extremely poor that conquers the kingdom, either through marriage or by conquering a (series of) challenges (Moses, Aladdin, Oedipus, Theseus, Heracles, Cinderella, the foundation story of islam/mohamed, ...)
The vizier/prime minister who decides he'll be king and becomes incredibly evil to achieve it. (robin hood, paradise lost (ie. the story of lucifer/the devil), queen esther, aladdin, MacBeth, ...)
The not-evil-but-quite-evil mother who sees her king/husband make a child, either with another woman or sees/fears she or her child will be put aside because of the other child and ... (Hera, Snow White, Medea, biblical story of Abraham, ...)
And then there's stories like Game of Thrones that are in large part a combination and integration of a lot of such stories: Circe and her children. John Snow, after being rejected, climbing up and up and up. Bran becoming king. Arya living through a sort of Herculanean heroic epic. And the kings fighting constantly.
Humans are clearly still a big step up even from state-of-the-art AI, but we are not infinitely creative like we like to think we are. It's a difference in scale, not a fundamental difference.
Then I was asked to review the questions for the final exam and I noticed some pretty weird constructions in the sample texts and sentences. Nothing completely grammatically wrong just unusual or semantically off. The example I recall was "... disproportionately affects men and women" but there were other worse examples. You can guess where this is going. But I knew that another native speaker had written the listening texts so I didn't want to directly criticise these, and I thought they might have been pulled from magazines or some corpus. But, of course, the course leader had generated them with chatgpt.
I guess nobody tried to adjust the temperature via the API then...
You can make something really unbland then!
Eg. "Write a 3 sentence story" Temperature=1
The old cat, Bartholomew, stretched languidly in a patch of afternoon sun, his purr a rumbling motor. A mischievous bluebird, emboldened by the cat's sleepy demeanor, swooped down to steal a whisker for its nest. Bartholomew's eyes snapped open, a silent promise that their game was far from over.
Temperature=2:
The old clockmaker found a gear he didn't recognize, its teeth shimmering with an otherworldly light. He fitted it into the grand clock tower, and as the hands struck midnight, the town square was bathed in a soft, silver glow. From that day on, no one in the town ever seemed to age.