Relentless advancement to produce new gen of 
seppos
I asked Wendy if I could read the paper she turned in, and when I opened the document, I was surprised to see the topic: critical pedagogy, the philosophy of education pioneered by Paulo Freire. The philosophy examines the influence of social and political forces on learning and classroom dynamics. Her opening line: “To what extent is schooling hindering students’ cognitive ability to think critically?” Later, I asked Wendy if she recognized the irony in using AI to write not just a paper on critical pedagogy but one that argues learning is what “makes us truly human.” She wasn’t sure what to make of the question. “I use AI a lot. Like, every day,” she said. “And I do believe it could take away that critical-thinking part. But it’s just — now that we rely on it, we can’t really imagine living without it.”
The whole point is that A) the goal of school assignments isn't to get the right answer it's to learn to understand the surrounding concepts and how to get the right answer in a more generalizable way and B) the students aren't learning anything if its copy pasted from an AI. And C) frankly the LLM doesn't usually "solve" it. Its outputs are often easily distinguishable, poor answers, that just look good enough at first glance to hit submit.
What about an LLM producing plausible output (the one thing it's built to do) in response to a prompt (the question/assignment) actually means the coursework is poorly designed?
I genuinely want to know your thought process here. Is it just that teachers should be expected to outpace cheating technology or that you genuinely think anything that can convincingly be done by an LLM isn't worth having a human do it?
Writing an essay on a topic is not just a way of assessing your knowledge of the topic, it's great practice for communicating your ideas in a coherent polished form in general. Just because an LLM can write something that sometimes passes for a human-written essay doesn't mean that essays are useless now...
where are you from this is not how schools operate in the US
unfortunately I am a burger american. There are good and bad teachers, and a lot of bad incentive structures and structural issues undermining education quality here, but that doesn't mean this isn't still ostensibly the goal of most assignments. Admittedly I went to supposedly good k-12 schools and the hit rate was probably still like 50% in HS, but it was better than that in college, even at a not at all prestigious school
the purpose of a system is what it does.
and my comment said that in >50% of my classes, what it did was genuinely foster learning in the way described. Not all of the evils of the US school system even conflict with this very basic model of learning, and regardless schools aren't a monolith. A blanket statement about how schools operate in the US isn't appropriate in this case, because it's a gross exaggeration of how useless they are and doesn't apply across the board.
edit: and to be clear I agree that classes/assignments that only foster learning in theory, are meaningless, but many still do in practice
If homework and other at home assignments are graded (not just scored for learning), the teacher is already following bad practice. As you suggest, homework is for practice and learning. So it doesn't need to be graded, just scored so that students know what to improve on. It is evakuations like tests that should be graded. Tests aren't foolable by "AI" so long as they arw done in person.
Also, as you mention, the better demonstration of learning is synthesis, to apply concepts in new situations (teaching is even better for demonstrating knowledge, but this is rarely evaluated). LLMs don't understand concepts, they are bad at any actual synthesis questions. They can merge vocabularies and patterns, ape narrative structures, etc, but are very bad at combining actual reasoning concepts. If a teacher's test grading (let's say it's online) is fooled by an LLM it is unlikely to be synthesis questions. More likely recall and patterns.
I'm not sure that I agree that the model "only tests are graded" is a good ideal. What is generally thought of as homework, sure, that is counter-productive to grade, but for things like essays, projects, etc. I'm not sure that's true. There is only so much essay you can write in one sitting, and the practice polishing, restructuring and generally just increased time spent thinking through that essay, that is afforded by it being done at home, is valuable.
And I think you underestimate what an LLM can (at least in theory) produce. Especially if you let students pick topics or take liberties with structure, etc. at all. What you're asking of teachers, when you say that students successfully using LLMs to pass their class is an indictment of their coursework, is an obligation to always provide sufficiently novel prompts and questions and such that an "AI" can't answer well.
I agree that an LLM probably can't convincingly synthesize two concepts that were both represented separately in its training data (though I expect they'll get closer to being able to pass this off for non-complex examples), but what if the synthesis itself was in the training data? In HS and undergrad level courses, how often are the topics at hand really novel enough to rely on that not being the case? Or how often is the syllabus really flexible enough to allow teachers to reframe all assessments into synthesis questions? And as these companies get better at incorporating fresh material, how often will teachers have to completely rethink their coursework to keep up. This isn't a treadmill that it's reasonable to expect teachers to get on or condemn them for being imperfect at detecting.
The problem isn't that teachers can't tell, it's that they can't prove it. The difference between a student who isn't really getting it 100% but is trying and one who used AI and the slop it put out doesn't quite make logical sense is not that cut and dry and they don't deserve the same grade.
As a matter of practicality, what you describe may become necessary for serious educational institutions, but I wouldn't lay that on the teachers or say that it's ideal in any abstract sense, absent LLMs.
Essays should be done in class if they are graded. Not to add a time crunch, but the opposite: at-home assignments leads to unrealistic and often classist expectations of homework time and parental/tutor support. Yes, spending time writing alone is valuable for learning. So is homework. Neither should be scored for a student's grade. At-home essays suffer from the same rampant cheating that homework does, which does a disservice to everyone involved in terms of learning. It's important to distinguish evaluations from the act of learning itself, the two are not synonymous: if students' essays are to be graded, they should be done under proctorship and with time and venue alotted for fairness, subject to special cases. Many standardized tests have essay portions and for all the problems with standardized testing, it is appropriate that they don't let test takers go home and mull over it for as long as their economic and support sotiation allows.
Writing essays in class runs into a time crunch in most primary and secondary schools because each class is alotted an arbitrary hourish window once per day. But there are schools that do 2+ hour sessions and have off days, making this practical. A student can write a rough draft one day, turn it in, get feedback, and then polish and turn it in for a grade. And universities can always dedicate appropriate amounts of class time, they just don't want to pay TAs for anything that can be turned into homework time. Too busy doing financialized real estate schemes instead.
Re: LLMs, they can produce essays yep. This is an indictment of a course that grades take-home essays. The course was already inappropriately constructed. The LLM didn't cause the problem here, it just exacerbated the existing problem that manifests as standard cheating (paying/bullying for essays), generally accepted soft cheating (parents write the essay), generally accepted classist legs up (parents help but don't write it/tutors do the same), and the inequalities in free time that impacts students heavily enough already.
Then it has a good chance of regurgitating it. But this is very close to reusing test questions, which is already bad practice and leads to cheating. It's true that an LLM will solve a problem that none of the students have seen if the teacher's strategy of synthesis is to Google for examples, though. That pary is unfortunate but an assessment shouldn't be done in the context where someone can use an LLM anyways. Either way it's not synthesis.
The topics aren't novel at all. But that doesn't really have anything to do with rote memorization regurgitation vs. synthesis. Synthesis questions are often new and different, even just changing the words used for a biological process for a question will strip memorization and force a focus on concepts. Add a follow up question to relate it to something else that was learned and you get synthesis. This is actually a very easy kind of test to write if you practice it.
Assessments should be done in-person. In-person assessments can include simple recall questions. At-home work that is simple recall questions can already be solved by just Googling things. And you're describing a problem in course design. It's not individual teachers' faults that schooling is broken.
Re: the rest, you seem to think that I am picking on teachers. Not sure why.