Live data from Hacker News

A high schooler writes about AI tools in the classroom

theatlantic.com

341–350 of 368 posts

Re: A high schooler writes about AI tools in the classroom

#341

Earlier quoted context omitted.

Neither? it's quite clear they're suggesting improving assessments. This will lead to upstream learning not being gamed.

So the option is to what, stop handing out homework? That would result in less education time. To clarify, I mean education time, not classroom time.

Continue to assign homework. Tie the homework to in class assessment such that if a student can do well on homework, without AI assistance, they are expected to do well on tests conducted in class, again without AI assistance.

Set homework grades to be a relatively small percentage of the final grade.

With the above framework, a student is incentivized to complete homework. If they cheat themselves and use AI, they'll do badly on the tests and badly in the class overall.

Tell the students about the above rationale. Tell them that they're not to use AI for homework, that you can't stop them from using AI, but that by using AI, all they get is a perfect score on homework and probably a bad overall grade.

Re: A high schooler writes about AI tools in the classroom

#342

Earlier quoted context omitted.

I’m not sure where the OP said that. Can you show us?

Sure. Current paradigm: Education time = time at school + time doing assignments OP said: > Obviously the answer to testing and grading is to do it in the classroom. So my question is, when is homework done? If it is being done at school, then our two options are to extend hours spent at school or give up time normally spent lecturing. I guess there's the alternative of getting rid of homework and only evaluating stu…

My take is here:

https://news.ycombinator.com/item?id=45141954

Re: A high schooler writes about AI tools in the classroom

#343

Earlier quoted context omitted.

> I'm surprised the answer of doing all exercises (including essay writing) in class is apparently not obvious. Because that results in less education time. If you do homework in class then you have to give up lecture time. Of course, the other option is to extend school time. Here's a good litmus test: if something seems very obvious, you're likely missing some hidden complexity. It's not a perfect test, but if it's…

> that results in less education time. If you do homework in class then you have to give up lecture time The most advanced classes I took, including in high school, made us do the reading and initial problem solving at home and then advanced problem solving in class. This was true for math, English and economics. Lectures with application combined. But that doesn't work if students don't do the reading. Just as lectu…

The most advanced and best classes I had were small, Socratic, and had take home tests. We were able to get through a lot of material, in a lot of depth, because people did the reading, but a big reason people did the reading is everyone liked the class and the professor (granted, this was college).

When I ended up teaching during my PhD I mimicked his style as best I could. Made my course very project based, made homeworks easy to get good grades but also included ways every student could expand on and gave lots of feedback. I like to think the students really liked me, as they would frequently stop by my office just to say hi and a bunch would show up the next term either showing me how they expanded their project or wanting to talk about how to do more or just general advice. YET only half the kids ever attended lecture, a third of kids chose to do a final project not much more complicated than homework, a few didn't turn in their final project, and 2 grad students complained to the department when I failed them for not turning in their final (they ended up being given Cs). This wasn't long ago, early GPT and tail/just post covid days.

There's just a time problem with doing the grading in class. You cannot cover as much material. An ideal class is students do reading before lecture, you go through the material together and have a healthy dialogue about where there is confusion, and then the students build on the solid foundation you created. This certainly works for high school and college, though I suspect not as well for lower levels due to lower independence. The unfortunate truth is that when teaching you're also teaching students a lot of auxiliary skills too, like time management and self-reliance. If you aren't teaching students these skills, where do you think they are going to get them? Sure, some will be able to learn them themselves, but you can't look at their success and claim victory through survivor bias.

But I don't think this is the whole problem.

I'll be honest. My experience with students, the big reason for them cheating is grades. Covid and GPT exacerbated the problems[0] (and created some new ones), but a lot stems from what was already there. We place so much emphasis on grades that this is valued more than the education itself. I've seen bright students that cheat because they feel overwhelmed. Because they know to get into the top colleges and top grad programs they need straight As. Strike that, they need a >4.0 GPA. They have to navigate the unknowns of which professors even hand out A+s, will forgo a better teacher for a teacher that gives more As, and so on. *They are not optimizing their education, they are optimizing their GPA*. Not because they don't care about their education, but because they do. Because everyone knows that the next rung of education is more important, so it is wroth forgoing some now to get access to more later. No one will say it out loud, but we all know even pretty mediocore students can play catch-up even up in undergrad and good students can do that in grad school. I'm sure if you randomly selected kids with GPAs >3.5 from high school and dropped them into your top universities you wouldn't see a big difference in outcomes[1]. I believe this stress is part of why some students just check out. But there is some aspect that is simple here: if grades didn't matter, there's no reason to cheat. I'm not saying to abandon grades, but I think it is worth reevaluating the system. I don't think patchwork solutions are gonna solve things.

All of this misses the entire point of education. Honestly, there's a larger crisis that's going on and it is that our world has just embraced Goodhart's Law as a good thing, not a warning.

[0] For example, that it is actually really difficult to punish cheaters. Any serious accusation needs serious evidence. Even more so when departments measure the amount of cheating by how many cheaters are prosecuted. That same metric hacking is why those students got Cs, just as much as it was that the chair was empathetic towards them. Part of that empathy being back connected to the importance of grades...

[1] Legacy students make this complicated but that's a whole other long conversation that mainly deals with connections.

Re: A high schooler writes about AI tools in the classroom

#344

Earlier quoted context omitted.

I have to present to a group of pissed off college faculty tomorrow. They're pissed because we ended our contract with the AI checking software. We ended it because it checks not for AI but for professional writing, good grammar and spelling, and professional non-conversational word choice. WHICH ARE THE THINGS WE'RE TEACHING THESE STUDENTS IN OUR CLASSES. I have to look at a room of mostly out of touch faculty and t…

Were I in your shoes, I would make a point of explaining what "style transfer" is. It was one of the first things LLMs were really good at. e.g. "Write a rap battle between Shakespeare and Plato". Academic prose is just style transfer. Academia filters pretty heavily for humans that can do this particular kind of style transfer well. Everyone on the college faculty will be 2 standard deviations of good at writing aca…

Update: I just cited 15 studies from prominent institutions indicating that these tools are ineffective and then explained that while they were not communicated with the way they want, there were 4 avenues they could've received that information if they were involved in shared governance on campus at all.

I decided just to close every door for a complaint. Not sure if it landed but no one said anything.

Re: A high schooler writes about AI tools in the classroom

#345

Earlier quoted context omitted.

> Do you harbour beliefs that conflict with your values? I'm not sure how long you've been human for, but this is in fact a common thing. Common for all living creatures really. Unfortunately we cannot always uphold the full idealized versions of our beliefs due to constraints of the world we live in. But on the other hand, if your beliefs weren't beyond our capabilities then we'd never improve. (I'll assume it is "n…

I can tell from the ad hominem that you have confused pop philosophy for the real thing. A book on ethics might be a good start. I suggest Blackburn. We'll just leave it at that.

Idk man, I didn't say anything that radical. We live in a world where the most conservative people tend to worship people that say feed the poor. That seems pretty hypocritical to me. All humans are to some degree, but there's a difference between internal and external based concessions

Re: A high schooler writes about AI tools in the classroom

#346

Earlier quoted context omitted.

When google came around it took me about 10 minutes to figure out how to use it. Further when I saw things in the search results that didn't make sense or were plainly wrong, I had the pre-gpogle critical faculty to question things. Do you think we are helping K-12 students by letting AI doing hallucinated thinking for them? What incredible "AI skills" will they be missing out on if we restrict the exposure? How to t…

The google comparison is superficial. The skill needed is understanding what different modes of AI can and cant do across different domains, knowing when to use it vs when not to, developing judgment about AI content that goes beyond simple fact retrieval. We are creating a massive competency gap by treating AI exposure as somehow more dangerous than social media, which we've already allowed to reshape adolescent dev…

I'm all for students learning the technology that makes LLMs work. That would go a long way, I suspect, toward students understanding what problems LLMs are a good fit for.

Likewise, using LLMs as an API for software they create to call sounds like it would give them insight into what LLMs are good for.

The act of just "conversing" with an LLM doesn't seem like much of a skill. I find it hard to reconcile the idea that one needs training or experience to use an LLM when contrasted with how LLM products are being advertised to the "everyman".

I simply don't buy that there's skill associated with using LLMs as an end user beyond the skills that you'd use for checking the validity of any other source. (Granted, everybody is pretty terrible at that anyway.) If anything, the LLM should be treated with more skepticism and subjected to more fact checking than human-created or curated sources.

The level of public LLM adoption tells me that they're not hard to use. The companies who make them are doing their best to make them useful for everyone. Any "moat" created by having "skills" associated with using an LLM will be drained. The companies want them to be useful to everyone, not just to people with "skills".

re: social media

Personally, I see "social media" as vastly more deleterious than LLMs alone. ("Social media" and LLMs, together, are a force-multiplier of badness.)

I already don't think there should be a place in schools for "social media", in terms of a curricular subject. I'd appalled if administrators approached "social media" as a part of the curriculum with the enthusiasm I'm seeing for LLMs.

Re: A high schooler writes about AI tools in the classroom

#347

Earlier quoted context omitted.

So the option is to what, stop handing out homework? That would result in less education time. To clarify, I mean education time, not classroom time.

Continue to assign homework. Tie the homework to in class assessment such that if a student can do well on homework, without AI assistance, they are expected to do well on tests conducted in class, again without AI assistance. Set homework grades to be a relatively small percentage of the final grade. With the above framework, a student is incentivized to complete homework. If they cheat themselves and use AI, they'l…

I think this is easier said than done. It pretty much sounds like you're suggesting there are pop-quizes. We've seen that style of teaching before. IME it isn't as effective. Though I'm quite suspect that a big part of that is simply coordination with other classes. All it takes is one teacher who thinks their class is the most important and not give enough wiggle room so that there is room for triage and expected life events. Just because there's theoretically enough time does not mean there is enough time. Think of it like lifeboats. Do you want enough lifeboats so that each person has a spot or do you want extra lifeboats so that in case one gets destroyed or in case a person is unable to make it to the other lifeboat that they will still survive? Over optimization ignores the noise inherent to reality.

Re: A high schooler writes about AI tools in the classroom

#348

Earlier quoted context omitted.

LLMs democratize personal tutoring. AI can massively improve education, if done right.

They are working well now because they live off the pile of human-created online content. Give it a few years. When all the AI has to feed on is the output of other AI bots, they will likely hit a reality wall and halucinate themselves into insanity. Forget new physics. It will probably be hard to get an AI to admit the earth isnt flat.

Web 2.0 is already doing a great job at "human model collapse". No generative content needed, people are consuming poor quality output of other humans and leading to a decline in discourse. Tik-Tokification.

Re: A high schooler writes about AI tools in the classroom

#349

Earlier quoted context omitted.

Don't move the problem. We weren't discussing world hunger.

Your objection to the first question is already captured in the ones that follow. Try to put some more thought into it, yeah? Okay, I'll tone down the snark (it's only there because you got my back up). Ultimately as you think about these questions, you will realize that the answers are not absolute, but based on degrees. You might think "hmm, I can't solve world hunger, but maybe I can help all the kids in my neighb…

I truly believe you do not know how you instigated the aggression. So allow me to clarify: people do not like to be treated like children. Even children don't like being treated like children, so regardless of if you believe we all are, you're going to get nowhere with that attitude. I also want to clarify that you aren't being arrogant, you're being pretentious.

Also, you seem to have forgotten we're having a very different conversation in another thread. Here you're acknowledging that people have desires beyond their capabilities. There you accused me of only comprehending pop philosophy for stating such inconsistency. Maybe it is you who were so focused on winning arguments that you lost what the arguments were even about.

Yes, I want to solve world hunger. I have no problem understanding that feeding children in US schools does not solve world hunger. But I also have no problem understanding the existence of time, as problems can't be solved uniformly nor instantaneously. There's no magic in the real world. Nor do I have a problem with understanding that this isn't a binary situation. Ensuring kids in the US get fed results in more kids getting fed despite there still being kids globally not being fed.

Idk man, there's no inconsistency here. Just because I'd like a billion dollars doesn't mean I won't be happy to get a million. Sure, long way to go, but it's a lot closer than I was.

It's only a line in the sand if you stop. Otherwise we call that "progress" ¯\_(ツ)_/¯

Re: A high schooler writes about AI tools in the classroom

#350

Earlier quoted context omitted.

Sure. Current paradigm: Education time = time at school + time doing assignments OP said: > Obviously the answer to testing and grading is to do it in the classroom. So my question is, when is homework done? If it is being done at school, then our two options are to extend hours spent at school or give up time normally spent lecturing. I guess there's the alternative of getting rid of homework and only evaluating stu…

I think your assumption is where this falls apart. To be clear, your assumption about where time is spent and how there can only be 2 outcomes.

Sure, you're right, we don't have to treat time as a zero sum game (normally I'm upset about non-zero sum games being treated as zero sum lol). But that's a different problem. Yes, we don't need to worry about time if our goal is to meet a fixed quality of education. You can increase the quality of education, getting more done in less time.

But that's a different optimization problem. My assumption here is that we want to maximize education, not meet a specific threshold. Especially if we're talking about the US. Maybe there is a specific threshold we want to reach, but I don't think we're close enough that this is the main concern.

So that's why I'm treating time as a finite and scarce resource.

And you're right to point this out. We're making different assumptions about what problem to solve and we should make sure we're not talking past one another. So I hope this helps clear up some of my assumptions.

Post reply on HN