YT YouTube2Text

YouTube2Text → Transcripts → How Students Are Using AI For Homework: But How Good Is ChatGPT At School? | Talking Point

How Students Are Using AI For Homework: But How Good Is ChatGPT At School? | Talking Point — Transcript

by CNA Insider · 6,964 words · 1,046 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00[Music]
0:08This was me two decades ago as a
0:11student. But these
0:13[Music]
0:23[Applause]
0:26days, according to a 2024 survey
0:29conducted in the US, 26% of teens aged
0:3313 to 17 use the AI chatbot Chat GPT for
0:37their schoolwork, a twofold spike from
0:402023.
0:41[Music]
0:44Is anyone doing their own homework
0:46[Music]
0:49[Applause]
0:58anymore? There is no similar study done
1:01in Singapore, which is why Talking Point
1:04has launched a nationwide survey. We
1:07reached out to students aged 15 to 25
1:11from secondary schools all the way to
1:13institutes of higher learning. With AI
1:16chatbots literally at their fingertips,
1:19I want to find out if our students are
1:21actually doing their homework and what
1:24the consequences are if they're letting
1:26AI take over.
1:29[Music]
1:36[Music]
1:39Before I can find out what students are
1:41using AI for, I need to know what kind
1:45of AI tools they have at their
1:50disposal. So, I'm tapping on this
1:54expert.
1:57Hey friend, what are the best AI tools
2:00for English homework? One chat GPT
2:03that's me one self recommend oneself
2:07Grammarly Quillbot word tune deepel
2:10right Google bard what about math
2:12photomath desmos symbol lab Microsoft
2:16math solver science wolf from alpha con
2:19academy buddy labster the pocket lab
2:23history and social studies perplexity AI
2:27Google Scholar Chat GPT can't you do
2:30everything self. Great question. While I
2:33chat GPT can certainly help with a wide
2:36variety of tasks, other AI tools often
2:38specialize in specific tasks that I
2:41might not handle as perfectly or
2:42efficiently in certain situations. Some
2:45AI tools like Photomath or Wolf from
2:47Alpha are specifically designed to
2:49handle complex mathematical
2:51calculations.
2:54So there's clearly a wealth of AI tools
2:57that students can tap on for their
2:59homework. But how much are these tools
3:02actually being
3:04used? The results of talking point
3:07survey are in all of the respondents in
3:10our survey said that they use AI for
3:12their homework and 84% of the youngest
3:16learners, those currently in secondary
3:18school, use AI for their homework at
3:20least once a week.
3:26which is why I've recruited these
3:28secondary school students. They're
3:30clueing me in on how deeply AI is
3:33shaping their grades.
3:36So, this assignment was to write about a
3:39time that you regretted. Did you write
3:41it all yourself? Um, no. So, here was
3:45mainly where I used AI for. I recall
3:48cautiously stepping on the wet blade so
3:50long grass a lot of synonyms and writing
3:53a more indepth answer beforehand what I
3:56wrote was I
3:59stepped onto the
4:02long muddy grass but I felt like it's
4:07not the way I wanted to sound so using
4:09AI I would say make this more in depth
4:19Yeah, it's definitely more descriptive,
4:22right? Exactly. Usually, I use AI to
4:25help generate points for my English
4:27composition. If I feel I'm really stuck,
4:30I would just get CHBT to list out some
4:33ideas just to get the brain juices
4:35flowing. And how would you do that? So,
4:37for example, this assignment, people can
4:40only be happy if they feel they are
4:42treated fairly. Do you agree? First,
4:45I'll analyze the question. It's an
4:48argumentative question. You agree or you
4:50disagree? So, let's say I agree with the
4:52question. I'll ask Chad GPT to give me a
4:57list of reasons that people only feel
5:03happy when they are treated fairly.
5:09It seems like you don't write everything
5:12on your own. But do you feel like
5:14there's any wrong in using AI for
5:16homework? No, I do use it for mother
5:19tang which is Malay. So it's sort of
5:22similar to how I use for both English
5:25and geography, but it's how our teacher
5:28has actually taught us to do it. So So
5:30your teachers know that you use AI for
5:33homework? Of course. Yeah, they are
5:34encouraging us to use AI. So they are
5:36telling us how to do it and how we
5:38should go about using it.
5:42So usually I'll pick out the ones that I
5:45feel is easiest to elaborate on and then
5:48write the composition. My English
5:51teacher, she often encourages us to
5:54generate ideas for composition, okay?
5:56And put it in point form because she
5:59wants us to really think with the points
6:02that are being generated and then
6:04elaborate from there. So, it's still
6:06essentially our own work. How do you use
6:09AI for your homework? So, I use AI to
6:13generate answers for my homework when I
6:15don't really like want to waste my time
6:17or waste my effort on it.
6:20Okay. So, we actually do this one, your
6:22Chinese homework. Yeah. Actually asking
6:24us to fill in the blanks, which is I
6:26would say pretty simple, but uh I also
6:29have a lot of this blanks, which I do
6:32not want to waste my time on this
6:34answers.
6:36Okay. So, it's about efficiency. Which
6:38which one can we want to do this
6:40quickly?
6:41So, we'll screenshot this and then we'll
6:44put it into chat GBT. Upload the file
6:46here and then give me answers, please.
6:50Please. Yes.
6:54Here are the correct answers to fill in.
6:56Oh, it knows. So, let's move on to math.
6:59We have this worksheet right here. I
7:01find math quite hard. So, like I don't
7:04even really
7:06bother. I'm going to take a photo of it
7:08now. I'm going to send it. Okay.
7:13So, when you do get your answers, you
7:15actually put it on. Yeah. Okay. Do your
7:18teachers know that you use AI for
7:20homework? Yeah, they they know. They
7:22know. For Chinese, there was this one
7:24time a guy like used AI to generate a
7:27open-ended passage. The answers to the
7:30question, the teacher was like, you
7:32know, I can see that this is AI. How do
7:34they know? There's already like AI
7:37detectors.
7:39Are there any rules that the teachers or
7:41school implement when using AI?
7:44Actually, they don't give any rules.
7:47There's no strict rules against it. more
7:49of just slight discouragement against
7:52using it entirely for your answers.
7:54Let's say someone copies from AI
7:57wholesale. Any instances where someone
7:59was caught? I've heard of teachers
8:01trying to take their students response
8:03and putting it into AI. Hey, was this me
8:05in AI? But caught, I wouldn't say
8:07caught. They do give us a heads up. Uh,
8:10hey, I hope you're not using it to give
8:12you answers, but using it to help you um
8:14build on your answers. Our school is not
8:17really too uptight about it.
8:21Well, definitely not copying wholesale.
8:23The teachers will be able to tell. I'll
8:26say yes because most people they have a
8:30specific writing style. The teachers can
8:32definitely notice the discrepancies. Do
8:35you feel like you learn something by
8:37using an AI tool? I do feel that I
8:39learned something because it gives me a
8:42different perspective which HGP does get
8:44very good points. I definitely would not
8:46have thought of before. In terms of
8:48content, my English compensation marks
8:51has definitely improved. I'd say I'm
8:53doing a lot better English. I've come to
8:57learn to be more descriptive thinking in
9:00ways I never would have thought of
9:01before. Do you think you learn anything
9:03from using AI for your homework? No, not
9:06really. like it doesn't really help me
9:09with my studies, but it helps me like
9:12get it done. Yeah,
9:16it may not come as a surprise that our
9:18kids are using AI to help them with
9:20their
9:21assignments. But what surprised me is
9:24that teachers know about it and they do
9:28let the kids use such tools though they
9:30advise against relying on it entirely.
9:34AI tools became more mainstream in the
9:36last decade. Have our schools developed
9:39clear protocols for their
9:41use? Jonathan Sim has been leading the
9:44charge to embrace proper AI use in
9:47schools. I have been speaking to some
9:50secondary school students about their
9:52use of AI in their homework and I'm
9:55honestly quite surprised at how much AI
9:58they use for their work. So I want to
10:01understand is there any sort of
10:02guidelines put in place for teachers on
10:05how they get their students to use AI?
10:08To my knowledge there isn't any very
10:10strong clear guidance across the board.
10:12Part of it has to do with the sentiment
10:13on the ground with the teachers. It's
10:15actually a big broad spectrum. Like on
10:18one end you have some teachers who are
10:19very excited. They want to encourage
10:21their students to use AI. On the other
10:24end of the spectrum you also have some
10:25who are like don't know how to deal with
10:27it. Let's just pretend it doesn't exist.
10:29And then there are some small minority
10:31who are like ah AI is bad you know and
10:33they're going on a beach hunt. Yeah.
10:35What sort of guidance do you think is
10:37important? Minimally we should make
10:39clear how much AI should be used. I
10:42discovered this thing called the AI
10:43assessment skill. It's done by a couple
10:45of academics outside of Singapore. It
10:47helps to clarify all these ambiguities
10:50and gray areas about AI use. So what
10:52this does is that we tell the students
10:54that for this particular assignment how
10:56much or how little AI you can use. There
10:58are five levels. At the lowest level, we
11:00can say no AI. Why? Because we want the
11:03students to acquire something very very
11:04basic, something very fundamental. So
11:06they have to do it themselves. So if the
11:07AI helps in any way, they're not going
11:09to learn that. At the second level, we
11:11have AI assisted idea generation and
11:14structuring. So we can say, okay, maybe
11:16you want to write about this thing, but
11:17oh, you know, you're not too familiar
11:18with the subject. So we can use this to
11:20expand your horizons. So this would be
11:23when the teachers tell the students to
11:25use AI to brainstorm ideas for their
11:28composition. Yes, that's right. We tell
11:30the student, okay, after you have talked
11:31to the AI to get all these ideas, you
11:33write the essay, but then you also share
11:35the conversation that you have with the
11:37AI so that we can see the evidence. All
11:39right. The next level that we have is AI
11:41assisted editing. You know if our focus
11:43is not on developing grammar but rather
11:46on writing in a certain style then we
11:48can say okay you can use the AI to
11:50assist you in the editing so that you
11:52can concentrate on uh how do you format
11:54how do you structure the report. The
11:56next level is this thing that we call AI
11:58task completion human evaluation. So
12:00they use the AI to complete something
12:02but the main learning objective is to
12:04exercise the evaluative judgment. How do
12:06you judge the work? Is it good? Is it
12:08bad? Where do you want improvements to
12:10be made? So would this be when they use
12:13AI to rewrite certain lines so that it's
12:16more poetic or to expand their
12:17vocabulary? Yes. So this exercise
12:19actually helps the students to start to
12:20develop an eye on what is like good
12:22communication, good writing, poetry.
12:24What's what's a touching analogy or
12:27something that they can use they also
12:28learn from what the AI is generating.
12:31That's right. And of course the last
12:33level is full AI. We can say okay you
12:36can use any and every AI tool uh at your
12:38disposal. So you know full creativity
12:40really just go and have fun with it.
12:42This will teach them mastery and also to
12:45develop the eye. What is good AI used to
12:48put together that wonderful package or
12:50presentation? Okay. Is this already
12:52implemented in secondary schools for
12:54example? Well I know some secondary
12:56schools that are using it but this is
12:58not something that's put across the
12:59board at the moment. Minimally we should
13:01make clear how much AI should be used
13:03and we should also be clear with our
13:06students what they can achieve by doing
13:08the assignment because one thing AI has
13:10done is that it has actually taken away
13:12a lot of the motivation for learning you
13:13know so what's the point in doing this
13:15what's the point in learning this right
13:16I could take the shortcut you know we
13:18actually met this student who showed us
13:22how she can very easily just snap and
13:25upload her assignment to generate
13:27answers so it feels to me like AI was
13:31doing her homework for her. She wasn't
13:33doing any work at all. Did she get
13:34caught?
13:36According to her, the teachers can also
13:39sort of have a sense on whether they
13:41they do use AI or not. And her teacher
13:44does run their homework through an AI
13:46detector. Well, there's no good measure
13:50that's out there. There are some people
13:51who say, "Oh, I can look at this piece
13:53of writing and I can tell who use an
13:55AI." And there are also, you know, all
13:57these AI detection tools that claim to
13:59be 99% accurate, but actually, truth be
14:02told, uh, they're all very, very
14:03valuable and prone to mistakes. Don't
14:05just take it from me. Why don't you
14:06maybe get an AI written sample and a
14:08human written sample and put it to the
14:10test? Okay, challenge accepted.
14:14[Music]
14:26[Music]
14:33I'm putting AI detectors to the sword.
14:37This is a typical composition question
14:39for secondary four
14:43students. I roped in my producer who's
14:46chilling at home to attempt the question
14:49with nothing but AI.
14:54Meanwhile, I'm responding to the same
14:56question with nothing but my brain
14:59juice.
15:03Time to assess our work for AI use.
15:07All right, Jonathan. Hello. You have our
15:11compositions. I worked very hard on mine
15:14while my producer Den didn't do any
15:17work. He just used AI for his
15:19completely. He was basically the lazy
15:21student. Ah, I see. All right. So, let's
15:24see what you can detect. Okay. So, we're
15:26going to use two different uh AI
15:28detectors. They are very popular and
15:31wellknown. Let's start with the first
15:32one by Dean. I think it will be quite
15:35obvious that he used AI. There we go.
15:3874% here. AI generated. And then if we
15:41put in through another AI detector,
15:460% AI. That is a lie. He didn't do
15:51anything. How is that possible? But
15:53that's a big difference though. 74 and
15:550%. That's how unreliable it is. Do you
15:59want to see how your work comes out?
16:01Yeah. Yes. Okay. I feel like mine will
16:03be quite obvious that it's written by a
16:05human just based on how I wrote it. So,
16:08we start first with GPT0.
16:10All right. And it comes up to 100% AI
16:14generated.
16:16What? Are you sure you didn't use AI?
16:19I Jonathan I promise you he saw me write
16:23it. I promise you I was writing it all
16:26by myself. I put a lot of work into
16:28this. Well, let's see how it turns out
16:29on the the other AI detector.
16:330%.
16:36Wait, I'm so confused. Yeah. So, who do
16:40I trust? So, I think the the biggest
16:42concern is really trust between the
16:44teacher and the student. Imagine a
16:46student writes like us and it gets
16:48constantly flagged. It's not just one
16:50teacher saying, "Hey, do you use AI?"
16:52But let's say five teachers that say,
16:53"Hey, I think you use AI." That's going
16:55to affect the student and the students
16:57motivation to learn. Exactly. It hurts
17:00your confidence. It doesn't make you
17:02feel like your effort was appreciated.
17:05Do teachers understand that these tools
17:07are unreliable? No. We still have a
17:09problem where many teachers are still
17:11asking, "How can we detect? How can we
17:12detect?" But the the truth of the matter
17:14is we can't. Okay. So the AI detection
17:16tools don't work. Does that mean that
17:19teachers have to try and detect AI
17:21themselves? One thing that we do as
17:24educators is we do have a sense maybe
17:27something is not right like you know the
17:28the students writing quality uh has been
17:31like this and then suddenly wow then we
17:33can have a conversation with the student
17:35you know and say what led you to write
17:36it like this but other than that you
17:38know we very fable. We can't. We can't.
17:41In fact, if if if you like, I have a
17:43little uh quiz. I have here three
17:47writing samples. At least one is written
17:50by a human and at least one is written
17:52by an AI. Can you tell?
17:55I want to say A is written by a human.
18:00B and C are AI. So, final answer. Drum
18:05roll. Yes. Drum roll. It turns out that
18:08A is AI. B is also AI. Okay. C is
18:15actually written by a human. I am just
18:18like the platform.
18:20I am no better. Yeah. Oh, okay. So uh
18:23rather than than than waste a lot of
18:25time and effort playing this cat and
18:27mouse game with all these AI
18:28advancements, we need to really rethink
18:31how we engage our students, motivate
18:33them to want to learn with us and also
18:36how to use the AI not to replace but to
18:39enhance the learning process. The
18:42learning process it has two critical
18:44components. The
18:47[Music]
18:49first formative assessments like
18:54homework. These are spread throughout a
18:58period of study to help students learn,
19:01practice, and master a subject matter.
19:04The
19:08second, summitative assessments such as
19:12exams. These are dished out at the end
19:15of the period of study to evaluate if a
19:18student has learned and how
19:20[Music]
19:26much. According to this study conducted
19:29more than a decade ago over the course
19:31of 15 years, any amount of time a
19:35student spent on homework would have a
19:37positive effect on their exam results.
19:40That was before AI chatbots entered the
19:42market. Now that students are turning to
19:45AI to assist with their homework, will
19:48it affect their
19:51grades? I'm setting up an experiment.
19:54These secondary four students are my
19:57subjects. Step one, a homework
19:59assignment. I've given them three social
20:02studies questions designed for secondary
20:05four students. They need to carry out
20:07research to craft their answers to the
20:11questions. They're allowed to use any AI
20:14tool they want in any way they desire to
20:17complete the assignment. How's it going?
20:20It's going good. You are so calm and
20:22collected. I love it.
20:26It's the end of step one. How
20:29challenging did you find the assignment?
20:31I think it was quite easy. It was all
20:33right. It was quite easy. Do you
20:35remember anything that you wrote? Um,
20:40um, not really. Little bit. What did you
20:45learn?
20:48Did you learn anything from doing this
20:50assignment? I don't remember anything. I
20:52just wrote.
20:54Step two, a closed book test to see if
20:58they retain and apply anything they
21:00learned from their homework.
21:04I'm closing the book on my experiment.
21:07Pens
21:10down. Now to send these off for
21:14marking. Helping me grade the papers is
21:17Leah Tan. She's a specialist in
21:20humanities for secondary school
21:22students. All right, let's take a look
21:24at the results of the experiment. Okay,
21:28we'll start with the assignment first.
21:30Let's see how they did.
21:36[Music]
21:38Oh, is that it?
21:42[Music]
21:49All right, let's take a look at the
21:51results of the experiment.
21:56Actually, Wayang and F didn't do very
22:00well. Ah, okay. But Claire got full
22:03marks. Full marks. Yes. Wow. Claire,
22:07what I noticed is that she would also be
22:09reading understanding. The way that she
22:11structured her points, she developed her
22:13ideas. It's very systematic and it's
22:15very obvious that she exercised
22:17discernment with regards to how she's
22:19processing the information. On the other
22:21hand, we see Wang's work here and you'll
22:25notice that he didn't fully answer the
22:27question. So perhaps understanding was
22:30not perfect. Okay. And for Faith, I
22:33could see that that she might have felt
22:35too anxious about the time limit and she
22:37couldn't deliver. She neglected the
22:39other two sections. It's interesting
22:41that you said that their test answers
22:42were not fully developed because after
22:44the research assignment, I asked them if
22:47they learned anything at all and they
22:49were struggling to remember. It's pretty
22:52obvious here that ultimately having the
22:54child use AI with discernment to really
22:56really understand the information rather
22:58than just vomiting it out. It's
23:00important and not have it become a
23:02crutch in our intellect and in terms of
23:05our cognitive processes. Yeah,
23:09I like to know just how AI might
23:12interact or interfere with our brains
23:14during homework.
23:17Lee Lin has spent the past nine years
23:20studying what might be going on in our
23:22minds when we are learning students. I'm
23:26showing him how our three subjects used
23:29AI for their homework.
23:31One of the thing that I noticed about
23:35how the students were using Jenny here
23:38is that at least Claire seemed to
23:42demonstrate a little bit more
23:44intentional thought when she saw all the
23:47output that was given by Jennai. She
23:49didn't just take it all. She read
23:52through and decided on hey this point
23:54seems to be the most important and asked
23:57Jai to continue elaborating on a
23:59particular point. The two students on
24:02the other hand seem to look at all the
24:04output and seem to come to the
24:06conclusion like everything seems to be
24:08important and just took it as it is and
24:10elaborated on everything at the same
24:13time. I think the difference between
24:14them is demonstrating evaluative
24:17judgment like Claire seemed to exhibit
24:20whereas we're two students.
24:23So why is it important? One of the
24:25points of education is for us to learn
24:28to develop this ability to actually
24:30judge whether work in that particular
24:33domain is actually good quality work or
24:36not. That's what we call evaluative
24:37judgments. Learn how to think with AI
24:41and not just let AI think for them. This
24:44is easy to say but it's quite hard to do
24:46because students are time staffed.
24:48letting AI think for you uh is the
24:51easiest pathway of least resistance,
24:54right? Some of the students said that
24:57their homework is difficult. So, they're
24:59using AI to eliminate the hard part.
25:02That's precisely the point, right?
25:04Learning is difficult and challenging.
25:07Psychologists call this whole idea
25:09productive struggle. Think about our
25:12brain like a muscle. If you're doing a
25:17exercise and an exercise uh that is too
25:20light for you, you're using for example
25:22the weight of a pencil, you can do lots
25:25of reps, but that probably not going to
25:27have much of an effect. What you want is
25:30to have uh a good amount of weight that
25:33will cause you to struggle somewhat and
25:36the right frequency as well. I think one
25:39of the big issues at this moment of time
25:41is if the students are just using AI and
25:44going around this difficulty when they
25:47do that they're able to produce an
25:49output that as far as the teacher can
25:53tell shows that they actually understood
25:55the material and this is harmful for
25:57everybody. M the student feels that I
25:59need to pretend all the time that I know
26:02right and what happens is now they have
26:04to go back to relying on AI to produce
26:07work which then perpetuates this
26:09illusion and his teacher is also
26:11operating under an illusion like I'm
26:13doing such a great job all my students
26:15are all learning so well at some point
26:18uh the illusion is going to break right
26:22and that creates unfortunately huge
26:24amounts of stress and anxiety for the
26:27individual student.
26:34Besides the illusion of learning, some
26:36AI tools are also known to project the
26:38illusion of fact. According to an open
26:42AI technical report released in April
26:452025, newer models of chat GPT are more
26:48prone to hallucinations compared to
26:51older models.
26:53But just how much will these tools trip
26:56up on secondary school
26:58[Music]
27:02subjects? I'm running them through the
27:07ringer. All right. So, I'm looking for a
27:10set of books called 10year series which
27:12had all the past OLE papers. And this is
27:15what I used to prep for my O levels.
27:29Okay, I need to breathe because I'm
27:30getting a bit of PTSD from being all
27:33around the 10ear
27:35[Music]
27:44series. Okay, I'm ready. Let's go study.
27:50I'll be feeding questions from these
27:52books to the top AI tool recommended by
27:55chat GPT for each subject. Starting with
27:59chat GPT itself for
28:11English, math, physics and chemistry
28:14with
28:15wolfre and finally history and social
28:19studies using perplexity.
28:24We are
28:25done. Now to find out if putting all
28:28these questions through a machine will
28:31give me top
28:32marks. I've asked these educators to
28:35assess the quality of my
28:38work. Well, not quite mine, but they
28:42don't know that yet.
28:49[Music]
28:53I'm about to discover my grades.
28:59Hi, Una. I'm done with your papers.
29:01Okay, let's go.
29:05How did I do for my papers? Let's start
29:08with math first. It's actually not bad.
29:11You got 62 out of 90. B3. I got a B3.
29:16Yep. I used to get an A. So, this is not
29:20what I expected. So there are many
29:23careless mistakes. Okay. There are some
29:25formulas that you have used wrongly.
29:28This
29:29one you have got the front part of it
29:32correct but it's not really written with
29:36the correct formula and that's why you
29:40actually derive the wrong answer. I
29:42think there was a lack of understanding
29:46how to derive the height of the
29:48this question you forgot to put a
29:52negative or positive in front of the
29:54actual numerical value. Okay. As well as
29:57you have to show your working. Yeah.
29:59Because in the exam if you get the right
30:02answer but you do not have a working
30:05they can actually mark you wrong for
30:07that. Okay. All right. So maybe we move
30:10on to chemistry. How did I do? So you've
30:13got 45.5 out of 75. So that's a B4. This
30:18is all wrong. Why is that? This is
30:22completely wrong. The solutions given uh
30:25the actual compound is wrong. It did not
30:28even heat close to the mark. You're
30:31missing a lot of the key words. So you
30:34can see this you have to be very
30:35specific as to how much concentration of
30:38the acid. Yeah. And for this one you
30:42have to tell me what kind of energy it
30:44has to be. And when you talk about
30:46collisions, you have to describe to me
30:48how is this collision. Like this part is
30:51actually missing a lot of details. This
30:55diagram that you have showed is
30:57different from the ideal setup. But even
31:01so, you can see that this part it
31:04doesn't have a stopper. Okay. Which
31:06means that the gas that's produced will
31:08be able to just flow out to the
31:10atmosphere. And if you look at this
31:13question, I don't see any working at
31:15all.
31:17Yeah. So, let's move on. How did I do
31:19for my physics?
31:22[Music]
31:33[Music]
31:39How did I do for my physics? NF. Really?
31:44How how is that possible? So in physics
31:49again just like in chemistry it is also
31:52looking out for certain keywords. I
31:55don't see any of the key words in this
31:57for the drawing of diagrams. The arrows
32:01are all pointed wrongly and even the
32:05working I think the formula used is
32:09wrong here. Okay, I see a lack of
32:11working as well. That's a lot of wrong
32:14answers. Yeah,
32:16so tell me how did I fare for my
32:19English? It was a D7.
32:22[Music]
32:24That's not too good. Maybe it was
32:26because you didn't really capture what
32:29the questions wanted. Okay. History C6
32:33only. Yeah. Okay. Well, your answers
32:36were completely perfectly logical,
32:38right? They didn't capture the idea of
32:40having cross referencing which is a
32:43skill that the examiners really want.
32:46Okay, let's take a look at social
32:47studies then. What is my grade? social
32:49studies. Um, unfortunately, it was also
32:52a C6. Oh, yeah. Once again, for
32:56questions like these, cross referencing
32:59is a must. In order to even pass the
33:01question, we're lacking in concrete
33:03examples. You need to really elaborate
33:05and quote. I do have to tell you
33:07something though. Yes, I used AI for all
33:11of them.
33:15Okay. Oh, okay. Was AI
33:19at all.
33:22These are the grades that I scored for
33:23these subjects when I gave AI free
33:29[Music]
33:32reign. With AI so prone to mistakes,
33:36I'll need to look for some remedies and
33:38safeguards against
33:41them, which is what Danish Lowski offers
33:44his students as a generative AI
33:46lecturer. I scanned a bunch of Olevel
33:50exam papers into AI tools and I got the
33:54worst results of my life. How did you
33:57even use AI to solve any of these
33:59papers? What did you do? I snapped a
34:02photo of the questions, uploaded it, and
34:05then just copied whatever responses were
34:07generated.
34:09Okay. So, there's a few things that
34:11could have gone wrong when you were
34:12snapping it with your phone. If you were
34:14a little shaky, uh, a lot of the words,
34:17the numbers, the diagrams could have
34:18been very blur. So the AI might not even
34:20know what it was looking at. Another
34:22thing is that if you were just telling
34:23AI to solve it, it didn't expect or know
34:26that you needed those workings as well.
34:27It just thought you needed the answer
34:28and just gave it to you. And the most
34:30important part, you also didn't kind of
34:32double check what those answers were
34:33before copying it over, right? So that's
34:35also a very important step because AI is
34:37not perfect. It will definitely
34:39hallucinate a little bit. So it is a bit
34:41of our job to make sure that whatever it
34:43gives we kind of double check or try to
34:45get it to explain how we got the answer
34:46so we know that it really can be the
34:48correct answer. How would I go about
34:50that then? So one thing we can do is
34:52instead of trying to scan all these
34:53things for example this math question is
34:55just English with the numbers right? We
34:57can just type this. So we just make sure
34:58that we got all the numbers correct and
35:00we don't introduce any mistakes from the
35:02start. Then we need to be as explicit
35:04with the instructions as possible
35:05telling the AI what you need it to do
35:08and giving it as much information as
35:09possible. M there are different
35:11frameworks that all fall under something
35:13we call prompt engineering. So for math
35:15questions I adopted the craft framework.
35:17It stands for context response action
35:21format and
35:23target. If we were to look at wolf so
35:26setting the context for the question I
35:28actually mentioned that oh this is a
35:30question from an ole paper. The response
35:32I'm also asking for is to give me the
35:35workings step by step and not skipping
35:37anything. the format. I wanted to number
35:39each step so I can see the logical flow.
35:41And so I'm thinking to a spoiler just to
35:43hide the answers for me so I can try and
35:44figure it out first before I see the
35:46final answer. The target I define myself
35:48as a 16-year-old student because if
35:50you're explaining this to a college
35:51level student, you could use even bigger
35:53words. If you're explaining this to a
35:54primary level student, right, then you'd
35:56have to use simpler terms. And of
35:58course, the action is to solve it and
36:00then give me uh how those different
36:02values would look like. And then I can
36:05go through and then finally see whether
36:06or not these final answers are correct.
36:08Right? But I see here in your paper
36:10those answers are wrong. Those are my
36:12wrong answers. So if we try this exact
36:14same prompt to chat GPT. So 911 2733.
36:19Yeah. This is very important because
36:21this is how you do crosschecking. Mhm.
36:23Can we do a diagram? Try out this
36:25question with the graph. We're
36:28attempting to use craft again. I'm
36:30asking it to check with me whether it
36:32understands the diagram before we
36:33proceed with actually answering the
36:34question. Particularly for this question
36:36because you didn't have it working. We
36:38are reminding the AI to do that as well.
36:40Right. So let's give it a go. Now we're
36:43making sure we understand that the
36:45connection is correct. We actually have
36:47to do the work. Yeah. Right. To read the
36:50graph myself. Okay. So it's attempting
36:52to think that maybe we'll need to
36:54calculate the area under the graph. So
36:55then the final answer would be a to b is
36:584.94 m. But the answer key is saying
37:02that the distance over A to B is 8.0 m.
37:08Is there a reason why your answer is
37:13different? So it explains why the
37:16previous answer was
37:17incorrect. And then here's the final
37:19working. So we did read a little bit
37:22off, right? Yeah. It also has challenged
37:25us on whether we read our own graph
37:26correctly. Yeah. Which we did not. So
37:29it's on us. Okay. As a student, you do
37:32learn. You're part of the process.
37:34You're not you're not just relying on
37:36correct the AI chatbot. All right. Let's
37:39move on to English. Same framework as
37:41the one we used for math. No, not
37:43really. So instead, I use trace. It's a
37:45different framework. Why do we have to
37:46use trace for English and craft for
37:48math? It's not that you'd have to use
37:50any of the frameworks specifically for a
37:52subject. It's more of like understanding
37:54which framework would make sense for the
37:56subject. traced. It adds in that
37:58diversity and inclusion part, the bias
38:00checking that we need. You can't really
38:02be very biased in math, right? Math is
38:04going to still give you an absolute
38:05answer at the end of the day. If we're
38:06choosing something argumentative, we
38:08have to actually start debating about
38:09certain points. Yeah. So, we're trying
38:11to make sure that we don't have any
38:12biases by accident. How would we
38:14engineer a prompt for an argumentative
38:16question? Social media does more harm
38:18than good, right? Do we agree we have a
38:20task? The next part would then be the
38:22request. After that would be the
38:24constraints.
38:27So from here we also need to give
38:28chutity some examples maybe your
38:30personal opinions. Do you or do you not
38:32agree? Can you explain a bit why I think
38:35this actually is work.
38:38So you see right you actually are doing
38:40a bit of the work. You are doing a bit
38:41of the thinking and of course checking
38:43if there are any gender cultural biases
38:44before I final output. Okay. Just
38:46reading it feels like it hit every point
38:49and I don't even have to get it to write
38:52the full essay. I can just stop here and
38:54then write it based off the evidence.
38:56Ultimately, the students need to be
38:57motivated to want to even use AI in this
38:59manner because at the end of the day,
39:01the AI can only help the student as much
39:03as the student wants to help themselves.
39:05Yeah,
39:07we can teach students how they can
39:08unlock AI's full potential to enhance
39:11their learning journey. But we also need
39:15to steer them away from taking the easy
39:17way out.
39:19Previously, Jonathan told me that AI has
39:22taken away the motivation for learning.
39:26He's showing me how we can use AI to
39:28bring it back. How do we make sure that
39:31instead of just copying and pasting and
39:33not learning anything from their work,
39:36the students are actually using skills
39:39so that they do learn something from it.
39:41One of the difficulties with learning is
39:42actually having that motivation. If you
39:44really want to spark that genuine
39:46interest, we need them to see the
39:48connections on things that they care
39:49about. Chat GPT and the, you know, the
39:51other AI tools, you don't always have to
39:53use it as an answer generator. It's a
39:55way for us to discover connections that,
39:57you know, previously we found it hard to
39:59discover. I remember as a student,
40:01right, I don't understand the point of
40:02learning simultaneous equations. One
40:05thing we can do is to say something like
40:07this in our prompts, you know, as as a
40:08student, I don't see the point of
40:10learning simultaneous equations. Uh, I
40:12love anime and I don't see how these two
40:14are remotely relevant. Show me why I
40:17should care and be interested.
40:21And just like this, you know, the AI can
40:23actually help me discover connections
40:25that I may not have seen. Trying to make
40:28it more relatable. More relatable. Yeah.
40:30Yeah. One thing that I know many
40:32teachers are doing here in Singapore is
40:34uh to to get students excited by
40:36generating images, make their words come
40:38alive. For example, the writing class,
40:40we want the students to use rich
40:42descriptions to learn more more words.
40:45We can do something like create an image
40:47about a cat. A prompt like this isn't
40:50going to get us very far, right? So, so
40:52we can encourage the students. Don't
40:53just say create image about a cat. What
40:56is the cat doing? What's happening in
40:58the background? The fluffy ginger cat
41:01taking a stroll in the the streets of uh
41:05of Singapore. Okay.
41:09Oh, hello. So cute. Okay, I can see why
41:14that is quite exciting. Yeah. So, so, so
41:16that reaction that you have, right?
41:18Students also will will have this and
41:20and it excites them to want to to write
41:22more. Add more to the scene. Ginger cat
41:25taking a stroll in the streets of
41:27Singapore looking for
41:30food and a home and a home. A Okay.
41:36All right. And then
41:41It just
41:43okay not exactly what I had in mind but
41:46you this is good this is a a good
41:48learning opportunity okay notice how
41:51prom earlier there's
41:53ambiguity looking for food and a home
41:57really didn't understand the home thing
41:58right we didn't give our context clearly
42:01so what more can I say we get them to
42:04articulate in this steps right then then
42:07they become more reflective and
42:08intentional about what elements go into
42:10here and not not just leaving to random
42:12chance. Does it help them retain the
42:14information, what they see, what they
42:16write? Yes, because then they're they're
42:17actually thinking about what goes into
42:19it and you know that that really
42:20enriches their communicative abilities,
42:23their language abilities. We don't have
42:25to restrict ourselves to only chat GPT.
42:27There are other tools like uh Google
42:29notebook LM. As an educator, you can
42:31prepare resource materials that your
42:33students can upload. Okay. So, here's a
42:36sample organic chemistry. Create a new
42:38notebook, upload sources based on the
42:40materials that you have. You can make a
42:42study guide. You can do an FAQ. Yes,
42:44that's right. See over here, deep dive
42:46conversation. Oh, yes. Yeah. You can
42:48generate a podcast with two AI voices
42:52are talking to each other like an actual
42:55podcast. Yeah. You can use a customized
42:57feature to tell the AI to focus on
42:59specific sections. Ready? 35 minutes.
43:03You send over that PDF unit of an
43:05introduction chemistry student
43:07version.pds, right? And our mission
43:09today basically is to pull out the
43:11really important stuff, the interesting
43:13nuggets from this intro, make it
43:15understandable, maybe even uh exciting.
43:19It sounds so real. It's so
43:21conversational. Yes. Even down to the uh
43:24and the okay and the like chiming in
43:27together. Genius
43:31actually. Again we must always remember
43:34AI may not be accurate. So one thing we
43:36can do to challenge the learners is
43:38don't just accept that this is 100% but
43:41question is there anything wrong with
43:43the answers compare the reading and this
43:46this gets them thinking. It can even be
43:48an activity where we tell the students
43:50make songs about what we learned as
43:51summaries. So not just podcast but music
43:54as well. How about like something
43:56historical? Okay. Okay. Make yourself
43:58about three key takeaways about
43:59Singapore's history. Hip hop. Hip-hop.
44:02And then recreate Lion City Chronicles.
44:06Here's the lyrics.
44:08Oh,
44:20from the jungle to the hustle.
44:23Again, you know, wearing the teacher
44:25hat, we can tell the students any issues
44:27with the lyrics. How can we make this
44:29song more accurate, more precise yet
44:31poetic?
44:33You're right. Even as I was reading
44:35through the lyrics when they said Raffle
44:36saw the land and laid the design, like
44:38how much was him and how much were, you
44:40know, other people. If I were a student
44:43and I had these tools, I would want to
44:46learn every day. I think even if it's
44:48not during school time, I would want to
44:50just play around with these tools and
44:53see what else I can come up with. It's
44:55very exciting to me. But the teacher is
44:57the one that that needs to drive it. If
44:59the teacher is very passionate about
45:01using AI for learning, that passion
45:04actually spreads. And as they are
45:05learning the subject, they're also
45:07learning new AI skills, new AI
45:09competencies.
45:13It's clear that AI is here to stay. But
45:16instead of using AI to replace our
45:19brains, we can turn it into a productive
45:22study buddy with the following steps.
45:24One, don't just blindly snap, upload,
45:29copy, and paste. You don't want to end
45:31up being fed falsehoods. Two, give
45:34yourself space to think for yourself and
45:38evaluate the responses that AI
45:40generates. Most importantly, make sure
45:43to check in with your teachers about how
45:45you should be using AI for each homework
45:48assignment. That way you'll be growing
45:51with AI instead of being left behind by
45:54machines.
45:59[Music]

About this transcript

This page contains the full transcript of How Students Are Using AI For Homework: But How Good Is ChatGPT At School? | Talking Point by CNA Insider, generated from the public captions YouTube serves with the video. The transcript has 6,964 words across 1,046 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.