# Discussion at the Principal's Office

**Sridharan**: [00:00] The National Boys and Girls Education Society was formed by Sir Sivaswamy Iyer way back in 1932. **The Lady Sivaswamy School was in existence even before that—1869 was the Lady Sivaswamy. We are the 12th oldest girls-only school in the whole of India.** At a time when girls were not allowed to come out of home, right? So we are very proud of that heritage and tradition. In the 1920s, I think if I'm not mistaken, Annie Besant was the president of the Lady Sivaswamy School, the society that was running it. There's a big history of all the who’s who of Chennai and those days obviously all the lawyers were ruling, so they were all there as various presidents.

**Sridharan**: [00:57] In 1932, the previous society was not able to run the school. They ran into some financial difficulty. That is when Sir Sivaswamy Iyer... **Sir Sivaswamy Iyer was one of the most well-known advocates in Chennai, who owned the New Woodlands property. Next to it, there is the Amalgamations' Shivsailam bungalow. It's a one-and-a-half-acre property, I think. It was owned by Sir Sivaswamy Iyer.**

**Anand**: [01:31] Okay.

**Sridharan**: [01:34] He sold that property because he wanted to construct something new here because that building was a very old building and all that. **You can't believe what I’m going to tell you: he sold that property and came here and rented a premises somewhere in this Sureshwarar Street so that he could take care of the construction part of it. That was his passion.**

**Anand**: [01:57] Wow.

**Sridharan**: [01:59] And the story goes that... So what happened was apparently Maha Periyava came in the 1930s. Sir Sivaswamy Iyer and his wife Lady Sivaswamy Iyer, Kalyani Ammal, they went to have a darshan. At that time, apparently Lady Kalyani Ammal cried in front of Maha Periyava saying, "Puthira bhagyam illaye" [Translation: I don’t have the fortune of having children]. **So apparently he said, Periyava, Maha Periyava said to Sir Sivaswamy Iyer, "Unaku kozhandhai illana enna, aayirakanakana kozhandhaiku padipu nee solli kodukalamae." [Translation: If you don't have a child, so what? You can teach education to thousands of children.] That is how he took over the society. He formed the society in 1934.**

**Sridharan**: [03:00] To Kalyani Ammal, he said, "Unaku kozhandhai illana enna, evvalo per... puthira bhagyam illadhaval irukalo, avaluku hospital katu." [Translation: If you don't have a child, so what? There are so many people who don't have the fortune of having children, build a hospital for them.] That is how the Kalyani Hospital came. So it's a great, you know... I studied in Annamalai University. Raja Sir Annamalai Chettiar had millions of rupees at that time in Singapore and Burma and all that. I'm not belittling this, but **when he had so many crores of rupees, he took a portion of it and built Annamalai University. But Sir Sivaswamy sold his property to build the school. So, such big-hearted people have done this.**

**Sridharan**: [03:52] And then for the 50th year, in 1989, from 1932 to 1989—the 125th birth anniversary of Sir Sivaswamy Iyer—in commemoration of that, this building was built for the CBSE school. It was actually converted. This was a part of Lady Sivaswamy School. At that time, it was [inaudible] the other extension. My wife, Raji, would say we used to walk from there and come here and do all those things, the primary and all that. So that is the background to our school.

**Sridharan**: [04:31] **CBSE is now going through a very major transformation on education. After so many years, they have now realized that children have to learn on a conceptual basis, it has to be on an experiential basis, and that exams—we have been terrorizing them for a long time. So they have to be allowed to naturally bloom.** Those sorts of transformations are all taking place one after another. Every week, CBSE comes up with a new system. So we are all... we are going through that transformation to be able to deal with the changes that are taking place.

**Sridharan**: [05:12] **And now, none of my teachers can ask any child to... in the old days, they used to ask them to write two pages of notes. Go home, do homework, read, and write. Now we don’t say that at all because everyone puts it into ChatGPT and brings it back. So we are frightened. We don't know how to deal with this animal called AI in the teaching-learning process.** That is why you are here. Thank you very much for coming. Shall we go? Thank you. We'll go to the office.

# Session Transcript

**Sridharan**: [00:00] This is a place where Anand can see all of you as he speaks. We need a laptop to be connected.

**Teacher**: [00:15] You need a laptop to be connected?

**Sridharan**: [00:18] Yes, yes. We have connected this CPU. If others can connect their systems, that will be even better.

**Teacher**: [00:26] We can connect a laptop. Actually, I have a straight connection from this CPU to that interactive panel. If you want to demonstrate, we will remove that HDMI cable.

**Anand**: [00:40] Sure, I don't want to demonstrate; I would like you to demonstrate. You can even do it on the phone, no problem.

**Sridharan**: [00:53] Next time when we do it, it's better to have it done in a classroom type of setting rather than this. Sir, come, please come. We are waiting for the Principal and Vice Principal to come.

**Teacher**: [01:21] Madam Principal, he just came from the room. Call him, so we can start.

**Sridharan**: [01:34] So in our organization chart, we now have the Principal and Vice Principal, and below that, we have split the school according to the CBSE regulations. They have come up with the "Foundational Stage," which is from 3-year-olds to second standard. Then they call it the "Preparatory Stage," 3 to 5. Previously, all of this was called "Primary." Now there is no such thing called "Primary." It's Foundational and Preparatory. Then we have Middle School, transitioning between a junior to a senior. Then, of course, Secondary, which consists of 9th, 10th, 11th, and 12th.

**Sridharan**: [02:12] According to that, we have four heads here. Apart from that, we have people who take care of... there's an Academic Coordinator, there's an Exam Coordinator, there's a Technology Coordinator. So, likewise, some of the coordinators... then there's a Quality Control, what they call "School Quality Assessment and Accreditation" (SQAA), there's a coordinator for that.

**Anand**: [02:37] Oh, like they do in colleges?

**Sridharan**: [02:43] Yes. What they do is make sure that the school has seven domains. In the seven domains, CBSE has given some recommendations saying that you have these domains called Academic, Infrastructure, Co-curricular Activities, Technology and Infrastructure, Governance, and so on. Under that, they have sub-domains. They say if you will be considered to be good if you have the following. Then you have four levels of grading: Beginning, Evolving, Experienced, and the top-most. So, starting from next year, there is going to be... I think there is going to be an inspection because they are now going to rank the schools. Like NAAC, the National Association for Accreditation of Colleges, this has been going on for the last 10 years. I came to know about it and worked on it because I was a financial trustee in Ethiraj College for four years. So, NAAC, there they give A, Triple A, Triple A Plus, and so on. They are doing that for schools. At some point in time, they are going to say if you are A+, you can charge this fee; if you are B, you can charge this fee.

**Sridharan**: [04:09] Okay. Good afternoon, everyone. I am very pleased to introduce Mr. Anand. Anand is here on a visit from Singapore; he is based in Singapore. I was just joking with him; he is an expert on large language models. Large language models, sorry. **I was telling him, "We are struggling with ordinary language already, and you are talking about large language."** (Original: *naanga ordinary language-ke kashtaputtu-ndu irukkom, neenga large language-nu pesareengale*) He is here to help us navigate through what AI can do under a teaching-learning environment. Related to that, any other subjects that come to his mind. **How can the school use AI?** is the main topic. But I leave it to him to decide a start and an end point. Broadly, this is what it is. I've already explained that all of you have some specific responsibilities under our organizational structure. So, without taking too much of your time, thank you Anand for being here. He is here on a personal visit and he has taken two hours of his time to come here to address all of you, for which I am grateful.

**Anand**: [05:36] Thanks, thanks. I have a small request. Is it possible to turn these two fans off? My voice is not very loud. At least these two fans over here, in case it's possible. Whatever is convenient for everyone. No problem, in case I'm not audible, let me know. I will try and be a little louder.

**Sridharan**: [06:00] If you want to move a couple of rows to here... some people can come here, some other people can come here in these two or three rows. Or if you are comfortable and can hear Anand's voice, yes, please go ahead.

**Anand**: [06:13] Only then will I be audible. (Original: *apo dhaan audible-aa irukkum*) Okay, then maybe we'll take a small break and put on the fan in some time. No problem. (Original: *problem illai*) I first wanted to say thank you. **Teaching is not easy. And I can't think of anything more fulfilling.** I have tried it and I am trying it. I'm an instructor for a course at IIT Madras for the Bachelors in Science program. But I don't interact with any students. It is a completely online, AI-driven course which I had designed. What you are doing, on the other hand, is so much harder and so much more meaningful. So, firstly, thank you for that.

**Anand**: [07:11] My voice is only this much, please come forward. (Original: *En voice-ye avlo dhaan, neenga munnadi vandhudunga*) Is everyone comfortable with Tamil? Is there anyone who is not? Everyone is saying AI, AI. It is important, but at the same time, **it's okay not to use AI**. Sometimes we should use AI, sometimes we need not use AI. **One of the things I'm very curious about is—everyone says AI will do everything—is there something that it will not do? Because that is what I want to teach the children. What it cannot do in maybe 10 years is a good thing to teach.** What it can do in 10 years may still be important to teach; nothing wrong with that. But what is it that AI cannot do? We are all trying to understand. Some things we know it cannot do—relationships, right? It cannot replace—it's not like you can marry a robot. You can, it's okay, maybe one day that will become a law, it's okay. But many of us are more likely to feel a father-child relationship with a human being, a parent-child relationship, a natural one. Like a friend... for many people, they want to treat humans also as friends.

**Anand**: [08:35] My aunt used to ask me, "Why didn't you come to the funeral?" (Original: *En chithi sanda pottu, "En da pona vaikku varala?"-nu ketta*) Who knows me, Auntie? So many people, everyone is waiting for you. That felt so good. So there are many relationships that exist because we value humans. That will last. In which subject do we teach this?

**Teacher**: [08:52] Environmental Sciences.

**Anand**: [08:54] Environmental Sciences maybe, Social Sciences maybe. So those may become a little more important, perhaps. Or some aspects of these syllabuses will become more important. Mental mathematics used to be in the syllabus, and it was a large part of the syllabus. I'm sure when you teach, mental mathematics is a smaller portion of that syllabus. It's okay. And mathematics has moved on to more abstract concepts. So, the impact of AI—what we'll be talking about mostly is how can you use AI. We'll talk about that. But I also want to keep at the back of our heads, **what does this mean for learning and teaching in itself, fundamentally?** What we teach, what we don't teach. Nobody has answers—maybe some people have answers, I don't have answers. You may have some points of view. Good, keep those points of view in mind. We can test it out and based on that, maybe change a little bit of what we are teaching if we find that something is anyway AI doing for us, we may as well tell the children, "Okay, use AI for this."

**Anand**: [10:04] This is because there are two problems here. One, we don't know what the answer is. Second, the answer keeps changing. I tried to work on it last month and it didn't work; this month when I try, it works. **How many of you have tried ChatGPT's voice mode?** Okay, that's about half a dozen people. Great. Today when I tried, there is a newer version of that voice mode, which is much better. Previously, the model itself was not smart. It was using a knowledge cutoff till July 2024. Beyond that, what happened it doesn't have in its brain. It can still search and find out, but not that smart a model. From today, it is now a much smarter model. So, today when I was coming, I was asking the same thing... (Original: *Inniku varumbothu adhaan kettuttu irundhen*) about this workshop. "What should I discuss? What should I learn from this group?" And it was very nice. Previously, with some auto sounds and all, it would get interrupted by, or it would not be able to use my previous chat. Now it said, "In your class, you are covering these sorts of things. Check with this audience whether it will be useful for you." It was not able to do this yesterday.

**Anand**: [11:13] So, what it is able to do is also constantly changing. In that case, **what strategy can we use to stay up to date? One thing that I follow is: I try and ask it everything. Everything.** And some it does well, some it does not do well. That is learning for me. Two kinds of learning. First learning is: maybe this is something I can teach. Because if I can't do it now, at least there is some value in teaching that. Second kind of learning is: today this seems to be the boundary of what I can get it to do. Maybe somebody else will be able to ask it in a smarter way and get it done, but at least it is the boundary of it and me. Tomorrow I will try the same thing again. Maybe because I've gotten slightly smarter, maybe it has gotten much smarter, tomorrow it can do it, so now I know. So at any point, **as long as I have things that AI cannot do, I'm happy**, because I can learn more about where I'm useful for something. And I can also learn when the models improve. If I don't have anything where the AI is not able to do, I will never know if it is improving.

**Anand**: [12:21] In software, that has become a problem. I have been programming for 30 years. I know software very well. Now I don't have many problems which I can give where it is failing. I don't like it that much, I can go that far, but I have therefore gotten to the stage where it is better than me in an area that I'm an expert in. Thankfully, there are three or four other areas where I'm reasonably good at, so I'm feeling okay about myself. But the reverse is somehow interesting. Let me take an example. **Contracts... even if you beat me, I won't be able to read them.** (Original: *Contracts-ellam soodu pottaalum enakku padikka varaadhu*) It is very difficult. I have no legal training. BPB is a publisher. They said, "Anand, can you write a book about LLMs and finance? This is the contract." Who will go and read this contract? It goes on for about 10 pages.

**Anand**: [13:28] So, at that time, I was referring to an OpenAI study. That study said, "We have conducted a test of experts versus AI in the following way: We took a list of occupations in the US and they know roughly how many people are in each occupation. They have taken the largest occupations, some 70-80 of those. In those, they have taken the experts. Each of those experts created five tasks. Those tasks were solved by five experts and five models—maybe more, maybe less. And then they had the experts compare the answers. How many... just rank order them: which is the best answer, which is an okay answer, not so good answer, evaluate it. They created a list of: in which jobs is AI already better than the expert? How often? 50% of the time is it better than the expert, 70% of the time is it better than the expert?"

**Anand**: [14:32] My profession is largely software. In software, 70% of the time last year, August, it was better. Now 95% maybe it has gone to. **Contract reviews and legal clauses and all was an area where it was only about 40% at that time. Personal finance advice was 70%. Auditing was only 20%.** So it varied, but here's the thing: **it is better than 40% of the lawyers.** I don't know any lawyer as such who can do a contract review. So, I gave it to ChatGPT. "Tell me what I should... review this contract." It told me some things. I could not understand head or tail of its review. So I told it that. Like I said, my principle is: Ask it everything. If I don't understand what it is saying, I ask it, "I don't understand what you're saying. Explain it to me in simple terms." It told me something. I said, "No, no, see, this is not what I want. We've been talking for years together. You sort of know me."

**Anand**: [15:40] And how do I know it knows me? In the settings, there's a memory option. I don't know how many people have seen it. You can go and enable that. Then it will look at all of your history. You can even tell it, "Tell me everything that you know about me." It says that and then I started correcting it. "No, no, this you didn't get right, this you didn't get right." So it is literally there. Worst case, I can go copy-paste it and put it there. I even told it why I'm trying to write a book and all that. So I said, "You know me. You read the contract. I don't want you to actually explain the contract, even though now I realize that is what I asked you to do. No, no, don't do what I asked you to do. Do what I need, which is: what should I tell BPB?"

**Anand**: [16:29] And it said: In this clause, tell them you are also going to be using this for workshops. They have a clause saying that you cannot use any part of the book for the workshop. But because this material comes from the workshop material itself, this is "prior art" and it should be excluded. "This is the first point you should make." Oh, really? Okay. When I went to them, they said something. I didn't understand the head or tail of it. I recorded it and gave it to ChatGPT. (Original: *Thalayum puriyala, kaalum puriyala, adhapdiye record panni ChatGPT-kitta kuduthen*) "This is what they are saying." And I told them, "Look, I will check with my legal advisor and come back to you." My legal advisor very clearly told me, "When they ask you for this point and this point, tell them this and this and this." I just read it out to them or sent it in an email. They were convinced because it came up with what seemed like a reasonable solution. It seemed to understand their perspective; it seemed to understand my perspective.

**Anand**: [17:19] So now I'm very happy that in an area where I don't know anything, I don't have anyone to reach out to for help, it is helping me. Like Google is helping me. Like that, there are 30-40 areas. I have no problem using AI in those areas. There are some areas where I may not be very comfortable using AI. It's okay. Where we use it, we use it; where we don't use it, we don't use it. That's not a problem. Now a lot of people are asking me, "How come you aren't using it for this? Why don't you use it for this?" I didn't feel like it. (Original: *Thonala*) And maybe I don't like it; I don't feel like it. It's okay. Maybe I don't need it. Maybe I will need it tomorrow. Maybe the time isn't right. I think that's okay. And this is... at least in corporates, there is a lot of pressure: you have to use AI more. Let me be.

**Anand**: [18:00] So where I'm going with this is: I'll tell you 30-40 things. **Ignore everything if you feel like it.** Don't worry about it. In case it is useful, take it. If it is not useful, ignore it. Anyway, everything that I'm going to be saying will have a shelf life of three months, maximum six months. It will vanish. So I would love to ask you: **how are you using AI? Where are you using AI? Where is it working well? Where is it not working well?**

**Teacher**: [18:34] I use it everywhere like you. I even ask questions about my pet, "Am I a good pet parent?" and all those things. I keep asking everything, whatever possible.

**Anand**: [18:46] Nice. Any interesting uses, something that surprised you?

**Teacher**: [18:49] No, the pet was unwell. So I kept giving questions. Finally, it said, "You're doing very well." I was doubting myself. "Is it me?" "You're doing very well. You're a brilliant parent." It appreciated me. I said, "Thank you so much." And I ask a lot of questions about education and my husband as well. So a lot of questions about health as well. So I'm quite convinced.

**Anand**: [19:08] So straight away we have three professions: one general physician, veterinarian, and maybe a counselor or a friend that encouraged her.

**Teacher**: [19:17] How about astrology?

**Anand**: [19:18] Tell me more.

**Teacher**: [19:20] Yes, actually it matches horoscopes perfectly. Wow! It gives exact predictions. I can even put a name board "Lakshmi Gayatri Astrologer." I can have a separate service.

**Sridharan**: [19:35] She is the head of our AI department. (Original: *Ava dhaan head of our AI department*)

**Anand**: [19:37] Wow, even better. It didn't say you're going to head the AI department. Actually, I was seriously thinking after retirement maybe using AI we can do horoscope predictions. And palmistry? Okay, you take a picture of the hand and...?

**Teacher**: [19:53] No, no, no. You give the details, just date of birth and birthplace. That is enough. It starts doubting about astrology. Yes, it's the best. Okay, wow, okay. And there is a drawback in that. In case if you are giving the pic of that horoscope and other things, what humans can do is humans will be able to read whatever is given. But it needs a clarity of the picture. You should give a good pic and other things are there. So repeatedly it keeps asking you, "Can you give me a good pic? I'm unable to locate where is Rahu or where is Ketu?" and all. But otherwise, actually, it is working extremely... 95% good.

**Teacher**: [20:33] Also, for the learning purpose also I guess. As we grow younger in heart and older at age, I think we depend on this not only for... of course all that is beautiful. Even I would try it maybe after SG's session. If I go unwell, I'll ask, "I'm really unwell, so can you give me a thing?" It really lifts up the spirits. It really lifts up. And so we do use it for different dimensions of teaching and learning also. Because the more the number of years that we spend teaching, the more the doubt we get whether we know the subject fully or not. And with the rising clarity among children also, we need to know many intricate details about every subject. I think we, like you, we also go and speak. I think my system knows the voice, my voice. The moment I switch on, "Hi Sangeeta" comes there. So I think it also learns from us. It learned a lot from me. Like it went wrong many times in calculations.

**Anand**: [21:42] Calculation it goes wrong.

**Teacher**: [21:43] Then it says thank you to me. Plan for a trip... Anand has made everyone an astrologer for education, teaching, and learning. (Original: *Plan for a trip... Anand vandhu education, teaching, learning-kaga AI-ah pathi pesalaam-nu vandhaa, ellarayum astrologer aaki vituttaaru*) But I know where he is going. Yeah, go on. Sorry, Madam.

**Teacher**: [22:06] But after one thing, it keeps asking us, "Are you convinced?" We don't expect that question from being told. Nobody is going to ask us, "Are you convinced?" "I'm convinced, do you want... are you convinced or do you want me to give you more options?" That's very nice. "Can I modify it? Customize it?" That's what no one else... we make presentations, picture descriptions are made easy.

**Anand**: [22:32] Picture description?

**Teacher**: [22:33] Description. For children, when we give it, if we want to give it in a language, it will mostly be available in English. When we want to change it to Tamil, it will change it if we give some options. (Original: *Description. Kuzhandhaingalukku kudukkumpodhu, ippo language naanga kudukkanum-na, maximum English-la dhaan kidaikkum. Naanga Tamil-la maaththanam-num-bodhu naanga sila options kuduththaa adhu maathi kuduththudum*)

**Anand**: [22:43] It changes in the picture itself? (Original: *Padaththulaye maathiduma?*)

**Teacher**: [22:45] Oh, nice. And now the multimedia process has become very easy. I can do any sort of editing—audio editing, video editing. Which is powerful.

**Anand**: [22:55] How do you do audio editing?

**Teacher**: [22:58] Audio editing is possible in ChatGPT Pro. If I feed in an audio, I say from this second to this second, cut... (Original: *Audio editing ChatGPT Pro-la possible. Naan oru audio feed panni, indha second-la irundhu indha second vara cut panni...*) and I'll tell it to remove the noise and I'll tell it to fusion some musical instrument. It does everything.

**Anand**: [23:14] Okay, this I've never tried. This is a very powerful idea. Have you tried song generation? (Original: *Song generation try panni irukeengala?*)

**Teacher**: [23:21] Song generation, we have different applications for that. AI applications are there. So with that, I've tried. Coming to ChatGPT, in my exploration level, I prefer that as far as ChatGPT is concerned, I have a data well in my hand. It is used for the enhancement of the data rather than getting the data from... which is... it is used for that enhancement. But for the creation, I prefer a different application other than ChatGPT.

**Anand**: [23:48] I've tried it with Gemini. It was an IAS officers' sort of 25th-year retraining or something. So I took a photo of all the attendees' names. Then I said, "Go search who these people are, what they've done, and create a song thanking them, a three-minute song." **It called out every person's name and thanked them for specifically what they had done as part of the... just one sentence about each person. But when played in the group, that power was phenomenal.** (Original: *Ovvoruthar peraiyum solli, it thanked them for specifically what they had done as part of the... just one sentence about each person. But adha group-la play pannumbodhu, that power phenomenal.*) Now that leads me to think, now children, they are already social media influencers for all we know, and using this sort of a thing. Before, I was preparing sketchbook assignments, sticking stuff, and so on. Today I'm guessing a lot of their assignments are more creative, more multimedia. Anyway, they are more creative, no question about it. Anyway, they have no constraints and they have better tools. In that case, I'm curious: **how can we ask them to do things even when we don't even know if it is possible?** I'm asking because before things were not changing so rapidly; we kind of knew and therefore were able to guide. Now we are in a position where we are needing to guide without necessarily knowing what is possible, what will happen. That's okay. If we are to do it, we'll do it.

**Teacher**: [25:41] How are you thinking about this? We can go to the fourth dimension. See, one or two dimensions can be physically gauged. Right? You can gauge it. Right? It could be this, a diagram, motion of a diagram, or a graph. Suppose you go to the fourth or the fifth dimensions, where you not only have your three axes plus time and space and all, but you have so many other axes coming into play which is very abstract. If you've watched the movie *Interstellar*, he speaks about the fourth dimension there. Like it was very... I had my students sit beside me to explain, "What is happening, actually?" (Original: *Enna da adhu angai, what is happening actually?*) No, that's the way. So for these, maybe you can... if you all experts try out and tell us whether we will be able to gauge it, to understand it.

**Anand**: [26:33] What I'm thinking is, we are all talking here. AI is as smart as us. Why don't we invite AI to the table? (Original: *Enakku onnu thonradhu, we are all talking here. AI is as smart as us. Why don't we invite AI to the table?*) Does anyone have ChatGPT on their phone? Okay, great. Open it and why don't we ask it this question now and let's see what it answers? And please speak to it. Let us hear it. Let us invite it to the table. (Original: *Adha open panni, why don't we ask it this question now and let's see what it answers? And please speak to it. Let us hear it. Let us invite it to the table.*) Why should it sit outside? (Original: *Adhu en paavam veliya ukkaandhundu iruppane?*)

**Teacher**: [27:08] Yes, as certain as we are as teachers, we are expert and a counselor. So I handle feelings, I handle feelings, personal touch. I always think that the replacement of this is our goals. (Original: *In the... see with that it can give.*) I was worried ChatGPT started singing. (Original: *ChatGPT paada aarambichuduchonnu paathen.*) So, if I say I have an issue, how to handle anger management? I can get theoretically a lot of explanations: you do this, you do this. But I have my back answer. When I am a small child, when I just put an arm and I calm... I always get this... this touch, right? This creates magic. This creates magic with that child. But that happens only between the client and the counselor. That won't be seen outside. **Do you think that AI can develop to such kind of a...?** (Original: *Epavume enakku adhu vandhuttu irukkum. Indha touch irukku illaya, idhu vandhu oru magic create pannum. Idhu vandhu andha kuzhandhai kitta indha magic create pannum. But adhu vandhu client-ukkum counselor-ukkum mattum nadakkuradhu. Adhu veliya theriyaadhu.*)

**Anand**: [28:23] Very good point. What I think is: why will I even, when I have a counselor, ever go away from a counselor? I will go to AI when I don't have a counselor. Okay, only during situations which... but has nobody else other than that? In that case, then there are some unfortunate children who don't have access to a counselor or for whatever reason they're not able to reach their counselor at that time, whatever. It's one more support. Hugging a doll... hugging a doll and mom are one and the same. (Original: *Andha bommaya kattikaradhu... bommayum ammavum onnu dhaan.*) Definitely, like a teddy bear. (Original: *Kandippaa teddy bear madhiri.*) Correct. But will we stop giving dolls for that? Because becoming over-dependent on that toy can also be bad and all that. But with dolls, we have spent millennia. We know how to deal with toys. We don't yet know how to deal with this new toy. We have to learn. We have to learn that too. (Original: *Bommayoda we have spent, you know, millennia. We know how to deal with toys. We don't yet know how to deal with this new toy. Kattukanum. Adhuvum kattukanum.*) But no, I completely agree. For that, will it never happen in the future where AI can never ever replace... I think in this AI, the voice... will it console in that voice? (Original: *Naan ninappen andha AI-la, indha voice irukku illaya, andha voice-la console pannuma?*) It will! It will! I don't know this, that's why I'm asking. (Original: *Pannumey! Pannum. Illa, idhu theriyaadhu, adhunaala dhaan ketkuren.*) It will, yes. And there's a difference between touching and speaking. (Original: *Pannum, aama. Adhu... and thodaradhukkum pesaradhukkum oru difference irukku.*) There's a difference between speaking and writing. So step by step it is also improving. And yes, that means that maybe there are 10%...

---

**Anand**: [00:00] ...that 10% of not-very-good counselors whom AI is better than. For example, the difference between speaking to a counselor over the phone versus speaking to an AI—maybe the gap is not that much.

**Teacher**: [00:13] Tele-counseling is the talk of the day.

**Anand**: [00:14] Correct. It is already happening. So it will happen to some extent. It also depends on what happens.

**Teacher**: [00:21] Children, they use it for... they give their lessons to get explained in simple terms. It does the job. So in that way, it is very helpful. But the same set of children, they give the questions that are given in the textbook. They get the answers and finish it off.

**Anand**: [00:33] That is a very good point.

**Teacher**: [00:35] So I'm very concerned about that. **How do I stop the AI, so that the student—the student cannot get answers from the AI? That's not right, no?**

**Anand**: [00:43] Let us talk about that. We'll talk about that first.

**Teacher**: [00:45] AI is sitting inside the room! (Original: *AI ippo room-kulle vandhuvittadhu*)

**Anand**: [00:47] Very good. But if anyone already has answers, please do share any thoughts on anything that it shared.

**Teacher**: [00:55] Whatever we've been saying, it just comes back in the same way.

**Anand**: [01:02] Sure, but the question being: **how do we get students to do things when we ourselves don't know what is possible?** Has it given any tips for that? (Original: *Adhukku edhaavadhu has it given any tips?*)

**Teacher**: [01:13] How do we... sorry?

**Anand**: [01:15] How do we teach things that we don't even know exist?

**Teacher**: [01:21] One of the... I want to share with you something. One of the points that comes up very frequently is, which I also keep mentioning to parents: today we are in 2024. Take a student who is in 5th standard. He has another—she or she has—seven years of schooling, four years of undergraduate—that's eleven—another two to three years of masters. So does any parent know how the world is going to be in 14 years' time from today? With the pace with which AI and other technological and other developments are taking place? So therefore, no point in... you know, you were saying that it is valid only for three hours or three months or whatever it is. So **all the things that we teach, if it is going to be valid only for a certain period of time, how do you prepare a child for 14 years from today?**

**Anand**: [02:30] Should we be asking that question to ChatGPT?

**Teacher**: [02:33] Let us. How do we talk to it? **Someone sit here and type.** (Original: *Yaaraavadhu oruththar inga ukkaandhu type padinga, madam.*)

**Anand**: [02:37] **Speak to it!** (Original: *Pesungov!*)

**Teacher**: [02:38] Speak?

**Anand**: [02:38] You speak, sir. Voice.

**Teacher**: [02:40] **I'm hesitant in English.** (Original: *Enakku dhaan English-la thaduppu.*)

**Anand**: [02:41] **Speak in Tamil, what's wrong with that?** (Original: *Adha Tamil-la pesungo, idhula enna?*)

**Teacher**: [02:44] Voice is available? And she can speak through the microphone.

**Anand**: [02:48] Let that be. (Original: *Adhu varalaam vidungo.*) No, no, we'll just take someone's phone and just talk to it. Sir, give it. Just a...

**Teacher**: [02:59] Just tell me, tell it that you're not Lakshmi Gayatri but someone else.

**Anand**: [03:02] It's incognito, sir. It won't remember also. (Original: *Incognito-dhaan sir. Adhuvum gnaabagam vachukkaadhu.*)

**Teacher**: [03:08] No, no, no.

**Anand**: [03:10] In English? (Original: *English-la-ya?*) Anything is fine.

**Teacher**: [03:15] How do we teach...?

**Anand**: [03:19] How do we prepare them for the future?

**Teacher**: [03:22] And can ChatGPT produce this quality student seven years from now?

**Anand**: [03:27] That's here. So how do we prepare them for the future in life? Let us find out. But this, I think, is a useful practice. Sometimes we'll feel like this. Just ask it. We don't have to read the answer. What I'm finding is I'm asking it so many things. **The answers are sometimes very difficult: do this, do that. 20% I take, 80% I leave. It's okay. Better not to take too much pressure on all of these things.**

**Anand**: [04:07] I'll tell you how I'm doing it. And I definitely want to learn from ChatGPT, which may take me maybe a minute. We'll give it a minute. What I'm doing is two things. One, I'm being very lazy. I tell the students: I've got a set of exercises and all that. I've told ChatGPT to create a bunch of other exercises. **But can you come up with something that your classmates haven't seen? For every one you do, one bonus mark.** 10% of the class will be very enthusiastic. That itself gives me a signal: who are the enthusiastic ones? "You've done this much better than I have." I'm just saying that it can be a part of how you teach as well. They are able to see things with less filters. Right or wrong is a different issue, but they have less filters than we do. So when I invite them to help the others to create a learning system for students, at the very least, it is less filtered.

**Anand**: [05:13] Second, I make sure that I get out of the way. When I am there, then I'll say, "Ah, no, don't do this, do that." (Original: *Illa, idhu pannaadha, adhu idhu solluven.*) We do this as a group exercise. One of the things that I find that the students do well, both from a teaching and learning perspective, is gaming. Today, co-creating a game is very easy. Publishing a game is very easy. So one of my assignments was: everybody likes games. **Can you create a game that will broadly teach this particular concept in data analytics?**

**Anand**: [05:51] The concept was: you need to make sure that you have all the relevant data before you do the analysis. And you need to know when you have enough data, when you don't have enough data. Okay, how will you create a game? It was only... I said, "Okay, fine, just create it and show me." It created a maze game. One web page, very simple. And in the maze, it's all like... as you move, you can see just the areas around you, those eight boxes around you. And that maze may contain a piece of data or not. If it contains a piece of data, you can collect it. Below, it will show as one row that you've collected. Like that, there are totally 50 possible rows, but you don't need all 50. You have to collect only the data that you need to answer one particular question.

**Anand**: [06:48] So the students would go wander, wander, wander, pick a data set—"Ah, okay, this is helpful," move, move, move. The trick is knowing whether they have enough to stop. Sometimes that can happen after 20% of data collection; sometimes that needs 70% of the data collection. And if they say, "No, I have enough," and they get it right, great. At the same time, if they take too much time, they lose points. They are playing a game. They are also sort of learning. Now how much of it is learning, how much of it is just fun... okay, this concept we could have taught them in maybe 15 minutes, 20 minutes. Is this an efficient way? I don't know. **It is definitely a different way.** (Original: *Fun-nu paapom-me, oru tharava try pannen.*) Good, at least now I know that I can create a game and just toss it out there. And I get a leaderboard of this one. And the students were very happy to include that as part of their assessment. In fact, I checked with them: you have a choice. At the end of it, tell me if you want this included in your final score or not. Overall they said, "Yes." Okay, let's do it. (Original: *Sari, pannuvom-me.*)

**Anand**: [07:48] So the kinds of things that we are able to create is increasing, that you already know. The things that we don't know that can be done is very high, that also you know. How to find out? My strategy is sometimes just ask the students. They are much smarter at this, they play with it all the time. But let's hear what ChatGPT says. Create a game? (Original: *Create a game-nu solla poringala?*)

**Teacher**: [08:08] Anything! Ask it what it wants to create a game about or how we should prepare children. Yeah, did you get the answer, Madam?

**Teacher**: [08:24] Yes, sir. It has given some answers. Learning how to learn, critical thinking, creativity, communication, emotional intelligence, ethics and responsibility, resilience. So "How do you prepare them for the world?" We don't even know what's going to happen in the next three months; how are we going to prepare them for life? Was my question. So it said all of these things. Humans skills. Humans can.

**Anand**: [08:57] And a good question... go on, please.

**Teacher**: [09:02] Next question is: how are we going to prepare children for the future based on professional suggestions? This is the first suggestion: AI and human collaboration, AI trainer, system designer, AI ethicist. Then in climate and sustainability: climate adaption engineer, renewable energy specialist, carbon management consultant, sustainable agriculture innovator. Then coming to digital and virtual world, it is like virtual world architect, digital content creator, game designer, XR experience designer. These are all suggestions from them. For robotics and automation, it is robotics engineer, autonomous system operator, smart factory designer, drone operations specialist, something like that. As far as education and human development, learning experience designer, personalized learning coach, educational technology developer. These are all its suggestions. **So how do we prepare children? We are going to induce: think well, learning, critical thinking, problem solving, logical reasoning.**

**Teacher**: [10:17] **Emphasis on creativity, design, innovation, entrepreneurship.**

**Anand**: [10:23] Interesting. Now we have an answer. What should we do next?

**Teacher**: [10:30] Train students on all those aspects. That's one possibility.

**Teacher**: [10:37] Humans skills. **Basic ethics are where AI gets it from.** (Original: *Basic ethics-laam eppadi AI-kitte irundhu kathuppaa?*)

**Anand**: [10:43] Good point. So let it be said, but we know what is important. Very true.

**Teacher**: [10:47] So categorize which is applicable to the facilities around us. See, depends on how... when we try to implement something, the possibilities to implement directly all these technical kinds of things, that should be seen.

**Anand**: [11:06] Got it. So in other words, filter for what we have facilities for. Good point. So...

**Teacher**: [11:13] The basic process, professionalism is the role we have actually. We have to channelize the process. The entire process—AI and student, AI and teachers, AI and human kind—we have to channelize the entire process. Yes, we can use it, but how are we going to use it? Teach them how to work with or use AI of the age. Okay. And on the other basic core engineering like the mechanical engineering, the basic computer science or finance for that matter, or power—all these sectors are the basic things for all what we need to do. So shouldn't we be training them on these lines also?

**Anand**: [12:00] Training on the fundamentals. Good point. So let me just pause here. The set of responses I see as being divided into: what should we teach and how should we teach. In 'what should we teach,' I'm hearing a range of four responses which is: one, it's given a reasonably good set of things—we should teach those. But maybe some of those, we need to look at the environment whether it's even possible to teach and filter. On the other hand, there are some things that we know are important, like human skills or foundations of engineering. **So it doesn't matter what it says, we know we should teach these.** That's another response. So it's ranging from 'let's teach what it says' to 'let's teach some of it' to 'let's teach what we know is important.' And all of these are valid.

**Anand**: [12:51] Now the good part is getting one extra opinion is helping our thought process. Even after hearing it, if we say, "No, this is not what we should be teaching," good. And if we say, "Oh, okay, here is something else I could teach," or half of it that I could teach, good. Either way, you have a point of view that gets built. Right or wrong, nobody knows anyway, and it's probably right on all sides. The other aspect of it is the process side of it, which is the 'how do we teach.' That becomes tricky and... let's talk about that also. **One of the things that you said, Madam, was assignment given—they are going to put it into ChatGPT.** "Write it and bring it." (Original: *Ezhudhi kondu vaadaa-nna...*) Okay, they will put it into ChatGPT and write it and bring it. Is that a bad thing? It's like saying, "Don't use a calculator." Exam hall, you can enforce it. **Sitting at home... if you say don't use a calculator, I can't enforce it, nor should I enforce it.** (Original: *Aathula ukkaandhuttu don't use a calculator-na, I can't enforce it.*) Come to class, write the essay, I will grade it and I know that that is without AI—that's all right. But the work that I ask you to do at home, I should assume that—not sorry, I shouldn't assume—**I should tell you to use AI.** What is the point of you taking a take-home assignment and doing it without AI? Then I am teaching you to not use a tool that you have access to, and it is a good tool. There's nothing wrong with ChatGPT. Now my problem is: how do I give an assignment? How do I evaluate it?

**Teacher**: [14:23] We actually know when... for example, I'm taking my subject. I'm a computer science PGT. So I used to teach a concept called 'stacks,' where one will come in front, the next person will go in back, where nobody can come go in between. First come, first served, in that basis. So if you ask ChatGPT, it can give only very limited examples. I can give that part... **because I have explored ChatGPT, I know that it has very limited examples.** So I have to give those types of assignments to my students where they will actually explore what are all the examples. So I was able to find out, you know, who has the actual original creation. One girl was telling me removing the bangle from hand, where only for... you know, it will not escape in between. Whatever I am actually feeding my hand, it will come in the same way outside. So that is a original idea, so I was able to appreciate her.

**Anand**: [15:30] Interesting. So what you're saying is: I have, at least in some areas, a way of giving an assignment and evaluating whether the student applied their originality or had delegated it to an AI, because in this area, I know that humans can have higher originality than AI. Which is great. **The number of people who have such problems they can give, and the number of problems that they can give, is small and shrinking.** We are, in some sense, fighting a good fight against a calculator. "I will give you a problem which you will not be able to solve correctly on a calculator." My math teacher would simply say, "Okay, what is 4 divided by 3?" Calculator users will say 1.33333333, but those who were sitting in the class and attending would put 1.3 with a line on top. Okay, so I can tell who has used the calculator, who has not used the calculator. Great. But did that help? Maybe to some extent. Was the point to teach me not to use a calculator? I don't think that was the case, because she asked us to use the calculator later. Point was also not to test whether we could do that particular division; there were other ways using, for instance, long division worked out mechanism that she was able to test it.

**Anand**: [16:59] So in this particular example, what was it testing? Not sure. And I think it was an evolutionary process. Calculator were becoming common enough that almost all the children were able to—had calculators at home and it was probably a shift. I'm thinking of this as: there are a set of problems that you have which the students can solve using ChatGPT. Okay, they are there. Maybe they are even mandated as part of the syllabus. You can't do anything about it. Okay, let that be. (Original: *Sari, adhu irukkattum.*) Then maybe there are some areas where you can come up with a problem which they are not able to solve with, or you can tell if they're solving with AI. Okay, let that be too. (Original: *Adhuvum irukkattum.*) I don't think that will be a growing set. **The set that I'm interested in is: when AI is smarter than the child—it's smarter than world's best experts in many areas, it will be smarter than the child in many areas—even then, I still want to teach the child something.** What do I teach?

**Teacher**: [17:59] Emotional intelligence.

**Anand**: [18:01] Emotional intelligence, yes. But my subject is maths?

**Teacher**: [18:05] Interest also. Absolutely. And the logical thinking will improve. Critical. **So when we ask what is the right question, the prompting is also important. Just by using AI is smarter. You have to use the right prompt.**

**Anand**: [18:20] Correct. Correct. Giving a concept and which we ask children to develop questions which is not already there in...

**Anand**: [18:28] Great ideas. Having the children develop the questions themselves. All of these are valid. You'll have lot of things to get into this topic. Probably that child would have known to certain extent on that particular topic. So once we give such exploration chances of that, he may go, he may end up with some 10 options. So he will know what are all there in that 10 options. Absolutely. Rather than telling them, "This is the thing that you are supposed to do," if we keep it open enough, children will come up with one final output. But then, during the process, he would have visited N number of options. Likewise, he'll also know where to stop.

**Anand**: [19:10] Fantastic ideas. Why don't we take this in two parts? Most of what you said is, I would say, how does one learn even while using AI? Today, I think this will be less than 50% of the syllabus, if at all. It's going to be very small fraction. And one of the things that obviously, based on what you're saying, this is going to be a growing fraction of that. Please do that. This is very important, but I want to discuss this second, in a short while. There is one other thing that I'm curious about, which is: subject-wise, I have to teach something, and I want to teach it in a way that they still learn what they need to learn, even with them sitting in front of ChatGPT. **Why don't we try and find out?** You can just name any subject and or specific thing that you're teaching in your subject, any one or two.

**Teacher**: [20:08] Weather conditions.

**Anand**: [20:09] Weather conditions? And what are we trying to teach them?

**Teacher**: [20:12] Factors determining and how we measure—measuring instruments of different rain gauge, anemometer.

**Anand**: [20:21] Wonderful. So here is my request. Let us all ask ChatGPT broadly along the lines of this question. This is... is there a better...? (Original: *Idhu vandhu... is there a better...?*) Let me then switch over to my machine. Okay, one thing that I will share... how do I connect my... one of the things that I strongly encourage is using a paid model. The main reason is... paid model. The 400 rupee or 500 rupee... (Original: *Paid model. Andha 400 rupee or 500 rupee...*) Airtel? No, that is the 'go' version. (Original: *Airtel? Haan, adhu vandhu go version.*) Even that is better than the free version, mainly because you have access to a smarter model.

**Teacher**: [21:09] Is that's an unlimited or...?

**Anand**: [21:12] There are limits. How much you use? I would recommend the $20, that is almost close to 2,000 rupees per month. If you ask me, **that is the best 2,000 rupees anyone can ever spend.** But the 500 rupee one is okay to start with, and then if you find that you're not using it enough, then it's okay. Why? Because the latest models are like professors, intelligence-wise. The lower-level models are roughly the level of a graduate or a college student. There is that much of a difference.

**Anand**: [21:52] So let me ask a question to this, and I will dictate it. So the question that I'm going to ask... is it online? Yeah. That I'm going to ask, I'm going to be very verbose. When I talk to it, I ramble. It's a good thing to ramble. See, I want to give an assignment to students. This is to a CBSE middle school or high school student in geography. So make sure that your answer aligns with the syllabus and subject and all of that. And we are teaching them weather conditions and factors that influence weather, as well as the instruments that are used to measure weather. All of this while we've had a way of teaching, and that way of teaching has kind of been messed up by AI, because they can go ask AI the answer to almost any question and come back with an answer, and I have no way of figuring out whether they've really learned anything. **My aim is twofold: one, they should really learn something; second, they should also score high in the exams because this is part of the syllabus.** Keeping both of these objectives in mind, I want you to brainstorm, ideate, come up with three or four things that I can do that in as take-home work assignment, activities. Doesn't also have to be individual, though I'd kind of prefer it to be individual, but it could also be group work, something that will, yeah, maximize both of these. Can you give me some ideas?

**Anand**: [23:40] Let's speak. (Original: *Pesuvom-me.*) I also don't... sometimes I just feel like rambling. What I find is that when I ramble, my thoughts also become clearer. And the smarter models, the paid ones, they are able to make much better sense of what I'm thinking. Now there is one other thing that I have learned over time, which is: sometimes I learn how to prompt well through experience. I capture all of that experience somewhere in a notepad. And then I paste it. One of the things that I have learned through experience is how it should brainstorm. There is a 'converge-diverge' theory, which is: you come up with lots of ideas without any constraints, maximize the variety of ideas that you come up with, and then identify which are the best against some criteria. I have put it into one standard prompt, and now I will put it as the brainstorming protocol. So it is automatically typed, the things that I normally use.

**Anand**: [24:46] I am using GPT-5.5 with high thinking. In the earlier screen, we only saw 'instant.' 'Instant' is very good for fast answers. This is better for smart answers, but it will take its own sweet time. Let it. (Original: *Pannattum-me.*) **What I'd love for all of you to do is try this with whatever AI, whatever model you have.** Let's try it. (Original: *Summa pannungov-me.*) Just ask it what is this kind of an exercise you can give in your subject, any one topic. Preferably the one you might take up next.

**Anand**: [25:31] Oh, okay, wait. I want to give an assignment to students... okay, publish this. Let's try again. The problem may simply be that my network connection is not that good. But we'll see. And I will also... so one of the things that I do is I never ask only one model important questions. I ask multiple models. So let's ask Claude also. (Original: *Claude-ayum kettu paapom-me.*) **The difference that I find is: ChatGPT will do exactly what I tell it to do. Very good at following instructions, which is perfect when I know what is the right thing.** Claude does the right thing even if that is not what I wanted it to do, but it may not do exactly what I tell it to do. Correct.

**Anand**: [26:42] So sometimes this is better, sometimes that is better. Sometimes I know which is better, sometimes I don't. Let's ask both. (Original: *Rendaiyum kettu paapom-me.*) And then what I do is take the answer from one and put it the other. "Claude told me this. ChatGPT told me this. What do you think is better? Take the better ideas, drop the worse ideas, you give me a better answer." That... and the models have very little ego that way. They will take the input from another... if I say, "This is my idea," then they overreact: "Oh, your idea is the best, you are the best person in the world!" (Original: *Indha idea... "This is my idea"-nu sonna then they overreact: "Yo, your idea's the best, you are the best person in the world"-nu aarambikkum.*) So I always make sure I tell it: this is some other model, same as you. No need to be so sycophantic and all that, just pick what is best. That is a good approach.

**Anand**: [27:28] And there are two reasons why I do it. One to get different opinions. Second, also to save me time reading that answer. (Original: *Paththu பக்கம் ezhudhum.*) **I don't have time to... just tell me what to do. Actually, don't even tell me what to do. Do it without telling me. That is the best.**

**Anand**: [27:50] So sometimes we should probably also try and see how far you can push it. Maybe after this, I will say: whatever you've chosen, pick the best one out of that, create the assignment and give it to me in a form that I can just print and give it to the students. **If we do it like this...** (Original: *Ippadi pannom-na...*) that leads me to think, and this is something that I do in my classes: I just give them some problems which I have not really carefully reviewed. Let the students take a look at it. Sometimes they say, "Sir, you said that there will be one attachment, there is no attachment." **Let's see... okay, ChatGPT, where is that attachment?** (Original: *Paakalaam-pa... sari okay, ChatGPT where is that attachment?*) "Oh, I gave it to you earlier." (Original: *Oh, I gave it to you earlier-nu sollum.*) So error hunts! I do it, which is an interesting thing, right?

**Teacher**: [28:36] But mostly in that case, if you generate the marking scheme, the marking scheme is full of holes. Though we say it's filterable format, in physics and all, it's such a code-based marking scheme that we get, that we need to analyze it and rewrite it.

**Anand**: [28:55] Which is a problem, right? I just tell it that. I tell it, **"The whole thing is rubbish. Fix it."** And tell it how. And after that, if it is not able to do it, then okay, I have only wasted few minutes—that way I'm not very unhappy. I'm happy that I've discovered something that it is not currently able to do. Write it down. (Original: *Ezhudhi vachuppen.*) Here is something on the list on 9th of July, I tried this and it is not able to do. Three months later I will try it again. If it has started working then, "Ah, one more thing that AI has improved on, now I have learned the..." and I have learned something new also.

**Anand**: [29:36] As you noticed, ChatGPT is taking a long time. I'm not sure it is actually doing what I wanted it to do because it may have bungled up when starting. (Original: *Aarambikkumbodhe it may have bungled up.*) But what I'm going to do is take Claude's response. And you can see that this is a reasonable response, but okay, let us take one of these. [30:00]

---

**Anand**: [00:00] ...**7-day home weather log and prediction game.** Okay, that is not a bad idea. You measure the weather at your house. Each student records the weather at their own home twice a day for a week. This is slightly biased because I am a data scientist. I love data. So it is looking at my history and this is the kind of assignment that I would give.

**Teacher**: [00:26] It knows you.

**Anand**: [00:27] Correct. It may not be the assignment that you would want to give, but it is not a bad assignment. So they keep temperature using a cheap thermometer or a weather app—well, okay, there are many ways you can guide them, and we can also ask it how they can easily do this. Wind direction is very easy. Okay, cloud type—**take a photo of the sky with a timestamp**—or signs of humidity. And then they predict tomorrow's weather. Okay, and write one reason. The next day they score their own prediction. Prediction-wise, we can just ask them to upload it or submit it or whatever, so they cannot back-fill the predictions. That is easy to implement. And what made my predictions go right or wrong? For this, please ask AI for as much help as you need. That's a good one. And yeah, AI cannot fake sky photos—well, but if they are smart enough to get AI to fake sky photos, good, give them a bonus mark for that! Not a problem.

**Anand**: [01:31] Bring one instrument and then report its error. Every student brings a simple instrument. Oh, okay, effectively you are almost making an instrument, and don't just build it, but actually compare your reading with the official data. Okay, slightly more effort on their part, but will still teach them something fundamental. **AI error hunt: You ask AI to write an answer about weather instruments and factors, but then you plant five mistakes in it. Students must find and fix every error, citing the NCERT as proof.** Okay, this is again teaching how to use AI. You give them this particular thing; are you able to find all the errors? Use AI to find the errors. And in this, if we do it smartly enough with very subtle errors and possibly non-errors...

**Sridharan**: [02:23] This is just a bell to say we have 25 minutes more to go.

**Anand**: [02:27] Perfect. We are going okay on time.

**Sridharan**: [02:29] You can take as much as you like. Please, it's interesting, going in a very interesting manner. Please go ahead.

**Anand**: [02:35] So, if four or five ideas like this come up, out of these, my success rate is about 20%. That is, it gives me five ideas, and I take one out of those on average, which is 20% more than zero. And the effort that I put in was that initial rambling, but that initial rambling was helpful because I kind of know what I want to do. **Sometimes I just tell it a problem and then I don't even look at the answer. Telling it the problem itself gave me clarity on what I want to do.** Sometimes I look at the answers; sometimes it is helpful. No big deal. Just curious, has anyone tried this? Has anyone found any...? Yes, please.

**Teacher**: [03:13] It suggests AI crime investigation. My question is on evaluation metrics in AI for Class 10. So it suggests some things. **An escape room, an AI news reporter... so it is going to... and design an AI board game for confusion matrix using cards.**

**Anand**: [03:36] Oh, an AI board game using a... okay, creating a confusion matrix using cards, which they will actually play with cards? Okay, yeah, that's nothing, just write it on paper, you can just cut it and ask them to mark it and actually play a game as a group. Yes.

**Teacher**: [03:50] Then, **TED Talk challenge.** Why accuracy can be dangerous, precision saves lives, recall vs. precision—the ultimate battle. Should we trust AI? So these are all the challenges.

**Anand**: [04:02] Very nice. I would ask them to record a video of them delivering a TED Talk, or a five-minute video explaining a point of view. Now, here is my other... I'll come to the other second part of it, which is: everyone sitting and recording videos, I don't have time to sit and watch all the videos.

**Teacher**: [04:22] I've got an assignment for supervised learning model. "Look into your surroundings—healthy versus unhealthy food, label it, let them identify that." It has given me "indoor plants versus outdoor plants," "clean classroom versus untidy classroom."

**Anand**: [04:40] Very nice. I'm very curious, could you please ask: is there an exercise that they can do while sitting in the class? And maybe if you want to describe the class... that could also be done. Because then it would also be interesting to see if different students come up with different labeled datasets in the classroom. So that could be another exercise. **So, what we may want to take away from this is: sometimes the answer to the question "How do we design an AI-proof question or a learning experience?" is to ask AI.** It is...

**Teacher**: [05:14] Mystery object classification. Build your own dataset within the classroom with explaining the features of the object. Then how can you train AI? That was one of the questions that came inside the classroom. **Spot the wrong label.**

**Anand**: [05:32] Which means now that classroom exercises are a whole lot easier. You know, there is no shortage of classroom exercises that we can run. We can just run one every day. **I would even, on my lazy days, give it a prompt like: "Look, I want something that the students will be able to run by themselves in a reasonably organized fashion."** And I will just sit back, think about whatever I want to think about that day, and give me at least a 10-minute break. I suspect it will be able to give me some ideas. Put another way, I have enough work. You have enough work. It's not like we are dying for more things to do and all that. **For how to do it with efficiency, where this is able to help, I think it is definitely worth taking that, especially when one of the side benefits is also that the students are getting to learn the important things.** (Original: *Efficiency-a epadi pannalaam-nguradhukku, where this is able to help, I think it is definitely worth taking that...*)

**Anand**: [06:33] And in this list, one of my major pain points is evaluations. I will not evaluate. (Original: *Ennaala mudiyaadhu.*) It is way too difficult, way too painful. So, one of the things, one exercise that I set up was: "Look, I want you to write an essay. I am going to... you use AI, you use what you want, I don't care. I am going to give it to an AI model—this is the model I'm going to give it to, Gemini—this is the prompt that I'm going to give it to, and I will ask it to give you marks against this rubric. **Find out how to score the highest. That is the exercise.**"

**Anand**: [07:16] And at first when I did this, I felt a little bad that I didn't evaluate. (Original: *Konjam felt a little bad namma evaluate pannalaiye-nu.*) After that, it struck me: when they all write a CV and are submitting for a job, it is an AI that is going to evaluate it. **Before that, we were teaching them how to write so that a human will read. Today they are writing for an agent to read also.** An agent has become a member of our society. So this not only is this not an invalid exercise, this is the exercise that they are not being taught.

**Anand**: [08:00] Art. One of my exercises was: create a piece of art. Use AI to create a piece of art. They are very welcome to submit human-generated images, but it should depict a contradiction. And I will have AI evaluate it. So in this, what are they learning? **Prompt engineering so that they can come up with an image that the AI will recognize as a contradiction.** And he tells them clearly which model, which prompt, and all that. (Original: *Endha model, endha prompt-ellaam avanukku clean-aa solli...*) You can test it out yourself, see if you are able to score higher than your peers. So one class of evaluations, I'm explicitly telling them: AI will evaluate. In the future, AI is what is going to evaluate you anyway, that is what I am training you for, and I'm not going to evaluate it. If we give it like that, then at least some part of the assessment workload becomes less. (Original: *Appadi konjam kuduththom-na...*)

**Anand**: [08:50] Second, sometimes it is possible to deterministically evaluate. That is, can you give me 10 words which, if present, I should give it full marks? If not present, I will not.

**Teacher**: [09:02] Like keywords.

**Anand**: [09:03] Like keywords. Or if it is the sciences or mathematics or whatever, then there is a correct answer, there is a wrong answer. Those are not problems. If it was like that, then I can just pass it straight to AI and say: "You write a program that will check all the answers." **In a program, there is no hallucination or any such thing.** It either gets it wrong, in which case you tell it to run again, or it will get it right and it will always work the same way every time. That is a very powerful approach.

**Anand**: [09:37] In mathematics, I will simply say: "Create a question paper, the students have to answer it in class, objective type, yes or no, whatever. I will just feed you the result, you correct it." Simple. Like that, objective type in-class tests are very easy because you give them a list of options and you just have it validated. But in this... now I'm branching into two areas—three areas. **One: how do I—assessment by an LLM.** Life is very easy. You tell them AI will assess, here is how AI will assess, score against an AI. **Second is subjective evaluation.** AI can help, but we need a little bit of monitoring—and I'll come to how I'm trying that. **Third is objective.** Again, life is reasonably easy; usually, I tell AI to write a program that will solve it. But in all three, there is one layer of difficulty, which is: if the students are putting it on paper, it is a nightmare.

**Teacher**: [10:39] Handwriting.

**Anand**: [10:40] Handwriting matters. No, who will then put it in the computer? (Original: *Yaar adha thirumba computer-la poduva?*) No, how easy is that for you?

**Teacher**: [10:50] No, it's definitely a double job. We did OSM (On-Screen Marking) now, sir, for Board exams. You know, if there are hundreds of papers that are being scanned, and there are possibilities where the bottom part—our children are very effectively using the paper, so they'll write everywhere till the end, below the end, inside the margin and all—so it becomes very difficult to scan it thoroughly well below and capture it.

**Anand**: [11:13] To everyone, I'll just say: "Do this on your phone or laptop." (Original: *Ellaarukkume "Do this on your phone or laptop"-ne sollidalaame.*)

**Teacher**: [11:19] You mean writing the exam itself, sir?

**Anand**: [11:21] Ah, well—we'll have a separate exam for handwriting. That's okay. No, no. During COVID, they learned it. (Original: *Idhu vandhu, these are tests where we are saying, we won't give such opportunities of exploring the internet options. No, no, no.*) **Tell them to use AI and solve it.** (Original: *AI-ai use panni solve pannu-nu sollunga.*) This is... what was it said? Weather, monitor the weather, you have to give photos and all. Take the photo, take a printout, scan it... just upload it on this site. Like, in Google Forms, if we give a form where they say, "Okay, today I recorded so much," we'll get a clean spreadsheet, and if we give that to AI, the problem is solved. You use AI, you have 10 tabs of AI open, do whatever you want.

**Teacher**: [12:22] The problem is we won't know how to analyze it, sir. Children are better off than us. No, we are not analyzing that they are better off than us. No, no.

**Teacher**: [12:31] Absolutely. No, when I'm standing there, it's just a change of attire. We were also there on that side and we've also created all sorts of problems. But when we are on this side, we expect somewhere something to be monitored by us, to be known by us. (Original: *Adhu namma-lukku theriyaadha-poddhu how we...*)

**Anand**: No, no, very correct. That is what we've asked AI: "**I want you to create something where even if they use AI, I don't have a problem.**" Like those exercises, at least one or two will be there where if you put it online, my work is less and there is no problem if you use AI.

**Teacher**: [13:07] So there, computational thinking has to work. How to do the... how to represent it or how to bring the output?
**Anand**: [13:19] Possibly. There are some assignments of that kind also. Correct. Sometimes the assignment is simple, like we saw: "What is the wind direction in your house today?" That they have to type somewhere. This is what I will do. I will tell ChatGPT to tell me how to create a Google Form that can record all of this. Or ChatGPT may say, "No, no, don't put it in a Google Form, put it into a Google Sheet so that they can record it every day." Oh, is that so? But what if they change yesterday's? At which point it may say, "Okay, then have them create a version." Google Sheets versions are possible. Okay.

**Anand**: [13:59] Now your question becomes important: "I don't know how to analyze all of this." Ask ChatGPT. "**I don't know how to analyze all of this. And it will probably say—I don't know what it will say, but my guess is, because this is something I've been doing for some time—it may say, 'Tell them to give you the links to all of their Google Sheets and share it with you, and give it to me.'**" (Original: *...appadi-nu solli-thannaa, I will do the needful-nu sollum.*) Very good. **That is how I'm running a course for 1,500 students at a time. I absolutely have no visibility—not nowhere, I can see it, but I choose not to see any one of their submissions.** I just tell them, "You put it in your Google Sheets. Give me the link." Or if it is code, submit it into GitHub, send me the link. Or if it is a photo, put it into any site, I don't care where you put it, just send me the link. Finally, I have one Google Form. In the Google Form, I have your email ID because you've logged into it, I have a link to Question Number 7's submission, Question Number 8's submission. I download that Google Form and put it into ChatGPT and say, "Give me the scores. Do whatever you want. Just get me the answer."

**Anand**: [15:25] What if it doesn't work? Fine. Try this as a sample assignment. Tell them, "Look, we're trying something, if it works, great." And if it works, continue in that direction. If it doesn't work, wait for two months. We will learn, it will learn. Everything comes with an expiry date, so it is beautiful. You can try out all kinds of things. **And what I'm telling you is, this is how I'm running an entire course at IIT Madras. All submissions are online.**

**Teacher**: [15:54] Sometimes ChatGPT fails in some logic. Yes. Because it fails, "No, no, your coding is wrong, this is a wrong logic," then it will say, "Yes, I'm sorry," and it will give another set of logic. Correct. So at that time, what you were saying will become a problem.

**Anand**: [16:12] Very true. Don't take the risk. (Original: *Risque eduthukkaadhinga.*) Where you are sure it will not make a mistake, go ahead. How can we be sure? For one, supposing we give them an assignment like what was the wind... each student builds one simple instrument and maybe we ask them to explain how it works. Now there are two ways of looking at it. Everyone, we are telling them to submit an essay, they will probably put it into ChatGPT, come up with an essay, they will... okay, no, that is not a good idea. Let me take a different one. Let me take this one itself.

**Teacher**: [16:55] Sir, actually when I prompt the group of audience, sir, I have seen something different when I tried to do my idea. "**This is a group of learned teachers with whom I am going to discuss,**" then the answer would be different, I have seen. For students, it will be a little lethargic. For teachers, it will be very bright. If I tell it, it will start being very obedient. Because she is a computer science teacher, she knows there is a mistake in this part of the program, that part of the program. See, even we have studied computer science long back, but I wouldn't be so efficient as a physics teacher to find out where the error lies and how I'm able to correct it. Absolutely. That group of text, it is able to do, ma'am. **If I actually computer science exams are going in piece of paper only, it is handwritten for us. Actually, I should not be telling this but I'm telling you, once I have to complete correction where there was a simple program—you know that a program can be written with 100 logics—actually I'm running short of time. So what I did, even at that time, you have to appreciate me, I was not even having the licensed this thing, paid ChatGPT. So I installed AI, Deep, Black AI, everything in my thing, I was not ready to pay.** Five-five images I started putting for that particular question. One after the other, it was telling me the mistake. So half a mark, one mark, like that I was able to finish it within an hour, all those questions actually, even without paid ChatGPT. And that was text-written, sir.

**Teacher**: [18:32] Suppose... (Original: *Ippo indha ellaam visayathaiyum exam-nu vekkalaam.*) Now I have a doubt. I upload my answer key. (Original: *Ippo enakku doubt, naan ennodaiya answer key-ai upload pannidalam.*) Sir just mentioned that today. I feed my answer sheet as well as the student's answer sheet. Will it evaluate and give it back? (Original: *Sarr-um inaiku adhai solli kuduththaanga. Ennodaiya answer sheet-um pullaiyoda answer sheet-aiyum naan feed panni, adhu evaluate panni enakku kuduthuduma?*)

**Anand**: [18:48] It will. We also make mistakes, it also makes mistakes. (Original: *Pannum. Nammalum thappu pannuvom, adhuvum thappu pannum.*) So I take it step by step. I'm going to correct it anyway. (Original: *Naan correct panna poren, epdiyum correct panni thaan aaganum.*) You also correct it. (Original: *Neeyum correct pannu.*) Let's compare. (Original: *Rendaiyum compare pannalaam.*) In case you find something that I have missed, okay good, I will take it. In that, some amount of understanding builds up. In this kind of question, okay, I've asked 10 questions; in four or five questions, it doesn't seem to be making mistakes. In four or five questions, I'm not so sure. Fine. Next time I evaluate that sort of paper, those at least the first three or four questions, I'll say, "Okay, you evaluate, you're not making major mistakes. The remaining I will evaluate with your support." I will take less of your support, more of your support, and over time you will improve. Okay, when you improve, I will keep an eye on you. **If we do it like that, we are using it like an assistant.** Like training a maid, slowly it too will understand us, we will also understand it. So there are safe ways of using it. One very safe way is like this: you create an assignment for me. Then I'm looking at the assignment, I'm reviewing it; its contribution is just the idea, saving me some writing time, typing time. Assistant work, it's very safe. **The question is largely when we are delegating our decision-making to it. Do that very judiciously. Do it when you are confident, but build your confidence.** Don't say, "I don't trust it so I will not give it any work." Don't trust it, still give it work. It is doing the work, you are not doing the work.

**Teacher**: [20:34] Sir, in Claude, actually, there are certain... I have seen... you have "God mode," "Ghost mode," "artifacts," this, that and all. See, when you change your prefix or wherever you use it, either in front or behind—I'm not sure where we use it—there are certain... and it beautifully organizes it. The moment you give your prompt, and you make your prompt using correct words, it modifies it very professionally. Yes. **So can ChatGPT do the same thing like how Claude does?**

**Anand**: [21:11] Today, not as well as Claude, but who knows tomorrow? Actually, who knows today? (Original: *Inaiku naan try pannala, theriyaadhu.*) Every day, because a new model is being released, GPT-5.6, people are saying it is better. But in this, there are many things that the companies are trying to improve on a daily basis. Claude generally they have been doing well on design and code. GPT has generally been doing well with data and analysis and is also cheaper. But there have been flips sometimes. So at any point, the answer to the question is: let's try it. **And experimentation now is generally becoming people's second nature. Ignorance is good. We don't know, we can ask it, we can get an opinion which we can validate, and therefore the process of finding out constantly and learning is becoming the skill.** Knowing is not the skill. Being able to learn is the skill. Increasingly... not in all areas, in human relationships it is instinct and practice and all of that, but in technology and being able to use technology, this is becoming a more relevant skill. As we practice, of course it is helping us. Students watching us... see, there is a segment that is very far ahead. There is a segment that is very far behind also. That segment will also learn. (Original: *Oho, miss ippadi panraala, sari namma-lum panni paapom... illa ippadi thaan pannuvaala irukkum-nu that group also gently starts learning.*) Or when we encourage some of the students who are really doing this well, they say, "Oh, okay, let's do it like him, do it like her," and it starts improving. **In short, yes, AI hallucinates. Where the hallucination does not matter... sometimes hallucination is creativity.** It will say something wrong, but "Oh, that's a good idea," and then we get ideas from its ideas. In that kind of situation, or doing some experiment—whether it's right or wrong—just let it go. (Original: *Vittudanum.*) We just let it go loose. Where things matter, we will keep an eye on it, but not hesitate to ask and take an opinion.

**Anand**: [23:29] In short, **any task that you have, give it to AI first.** (Original: *Pannattum.*) Let it fail. Where it succeeds, maybe use it more. That's all. This itself is an exercise. I'm learning every day how to give more and more to AI, and a lot of people tell me, "Anand, you're using AI like crazy." And I find that every day I'm learning so much more that I'm probably using 1% of its potential. (Original: *Avlo irukku.*) Not that I'm saying you should, but there's no harm if you do.

**Teacher**: [24:12] And you are training your own small ChatGPT also for yourself. You are making it ready to adapt to our needs.

**Anand**: [24:23] Correct. The "Chotu" concept. (Original: *Andha "Chotu"-nu oru concept.*) Just having a small boy as an assistant. (Original: *Oru assistant vachundu.*) "Hey, go to the shop and buy this. Don't you know how to buy sugar? This is how you buy it. Go, go and give it." And he gets trained. ChatGPT is like that.

**Teacher**: [24:43] We are all becoming prompt engineers more than our teacher's teaching. We are all becoming prompt engineers to do and modify. If you see our tablets and all, sir said he provided us with laptops and stuff. You can see, sir... there's all the...

**Sridharan**: [25:01] One group of data scientists was telling me earlier that the data scientist mentioned... sir was about to open his mouth but then you accepted it! Like, easy to speak.

**Teacher**: [25:12] How safe is our data getting exposed?

**Anand**: [25:14] A very good point. I mean, my data is already on Gmail or Dropbox or OneDrive. **There are some companies that I, out of lack of choice maybe, am forced to trust: Microsoft, Google, Dropbox. In these three, my bank statements, everything is there. So now I have recently added Anthropic and ChatGPT to that list of trust circles.** So to me, there are three levels of trust, maybe four. One: things that I will not even keep in my computer or let anything go outside of my computer.

**Teacher**: [25:56] Using ChatGPT for that, for that access, the control—are we compromising on that?

**Anand**: [26:03] If we share with that, yes we are. Just like if you share an email—if you send an email attachment, Google has access to it. **So I keep it on par. If I will send it to my Dropbox or put it in my Google Drive or somewhere, the level of what I'm okay to put there is the same that I will trust putting into ChatGPT or Claude.** If I won't put it there, I will not put it here.

**Teacher**: [26:29] Again, ChatGPT is a learning model. It learns from your data, right? But we can keep it in a Google Drive, it doesn't learn. Actually, our prompt and our data, our security from that.

**Anand**: [26:39] **I don't trust the people in Google. People are learning models also.** They'll say, "I will not learn from this," and they'll say, "I will not look at your data." Models also say—I mean they have clauses saying that "I will not learn." But ultimately, I say, "Look, either this is something where even if you look at it—human or agent—I'm okay. Or I trust you not to look at it." But basically, I'm not differentiating between human and agent at the other end, I am differentiating my level of trust on the company. And it is there to some extent, not beyond that. So I'm managing it.

**Teacher**: [27:15] Sir, we can always tell it, "I don't want you to remember," and other things. That is actually working, I think so. Yes, that option is there. And since you are saying that you're using so much of ChatGPT and AI, I was like... do you expose your...?

**Anand**: [27:32] No, I go to the other end. I say, "Why should I only share it with you? **I will put it publicly on my blog.**" My phone number, email address, Aadhaar number, practically everything about me is online, including details of several of my bank transactions and so on. At one point, I was thinking—some of my friends and I were discussing—**this privacy concept is a very 19th-20th century concept.** (Original: *Idhu enna-na...*) Why? Because in villages, every night what happens is known to everyone in the next house. If a new person comes to town, who, what... they investigate and just completely destroy any shred of privacy you might have. What is privacy? You're right, maybe we are going back to that model. No, I'm not saying there is a difference because a few organizations or governments are getting access to much larger data, so there is a difference—not denying that—but also seeing that some aspects of privacy are relatively new that we have grown up with. Nice to have, but it may vanish.

**Sridharan**: [28:46] Yes, sir. I have two questions. One is, it's not my own, I heard it somewhere. In general, they were saying: **A 9th-standard student of 2010 and a 9th-standard student of 2026 are very completely different animals. Correct?** Whereas a teacher who has been a teacher for the last 25 years, he taught 9th-standard student in 2010, she is also teaching in 2026. So is there a... is there a conflict or is there an adjustment that... you cannot ask the child to adjust because the child is sitting and working in that environment. Okay, its ecosystem is that all other children are... belong to that 2026 9th-standard student. **Whereas the teacher—is there a risk of the teacher getting frozen in the way in which...** [30:00]

---

**Sridharan**: [00:00] ...in the way in which a particular subject was taught 15, 16 years ago is a question. I thought it was an interesting question, but I don't have an answer. Now, what is going to happen? Depending upon the answer, we need to probably... the immediate thing that came to my mind was that **we need to bring our teachers up to be able to teach a 2026 year 9th-standard student. Not ask the student to go back saying, "Oh, you know in 2010 this is how we did."** [00:44] (Original: *Idhu indha 40, 50 vayasu aanaale andha kaalathula-nu aarambichiduvaanga theriyaadhu?*) [Translation: Don't you know once you turn 40 or 50, you start with "back in those days"?] But that should not happen to our teachers, I think that is something we must mind first. Whatever, if you are able to throw some light into that in terms of...

**Anand**: [00:59] Yeah. Actually, if I may, even before you ask the second question, this teacher, you can search and replace it with programmer, researcher. So many professions are facing exactly the same question. If AI can do my job, what is my role? We can look at it from two perspectives: from the teacher's perspective or from the learner's perspective. Just like you can look at it from a programmer's perspective or the customer's perspective; researcher's perspective or the person who is consuming that research.

**Anand**: [01:38] When we look at it from the other perspective, some people will adapt, some people will not adapt, and those who don't adapt should get replaced by those who are more effective. That is, for the system, a good thing. Nothing wrong with that. And therefore, for the programmer, the researcher, the teacher, one takeaway is: **If I adapt, I have higher longevity. If I don't adapt, my longevity reduces.**

**Anand**: [02:14] But what this also needs to factor in is: **Do I want to adapt?** See, there are some people who say, "My objective is to deliver education to the child, make sure that they learn." Some people say, "I believe that this is something that is worth teaching. What I teach is important. What they learn is also important, but my purpose is to teach what I'm teaching." It may be the method of what I teach. It may be the content of what I teach. And if I am outdated, fair enough, but I'm the person preserving the tradition. The thing that needs to be held on—if I don't do it, who will do it? [02:54] (Original: *Thappu illai.*) [Translation: Nothing wrong.] The forces are against us.

**Anand**: [02:58] But there are two stories that we can tell from this. **One is the story of the progressive teacher who is fighting hard on behalf of the student to prepare them for the future. The other is the story of the freedom fighter who is holding the ground against all of the trend, making sure that what is worth preserving for a future generation still stays intact.** Both are beautiful stories. Both are hard stories. [03:29] (Original: *Inga kathukanum. Inga nilachu nikkanum.*) [Translation: You have to learn here. You have to stand firm here.] But the difficulty is guaranteed, unfortunately. If you happen to know your story, good. Sometimes we don't know or we are in the process of discovering that story. That is also okay. Just know that there are two stories, both equally valuable or both worth singing stories or songs about.

**Teacher**: [04:02] Sir, I have a little view on this. **Without knowing the subject, just using ChatGPT is like without knowing math, when I'm trying to use a calculator.** So that is what I'm able to relate with what sir told now. Like the basic fundamental knowledge is with the teacher, even though the IT field is growing and all these ChatGPTs are coming. **How to do mathematics with 2+3, only the teacher will be able to teach with hand and tell the student that "hold the 2 in the mind and 3 in the finger" and all that, which a ChatGPT will not be able to do.** If I use the ChatGPT instead of that, I will not be knowing how that 5 came, but I will be able to get the result.

**Anand**: [04:50] Very true. Now, let me present the same thing in two lines. **One is, therefore, the child that does not understand how that arithmetic is happening will not lose just that skill, but a fundamental concept, that is how it doesn't matter whether it's apples or oranges and so on, there is an abstraction level called numbers.** And that abstraction level can be applied not just to physical objects, to digital objects, but also a certain degree of struggle. This is how we struggle with a problem at an early stage. We fail, we learn. The brain gets wired in a different way by struggling at this. And therefore, **by withholding the calculator or ChatGPT, we are not just training them on the subject, but on a process of learning and on a fundamental concept that can be used in many ways.** Therefore, it is important that we teach it.

**Anand**: [05:52] Even though—now let me present the other perspective—**in the job market, I have absolutely no reason to hire somebody who has spent their time learning how to do mental multiplication as opposed to using Excel to build something three orders of magnitude more sophisticated.** And if in the process they don't know how exactly to do the arithmetic but have learned very different kinds of concepts, like how the cells interact, how a change in one particular cell can influence through a network a very different kind of skill... that is the valuable skill in the market. And therefore, the same teacher may choose a different direction, which is: **"Here is something that is required in the future that the market values. What I'm preparing the child for is a set of different skills they will need, more valuable."** And yes, they're going to lose something in the process, but they're going to learn something in the process. The salary that they will get as a result, the market value that they will get as a result, will be higher. That is what I'm in service of. Both are valid.

**Sridharan**: [07:07] The second question. **The CBSE has moved towards a concept-based teaching and learning. There are no longer the lessons or a syllabus. There are concepts.** This child in this particular class, in this particular subject, has to learn X number of concepts. So, **evaluations based on AI on whether the child has understood the concept or not, there is a theory that it could release a lot of time for my teachers to focus on something better rather than spending time on corrections.**

**Anand**: [07:54] Therefore, how will teachers save time?

**Sridharan**: [07:56] See, when the examinations are held or when the assessments are done, whatever you call it—class test, assessment, evaluation, whatever it is—until you reach the board examination, what is my objective in the school? My objective in the school is that the child has learned every single concept that it is supposed to know by moving from class to class to class. **So today, my average teacher spends a lot of time on correcting answer sheets, notebook, right? Each teacher takes at least about four sections, three or four sections—five sections. Each class, there are each section, there are about some 40 children. So a teacher is supposed to evaluate, assess, correct 160 notebooks, depending upon multiplied by number of tests that she gives, right?** And that, it doesn't happen during the school hours. I hope you understand that too. They don't have... so there's enormous amount of time, private time, being taken away by the school in the form of this correction. **Is there a way I can release that time to my teachers?**

**Anand**: [09:19] Got you. Understood. **Absolutely not.** And no technology... can replace... [laughter] You will not guess what I am saying in all likelihood. **If you had more free time, will you say, "No, I will not do correction"? Is it that 100% of your work you are somehow able to do? Usually it is only 95%, 98%.** [09:52] (Original: *Edhaavadhu vittu pogum, "Aiyo, idhai panna mudiyalaye."*) [Translation: Something will be left out, "Oh, I couldn't do this."] Basically, you have pushed yourself to the limit of what you are willing to do, able to do, willing to do, etc. That is the real constraint. If you had more time, you would have done it. [10:09] (Original: *Vela irukku.*) [Translation: There is work.] There's no shortage of work. [10:11] (Original: *Illai-naalum neengala izhuthu pottupingha.*) [Translation: Even if there isn't, you will drag it upon yourself.] There is just no doubt about it.

**Anand**: [10:16] So if there is some technology that suddenly cuts your work in half, you will find the work and you will spend the same amount of time. There is nothing that is going to solve that part of the problem. **And it is not the work itself, it is the amount of pain that you can tolerate.** If it was fun, you would be doing corrections until 12 at night. It is the part that is painful that is slowing you down. So the amount of pain that you take is entirely driven by your ability to take that pain. No technology will change that. **What it can do is, for that amount of pain, how much good is being done to the children or school or society? That can change.** Your pain will not go away because you have chosen it.

**Sridharan**: [10:57] No, no. My objective, ultimate objective, is not to reduce the pain. **Shift it. Shift it from evaluation, you go into the creative methods of teaching, right?** Today, every teacher's ability to be a good teacher or able to teach a particular concept is restricted to only herself. But if the same teacher goes and asks ChatGPT or goes into Instagram or wherever it is and then says, "I want to teach this... this one," like what you did—we never thought about this home exercise being given for taking temperatures, wind, this one, etc. But my teacher should know that this is the type of experiment if I give... but on her own, having taught social sciences for the last 20 years, would she have thought about it on her own? **So that is why I'm saying the evaluation—it is the evaluation part of it is a physical, manual time being spent by my teacher. If she can upgrade herself with what are all the other ways of teaching, methodology of teaching... Ultimately, objective is concept should be known to a child. Once I teach the concept, it should not forget.** For that, what is it that I can do? If this is like a, you know, manual labor to an automated labor, that is where...

**Anand**: [12:32] Good point. So here is my guess. Evaluation you can give to AI, it will make us more productive. That is, I'll find that, okay, for one test I used to take 4 hours, now I will take only 2 hours. So what will you do? You'll give two tests. Okay, fine. And similarly, you'll take some of that time maybe and move it into creating ideas. What you will find, I suspect, is: **AI is very good at generating ideas. If you ask it, it will give you ideas quickly and well.**

**Anand**: [13:06] When verifying AI, initially we will not have enough trust. So that 4 hours becoming 2 hours? Yes, it'll become 2 hours. Generation for ideas, you would have sat for 8 hours; that 8 hours will become 8 minutes. So what happens finally? **The amount of time that you need to spend being creative shrinks a lot more, the amount of time that you need to spend doing verification is higher.** Net-net, you are two times more productive on verification, ten times more productive on generation. Everything—that is good. But the mix, I think, maybe you will be doing more verification than less. And both the thinking, idea generation and the thinking will become harder because it is doing all the easy grunt work. Now you have to be even more at a higher level on "Is this verification done right? How do I verify this?" and so on. Which takes us to management skills.

**Anand**: [14:02] So, it is going to be even more exhausting. You'll find that earlier you were able to spend 4 hours at home doing all of this correction; now in 3 hours or 2 hours, you'll get tired. But there will be more productivity, not saying that that is a bad thing. So I'm saying: **Please don't feel bad that you are spending less time but still getting just as tired when working with AI.** It is a normal reaction. The more you work with AI, the more productive you are, you will not realize it... [14:34] (Original: *Oru edathulathane ukkaandhutu irukkom, adhu thaan vela pannitu irukku-nu, but it is very taxing on the brain.*) [Translation: We are just sitting in one place and it's doing the work, but it's very taxing on the brain.] It's okay, that is also a form of exercise for us. And it will be higher in verification, less in ideation. And it may shift differently.

**Sridharan**: [14:52] Another question I want to ask the teachers is, what... somebody said that when I'm correcting, for that particular child, I am able to understand where the child is lacking. Where is it going wrong? Okay, that knowledge, spread over a period of one year—and if I'm going to teach the child the same child for two years—I am getting some... but that is, I can understand that is possible when you have 30 children to take care of. Is it possible when you have 160 children?

**Teacher**: [15:26] It automatically comes...

**Teacher**: [15:27] We know this child...

**Sridharan**: [15:29] **So that will be lost if I put it through automation for correction.**

**Anand**: [15:35] Oh, interesting. Because until you said that last sentence, I was going to conclude the exact opposite. 40 people, in a way, I can know who the people are. **160 people, I will not be able to figure out who they are, individualize, customize it. But AI can.** For it, whether it is memory of 50 people, 100 people, 200 people... my course is running for a massive number. If I went to it and said, **"Give me the personality of each student," it will tell you.** It is going to be able to. Is it better than me if I were doing it as a class of 40? No. Better than me if I'm doing it as a class of 160, 1000, 1,00,000? But yeah, definitely, no chance.

**Anand**: [16:34] So which is a very different dynamic. That is, it's like comparing us with machines and saying, "Look, I can lift 40 kg. Can it lift 40 kg? No, it can lift 4000 kg." There's no comparison. But can it fold something as dextrously, as delicately as me? Okay, maybe not. There's no robot invented yet for ironing, to iron and fold. It's not been found yet. So in skills, okay, it is not able to do it. But here is the thing: how many jobs involve that skill, how many jobs involve brute force? Which is growing?

**Anand**: [17:21] **If there is a shortage of good teachers and AI is an average teacher, I will use AI to teach all of the children who don't have access to even an average teacher.** [17:31] (Original: *Nalladhu thaane?*) [Translation: That's good, right?] And then there will be those premium children who are able to get access to good teachers. Then two things will happen: the below-average teachers will have to find roles that are complementary to AI. The above-average teachers will have to find, again, roles complementary to AI but find new teaching models where they teach groups in a way that provides value beyond what AI can provide, that way it bifurcates.

**Anand**: [18:04] But what I'm saying therefore is: **Personalized feedback has two dimensions. One: what is possible when you can give personalized feedback at scale.** That is not great, but not bad. There is such a model and schools will go towards it. You will face this, no question about it. Because economics.

**Anand**: [18:31] We are now trying to do the trend analysis as one of the bottom, lowest methods of assessing. See, we've done something, we've given our... the marks scored by each of the children from 6th standard onwards up to 9th standard. We are feeding it into a system for it to tell me which child is... is it consistent? [18:59] (Original: *Idhu music-la irukkum-nu nenaikaren, who is into music here? Adhula Aarohanam, Avarohanam, innunnu onnu irukku naduvula, Madhyamam-aa adhu? Something like that.*) [Translation: I think this is in music, who's into music here? In that, there's the ascent, descent, and then one more in the middle, is it Madhyamam? Something like that.] So you can only be constant or you can go up or you can go down. I want to know which child is going up where, which child is going down where, [19:18] (Original: *Adhukku trend analysis mark-a vechundu panren.*) [Translation: For that, I'm doing trend analysis with the marks.] But now, unless that mark is based on individual concepts, I wouldn't be able to understand whether the child has understood the concept or not.

**Anand**: [19:32] Very powerful things, sir. **For this, the biggest enabler, if I have one request that is slightly outside of AI, it is: maximum amount of data submissions, whatever, let it be digitized.** Whatever the student writes, whatever... any communication from the student to you, from you to the student, please digitize. **Record your lectures. One thing, just tie the phone around your neck and record it.** [20:05] (Original: *Onnum illai.*) [Translation: It's nothing.] There's a recorder in the iPad. Even better. Or if you record in video, even better. Let everything be there.

**Anand**: [20:11] So much becomes possible because that evening itself you can just tell it: **"ChatGPT, what are all the things that I covered today? What are all the things according to the syllabus I should have covered and I missed? Tomorrow can you just give me those three or four bullet points? Two students were absent, can you send them an email summarizing what I taught them? Can you suggest one or two questions I can ask tomorrow based on what I covered today at the beginning so it will serve as a refresher?"** There is this "spaced repetition" concept where if you keep reminding students of something at a regular basis, they learn better. "Can you just check last week what I taught and put in one or two things into that as well?"

**Anand**: [20:58] **Firstly, whatever you are teaching, if it goes into AI—and it can simply be the recording and you keep this as transcript—it is such a valuable resource because it is like having ChatGPT attending your class.** [21:11] (Original: *Aamam, adhukkum solli kuduppom.*) [Translation: Yes, we'll teach it too.] And if you have all the students' responses coming into it... "I taught this, which of the students have really caught on to some nuance in this?" What they are writing in class... okay, I have uploaded all of it. Writing in class, you have to scan it. Okay, fine. But at least scan it and put it there.

**Anand**: [21:34] Better yet, **tell them just go and type out what you remember from this lesson. What will they ask ChatGPT? Even if they ask, it's okay, take that also as input. Let them put it into one Google Form that you create every day.** Let's keep it. And then you can ask, "Overall, what did students absorb from what I said? What didn't they absorb?" So maybe I should take that portion that majority didn't understand and give it. What are the core concepts that they seem to have understood? What are the concepts they didn't understand? If tests are digitized, you can ask at the question level, "Which of these concepts does this question map to? And therefore you do an analysis and tell me which concepts people have understood, which concepts are not." **In short, whatever comes from you, whatever comes from the student, please digitize. [22:30] (Original: *Idha vida higher value activity kidayadhu.*) [Translation: There's no higher value activity than this.]**

**Sridharan**: [22:34] Very good. Good. [Applause] Okay. I know many of you have commitments back at home and I promised that we will... we will impose ourselves on Anand next time also, whenever he comes here, whenever it's convenient for him to come here. Some of these some of these thought-provoking ideas... Thank you very much, Anand, on behalf of all of us.

**Anand**: [23:09] Pleasure is mine. Thank you. [Applause] **But next time I would love to see a presentation of the various things that you've been doing with AI.** [23:37] (Original: *Naanum kathupen.*) [Translation: I will also learn.]

**Sridharan**: [23:39] Sure. Wonderful. That would be great. Thank you.

**Anand**: [23:41] Thank you.

# Discussion after the session

**Sridharan**: [00:00] Because of the external environment and also, the **school system in India is a compliance-oriented system**. *Everyone has to stand up when told, sit down when told, wear only this dress, do that.* [Tamil: *Ellarum ezhundirukanumna ezhundirukanum, ukkarnumna ukkaranum, dressnu inda dress-dhan podanum, adha pannanum.*]

**Sridharan**: [00:23] So, there is a general feeling that the **extent of creativity in the children is sometimes stunted** because you take all of them... *Yes, mother. I am looking into it, mother. Okay. I'll speak to you.* [Tamil: *Haan ma. Parkren ma. Seri. Naan pesuren.*]

**Sridharan**: [00:52] So, that creativity is stunted because your **focus is more on compliance than on creativity**. If a child in a class of 40 stands up when the teacher is teaching halfway through and says, "Ma'am, I have this doubt," she will say, "**Keep your mouth shut, sit down, and wait until the end of the class.**" That is compliance-oriented, but that stunts creativity. So, there are a lot of things that I... how long are you here for?

**Anand**: [01:25] I'm leaving on Monday.

**Sridharan**: [01:34] Absolutely. Thank you. Next time, maybe in another about three or four months, because we also, as you say, have to now come up with things that I want to show you. Next time when you come, it will be what it is that our teachers have been able to do—and not able to do. **Failures are also important.** It will be about what they have been able to do, what are the things that we failed at, and what are the things that we tried and failed. At that point in time, you can provide your inputs.

**Anand**: [02:11] Definitely. Thank you. Thank you, Anand. Thank you very much. I'll be in touch. Sure.