# Transcript

**Anant**: [00:00] It's kind of a disaster, right? Because the whole curriculum pedagogy for higher education and in post-graduation, **everything has to be re-thought because everyone is submitting a fantastic homework, every test is 100% marks, because none of them is written by the students.** It’s crazy. And so we are living in a very important time and so this is something Anand—he is doing it for a lot of people and he wanted to do it for the schools. So I felt what Harish is doing, Infinite Engineers is doing, is very important. And through that we can reach more. We are doing it out of passion, right? Anand and myself, **we just want to share this knowledge**, what is there with us and what we are trying to work along with some of the universities in the UK like Oxford, Cambridge. Also we are having discussions with teachers there and whatever we are learning there, we want to share with more and more people through the process, yeah.

**Harish**: [01:06] Wonderful. Thanks Anant for sort of setting the context. I think probably we could also have some sort of a conversation from the teachers, whoever wants. Maybe they can share the fears, challenges around in the world of AI, whatever they feel. Maybe after Anand comes they can talk or some of the teachers also can open up. I think for participants who have joined, I think you can feel free to sort of interact right now. We are sort of 44, we are going to like cross 50.

**Teacher A**: [01:54] You [inaudible] by mistake it's been raised up. Okay, no problem.

**Harish**: [02:00] Thank you. Yeah, if other teachers want to share something related to this topic or something that you want to cover—I mean, you want to address that is part of the session, please feel free. Because whatever you are feeling right now in the world of AI is far more important and we want to address that and probably you can ask questions. You can even put it in the chat or you want, you can unmute and talk and say a hi and feel free to do that. Or probably you can—if somebody wants to just do an intro, just put it in chat, the subject that you're teaching and just a little bit intro about you so we also get to know more about you through the chat, yeah.

**Teacher B**: [02:43] Okay, good afternoon everyone. This is Teacher B from Shree Niketan Matric School. Hi to everyone. When we were starting our ATL in 2021, we didn't know what to do. We didn't have any kind of experience over ATL. **Infinite Engineers, they guided us a lot. They have showed a path for us.** Even it was a small team—Anand sir, Harish sir, and Ma’am—I'm so grateful for you all. Now we have been a **top-ranked ATL lab in our school**. This is because of your team. Thank you so much for everything. Thank you.

**Harish**: [03:29] Thank you, ma'am. Thanks a lot for sharing it. Yeah. So we have the star of the day, Anand is there here.

**Anand**: [03:37] Wonderful. Hi everyone. Am I visible, audible?

**Harish**: [03:40] Hello Anand. Yes, you're visible. I'm just making you a co-host and I’ll also bring you the spotlight. Yes, so Anand is here. Thanks a lot Anand for deciding to do this and—

**Anant**: [03:59] Just a little bit of—I mean, whatever Anand shared—I mean, I'm also seeing you for the first time. But with your experience in the world of technology, you have been an alumnus of IIT Madras and you have been in the technology field and now you want to do something for teachers. I think whenever industry people want to do something for teachers, I get more excited because you know the reality and teachers are facing a lot of challenges and they are preparing for the future. I mean, whatever children they are preparing, they are preparing for the world that is sort of unknown. And you are living in the industry which is sort of driving a lot of these changes. So I always feel whenever these two people come together, that creates a lot of synergy and thanks for doing this session. And Anand, you want to share a little bit of—yeah, before we—

**Anant**: [04:44] Yeah, great. Just a quick introduction for all the teachers out here. Anand, who I’ve had the privilege of studying with him—**he is the ideal student all teachers will love in the world.** He, in IIM Bangalore where I studied with him, we have something called the academic topper and the extracurricular topper. In my view, the only person in history who's got both of them as a merit—usually academic and the extracurricular are two different people—his name will be there in the both. And similar such accolades, I think, must be in IIT Madras too. I went to NIT. It's great to have him. Fantastic student and a great friend, mentor. He's been a lot of names that people say in IIM Bangalore, but for me, I look up to him and whenever I get stuck in certain things I reach out to him and he's always—not just to me—I always find little scared because his time is precious because a lot of people reach out to him. So I think through thoroughly before I ask him.

**Anant**: [05:46] And when he said this, that "I want to do this," I was very happy because like I said, me being a parent and teaching my 11-year-old kid, and it's extremely difficult now because the distraction they have with mobile phones and even ChatGPT and other things. **Teacher's life has become extremely difficult, right? Extremely difficult before than what it was because kids are not reading textbooks or reading—reading habit itself is difficult and from there making them sit all together in one class and doing it is very difficult, right?** So I'm a big admirer of teachers. And over to you Anand, I can talk about you all day, but today I want you to share this knowledge. Thank you so much for doing this. And I like the rest of the teachers will sit and listen to you.

**Anand**: [06:31] This makes it very awkward and embarrassing for me to ask the next question because I'm supposed to be an IT expert and all kinds of things. But how do I find the chat in Zoom? Can anyone help me please?

**Harish**: [06:48] Okay. That's what I was also figuring out. Maybe I don't know if the school have disabled chat or maybe I’ll—just give me a moment. I’ll just figure it out in the settings.

**Anand**: [06:58] Okay. Tech is beyond [us?], right? I mean, this is the kind of thing I'm very tempted to just take a picture and share it with Gemini or ChatGPT and ask, "Look, where am I supposed to find the chat button?" Actually, I'm going to do that. Let's ask and see if we get an answer to this because, yeah, I mean the answer's probably exactly what you said, it's been disabled. But let's do that. I'm on Zoom as you can see from the screen, but I can’t find the chat button. Is it disabled or what do I do? Let's see. It will eventually tell me. My ChatGPT is pretty slow. But Harish, do you know if it's been disabled?

**Harish**: [07:46] Yeah, I’ll check that out. And just to—just a reminder for the participants who have joined, it would be nice if you can turn on your video if you are in a comfortable situation. It will be nice if Anand could also see you all. So wherever you are, if you're comfortable, you can switch on the video and I think your smiling face and energy will add a lot of value to this session. So just a small reminder. And Anand, I’ll also figure out—I think you can get started, probably I’ll just figure the chat thing.

**Anand**: [08:16] Not for the entire duration, but at least for some duration just to get a feel, otherwise it will be like me [at my place?], right? Just a moment then after that people can—I think I figured it out. Okay, but it is—see, there's another window apart from this where the actual Zoom workplace is. Home, Chat, Phone, Docs are there. "Sign in to chat with your team and continue your meeting discussion," it says. So I'm on the Team Chat tab and this—

**Harish**: [08:52] Zoom workplace, Home, Chat, Phone number, Docs. Yeah.

**Anand**: [08:54] Yeah, if you click Chat then you need to sign in. Okay, let me share my screen and—because I'm going to be asking everyone to type stuff on chat. So let's see. I'm on this page where I guess this is what you were referring to, Anand?

**Anant**: [09:21] No, not this one. It says sign in to the chat—it's a different thing. It's asking me to sign in again. I don’t know why. The Zoom workplace. I have to do it.

**Harish**: [09:37] I’ll also figure out if any other teachers have experience because we used to do a lot of these meetings on Google Meet these days. So Zoom looks a little bit different to me, so I’m sort of figuring that out.

**Anand**: [09:51] Stay connected. Okay.

**Harish**: [10:30] I'm also calling the school because they have this access control. Just give me few minutes and probably you can get started, Anand, with the initial conversation. Probably I’ll just figure this chat thing.

**Anand**: [11:34] Yeah, the conversation—the first thing I was going to ask was how are people using it and maybe invite responses on chat. But maybe we will do that on voice. Probably some people can unmute and talk for now, I mean if they want to express it, yeah. Not that you are new to AI, certainly we are not at a time when we are new to AI in the first place. So **how have you been using AI for teaching?** I'm just curious to hear some of your more interesting experiences. Please feel free to just unmute and share.

**Harish**: [12:15] So teachers now it's your thing because the more you share about what you're going—how you're using AI currently in your teaching-learning, I think that will set a lot of context also. So please feel free to unmute and you can just share your experiences. So right now your use cases, challenges, whatever. Yeah.

**Teacher C**: [12:41] Hi sir. This is Teacher C here from Shree Niketan Global. So, I mean, the question is maybe a little funny because as teachers we kind of get into the AI mode just to make worksheets or to see if our any of our sentences are wrong, you know, grammatically right or if anything else can be added to it. Some—sometimes in fact I've been working on a lesson plan. I type, I mean whatever I want to write, I write it down and I ask AI to help me change it into a better sentence. But one question that I have is—how do—we keep telling our children, you know, "Don't depend on AI all the time, don't go into ChatGPT, I think you should just use your creativity." **How do you think AI is like going to help teachers as such? Because don't you think we will also miss out on our creative way of teaching students?** Because mostly it's always the traditional method, right? We were like that. Now that technology has taken over, it's mostly AI and now that Claude is—in fact a fourth-standard kid is able to tell me, "Ma'am, don't go into ChatGPT, open Claude." So how much are they actually missing out because of technology? So—

**Anand**: [13:57] I think every generation faces similar question. Half a generation ago I faced a similar question with Google and Wikipedia. **If all of the knowledge is out there, then is it important for students to memorize and learn certain things? I think the answer is yes, it is important for them to learn and memorize certain things, but maybe 90% of the things that we used to learn and memorize became smaller.** For research, they faced a similar challenge. Research mostly was about being able to go to the library to find the relevant book, read it. That skill is no longer required. And the industry had two reactions to that. One: you're losing out on something that is important, which is true. But that important thing ended up being important only in 5% of the cases. 95% it enabled more people to do research. Same with I think Google Maps or the contacts in mobile phones. I don't—beyond maybe two or three phone numbers, I don't remember any phone number. **Have I lost something essential? Definitely. If in an emergency I need to contact somebody I cannot. But that emergency happens maybe only 1% of the time, so we find that loss acceptable because 95% to 99% of the time, the benefit is there.**

**Anand**: [15:29] With AI, I think it is the same and we are in that phase of adjusting to it. So should we force children to not use AI and learn how to deal with it? I think yes, as much as today for instance in Scouts we teach children how to go—wander into a forest without a map, just as we tell people how to solve in an exam without any internet tools. We are forcing them through that. But at the same time, we are also teaching them how to use some of the modern tools because 95% of the time that is what they need when they go to college, when they get into jobs. I think we'll be fighting a losing battle if we say we will never teach these. But at the same time if we identify small pockets where they can survive without those, it will be a good transition maybe.

**Anand**: [16:30] But having said that, **what if we see how we can put in our own creativity as part of some of these?** Like you said, ma'am, when you're creating a lesson plan, if we are using AI does that remove our creativity? Let us try something. I'm going to create a lesson plan, but I’ll do it with a slight change. Let's firstly use voice, I find it easier.

**Anand (to ChatGPT)**: [16:58] I would like you to create a lesson plan—actually no, I don't want you to create a lesson plan. I want to create a lesson plan for let us say introduction to trigonometry. Now, in order to do this, **I want you to interview me. Ask me questions that will help us together come up with the lesson plan.**

**ChatGPT (Voice)**: [17:23] Sounds like a plan. I'll ask you a few questions to get us started. First, who is this for? Like age or grade level? And what do they already know? Just algebra? Any geometry? Knowing that helps us shape the whole thing.

**Anand**: [17:35] Okay, not sure if you heard that—I will try and get the speaker phone sorted better next time—but it was asking me who it was for. So let me answer.

**Anand (to ChatGPT)**: [17:44] Let's assume that this is for grade 7 students and they know a little bit of algebra, nothing more than that.

**ChatGPT (Voice)**: [17:54] Got it. And what's the target? Are we trying to just introduce the concept, like what sine and cosine feel like, or do you want them solving right triangle problems by the end?

**Anand**: [18:06] Was that audible?

**Harish**: [18:09] Yes, it was audible.

**Anand**: [18:11] Great. So then let me continue with the conversation.

**Anand (to ChatGPT)**: [18:13] No, let's keep it simple. This is just for the first session and I would like something that will help them solve—well, just very basic things. Keep it—let's keep it a general introduction.

**ChatGPT (Voice)**: [18:26] Perfect. So more of a conceptual first taste. Timewise, how long is the class? Like 40 minutes, an hour? And is it a one-off or part of a longer unit?

**Anand (to ChatGPT)**: [18:38] Part of a longer unit and let's maybe plan for just half an hour.

**ChatGPT (Voice)**: [18:43] Okay, half an hour. Nice and focused. How comfortable are you with using visuals or physical demos? Even something like drawing a triangle or using a ladder against the wall?

**Anand**: [18:55] And already it's starting to give me an idea. **A ladder against the wall is a very nice way of introducing trigonometry.** I would not have thought of that. Now, at this point I start wondering: who is bringing in the creativity? Is it me because I recognized that a ladder against the—a ladder showing a right-angle triangle is a good idea, or it because it suggested it? Some of this is co-creation. Now, if we say, "Give me a lesson plan," that is one thing. If we say, "**Interview me and we will create a lesson plan,**" that is a slightly different take. I think the whole—in fact, in my mind there are a few prompts that are almost magic words and **"interview me" is one of those magic words.** It can completely transform the way in which we use a tool and this changes the relationship from "I tell you, you do" or "you tell me what to do, I do," which are both asymmetric, to a "let us collaborate and come up with something," making the—our use of the tool something that enhances it to beyond my creativity, beyond its creativity and something combined. So Teacher C ma'am, in short, to your question I would say I don't think we can deny the progress of AI, we'll have to use it, but maybe find spots where it's important to be able to survive without AI. And for us also to not lose out on the creativity by having telling it tap into our creativity.

**Teacher C**: [20:44] Okay sir. Thank you. That answers my question.

**Anand**: [20:48] I would love to hear how else are you using AI and that's an open question to everyone else as well, please.

**Teacher D**: [20:56] Hi sir. Myself, I am Teacher D from Shree Niketan Patasala, Mannivakkam branch. And actually, one interesting way I use AI—it's almost like I used to create a question paper or a worksheet, then **I ask AI to challenge it.** Now, where am I making assumptions? Is it really a higher-order thinking? Is there any other way a student could interpret this? So sometimes AI ends up questioning my questions, which is actually quite useful.

**Anand**: [21:33] That is a fantastic thing when you're validating your question paper, especially from misinterpretations. Makes the questioning much clearer. Let me share a slightly different point of view, not to say that you shouldn't do this, in fact you should do even more of it. But **so what if a question is unclear?** Now, you may say but if a question is unclear, then the student will not understand it. Correct. But if the question is very clear, then AI can solve it. What is the student learning? In fact, today one of the biggest challenges that we have with AI is we say something and it is not able to understand us. That is why we want to give it to a human. Now, if we are training students in solving things that are specified so correctly that they cannot misinterpret it, where is the room for interpretation? Now, obviously this means that it will make our headaches worse. Student will come and say, "Ma'am, but this question actually means this." The other student will say, "No ma'am, it means this." Headache, correct. But that is perhaps the kind of skill that they will need in the AI era. And maybe there is room almost perhaps as a separate section in any class where we teach them to deal with ambiguity.

**Anand**: [22:58] Some of the ways in which I try and teach this is **give the question in a foreign language.** At first, my students complained saying, "Sir, but we don't know French." My response was: what difference does it make when you're going to be using AI to solve the question? Your job is not just to use—I mean, your job is to use available tools. In I had one exam where you can use AI to solve the question. So maybe understanding is coming in their way. We will get into this a little bit because one of the themes I want to explore is: **if AI can solve the intelligence problem, what is left for humans that we should teach them?** It's worth thinking about. But it's fantastic that you're thinking about reviewing the questions from different points of view.

**Anand**: [24:10] Another thing you may want to try is: **what are the different kinds of misconceptions students might have?** And can we frame a question carefully aimed at that misconception? You may know that 80% of the time students make one particular mistake. Can we craft a question that very carefully almost makes them confused? But if they read that question very carefully, they would actually understand it. Let us try and do that. Since we are anyway crafting a trigonometry lesson, let us ask.

**Anand (to ChatGPT)**: [24:47] Actually, that is fine but let me jump on to a different request. **I want you to think about what are the top misconceptions students would have at this introductory level of trigonometry.** List the top three. And then I want you to—no no, listen to me—I want you to craft a question that will help the students tackle that misconception, meaning it should almost lead them along that path of that misconception which they would usually make, but obviously the question is intended to correct them. So can you give me the top three misconceptions—actually forget giving me the top three, just give me the top misconception and one question that will help me address it?

**ChatGPT (Voice)**: [25:35] Okay. So the big one for beginners is thinking sine, cosine, tangent change with the size of the triangle instead of depending only on the angle. So here's a question to surface it. **Imagine two right triangles, both with a 37-degree angle. One is small, opposite side is 3 centimeters, hypotenuse is 5. The other is twice as big, so 6 and 10. Ask them which triangle has the larger sine of 37 degrees.** Then make a prediction before computing. That’ll tease out the misconception.

**Anand**: [26:12] Very fair. Yes, I can see how the size of the triangle makes them get confused about the sine and cosine. Is this something I might have known? Maybe. Now, here I'm applying my creativity only in the direction. I'm saying, "Look, we want to teach them to avoid misconceptions. Specific example you give me." We could flip this around and we could say, "Look, amongst the misconceptions you decide—or actually even in this case we're asking it to decide the misconception also." But we could flip this around and say, "**Misconceptions are one way of teaching. Are there other ways of teaching?** And I will pick examples from those and we can include that." Let's see, let's find out.

**Anand (to ChatGPT)**: [26:59] Okay, that was a good example. Now, see the direction that I went in, you know, that is misconceptions as a way of teaching—I think that's an interesting direction and there may be similar non-traditional ways of teaching students that are still very effective. So I want you to give me maybe—actually I want you to think about three or four ideas, pick the best of those, something that is both novel as well as impactful, and suggest how I might use such—what such that technique is and how I might use it to teach.

**Anand**: [27:38] Here we are saying, "I don't even know what to ask. You tell me."

**ChatGPT (Voice)**: [27:49] Okay. Here's a strong one: **Productive Failure with Contrasting Cases.** You give students a problem they can’t solve yet, say figuring out how to compare steepness of two ramps of different sizes. Let them try, get stuck, propose weird ideas. Then show a carefully chosen set of triangles where their ideas break, except for the ratio that stays consistent. That moment—"Oh, the ratio doesn't change when you scale!"—becomes the hook. Then you name it: that’s Tangent. They’ve basically invented a piece of trigonometry. It’s novel for them and it sticks because they felt the need for the idea before you named it. In 30 minutes, one good puzzle like that can set a deep intuition that makes sine and cosine feel far less arbitrary later.

**Anand**: [28:44] So very good point. Why should we spoon-feed them? In fact, toss them a problem which we know they cannot solve—that is the whole point of what we are going to teach. Let them stumble and then when they learn, they will be far more receptive to it. Yes, good idea, something that I'm not applying enough and could be useful. So all I was doing, Teacher D, was taking your theme of having it clarify questions and saying we can also branch into difficult questions, identifying what kinds of questions to ask, even what pattern of teaching to apply. But thank you very much for that example.

**Teacher D**: [29:26] Thank you sir.

**Anant**: [29:28] I just want to—I just want to add here as part of introduction I missed, which is **Anand is actually teaching Data Science course in IIT Madras.** It is one of the largest courses and I don't know number of students, it's a large number of students. He’s been doing that for the last how many years, Anand?

**Anand**: [29:45] Five years, and "largest" is right as an intake, but large number of students don’t pass.

**Anant**: [29:53] Okay, yeah. True. But totally what is the student strength in your course?

**Anand**: [30:01] Per—every term it’s about a thousand, so 3,000 a year.

**Anant**: [30:05] 3,000 a year. And you teach for all the 3,000 students?

**Anand**: [30:09] I’ve never actually physically met a student because **I’ve designed the course so that AI teaches them**, or rather they use AI to learn, which is another topic in itself. But...

**Anant**: [30:23] Okay, great. I just wanted to say that to teachers that Anand is a teacher—as we speak, his course has been offered in IIT Madras for Data Science course.

**Anand**: [30:34] Would love to hear from more of you. How are you using AI?

**Teacher E**: [30:45] Sir, good afternoon. This is Teacher E from Shree Niketan Matriculation School. I’m handling the KG department. We are very happy to use AI because nowadays **we are creating our own stories, rhymes, even worksheets we are giving.** Previously, we would download from the Google, so now we were using the AI to create our own worksheets, our own manuals, and nowadays we are creating new rhymes, sir. So it is very happy as a kindergarten teacher. So today we were—came here to learn more about it.

**Anand**: [31:31] Which is wonderful. Teacher E ma'am, are you using rhymes like songs or rhymes like words?

**Teacher E**: [31:39] We were getting the lyrics alone and then we create our own rhythm with our music teachers.

**Anand**: [31:46] Very nice, very nice. And I think that is a fantastic thing which you should continue. Let us try something. I'm just opening—okay, this is visible. Let me see if there's a way I can—

**Harish**: [31:58] Is it blurred?

**Anand**: [31:59] Yeah, turn off the blur. Yeah. Now you should probably be able to see Gemini, which I've opened on my mobile. Earlier I was talking to ChatGPT. I find ChatGPT’s voice reasonably good, but maybe the others are just as good these days, I haven't checked recently. Let me ask—so in Gemini, there is a little plus button and that has a bunch of options. One of those options you might be able to see is "Music." So let's turn on that music option and I'm going to ask it.

**Anand (to Gemini)**: [32:32] Can you create a nursery rhyme about numbers? Keep in mind that the lyrics should be the focus, I don't want the music to drown out everything. And keep it short; this is for a kindergarten audience. Make the rhythm nice and memorable.

**Anand**: [32:52] I hope it heard me. Okay, fine. Now, this is a feature that has been in Gemini for at least half a year now, maybe even a year. Lyria is the name of the model that they are using, but basically **Gemini can create songs.** Gemini is free. You can put it on your phone, you can go to gemini.google.com. And I've found that this can be used in so many interesting situations. Once we had a workshop with IAS officers. The vote of thanks was sung by Gemini, effectively thanking every person in the audience for their service. When we have meetings, Zoom meetings like this, before that when we are waiting for people to join, during that time we have Gemini sing a song which is entirely based on the theme of that particular session. It's just fun stuff. But let's listen to it now.

**Gemini (Music)**: [34:00] [Singing: Numbers to find, some things in the garden greens. One snail crawls right here, two big birds appear, three green frogs they jump, four worms on a bump...]

**Anand**: [34:31] I'm not sure how good the audio is, I'll just talk through the lyrics: "one snail crawls right here, two big birds appear, three green frogs they jump, four worms..." something, I've got to check the lyrics out. But it starts with the lyrics in its mind and then creates the song. Now, from that point on, we obviously can perform along with it—the lyrics and students performing obviously has a lot of value in the way in which it brings them together. But **having the model create a song means now the students are free to perhaps perform to it.** Can one student be the snail? Can one be the frog? Can they make sounds along with it? It just takes imagination a slightly different direction.

**Anand**: [35:19] Now, there's nothing stopping us from creating a video based on this as well. Personally, I think the lyrics option is even better because it has more involvement, but there are times when you just want a video of this scene so that children can visualize what is happening. It's entirely possible that they've never seen a frog and a comic picture of a frog is not quite the same thing. You could show a YouTube video of a frog, fine. But **when you get to create a video of a song where the snail comes in just when it says "snail comes in" or a frog comes in just when we have the frog coming in, the receptiveness becomes so much higher,** I think.

**Anand**: [36:11] The other thing that I personally love to do for younger children—actually for anyone, to be honest—is **creating comics**. And once again, both ChatGPT and Gemini are able to do an excellent job of converting even complex concepts into comics. Simple one-panel comic explaining a concept—it is beautiful. So Teacher E ma'am, I love the way you're taking that session for kindergarten. I'm very curious: what else are you doing with kindergarten children?

**Teacher E**: [36:49] Sir, for myself, we are taking ELA [English Language Arts] concept. It's totally based on activity, hands-on do activities. So every day we will receive the manual based on this AI only. They will send a story, rhymes, and activity—everything they will create and forward to the teachers. We will display in our Educomp smart board and we will teach the children during the classroom, which was very engaging as well as they were happy nowadays.

**Anand**: [37:28] That is wonderful. Now, the whole notion of someone in kindergarten using AI for creation is fabulous. I think from there, it is a small step to—and you're probably already doing this, but here's something that applies I think to everyone, which is: **AI is actually a very powerful learning tool.** We know that. When the internet came, obviously the potential for the use of the internet to learn was enormous. **How can we teach children to learn from AI?** And what you said, Teacher E ma'am, is probably a good stepping stone—at least it struck me.

**Anand**: [38:08] One of the things that I would do is **ask students to teach AI something.** Give it a prompt saying, "I want you to act as a kindergarten child whom another kindergarten child is teaching, and respond like a kindergarten child." Now, can we ask the child to explain what a plant is? Can we ask—let's take a higher grade child, let's say a grade 6 or a grade 8 child—to **explain trigonometry to an agent and share the chat with you?** Because once you get all of those chats, you get a sense of what they understand, what they are not able to understand. The agent of course will share feedback saying, "Oh, I understood this. What do you think about this?" etc. It'll try and solve half of the problem. What it says and doesn't say is less important; **what the child communicates to the agent is what we are learning from** and what we are teaching them to do better. Ultimately, communicating with AI is an increasingly big part of this.

**Anand**: [39:18] Now, that is one mode, where taking your concept of "let children use AI" to "**let children interact with AI and let us observe and help them learn from that interaction.**" One is you teach AI something. Another could be: **you argue with AI.** Let us take a concept in history—something that you could debate either way: which was the greatest war ever fought? Pick some two wars, let us say Alexander versus Purushottam [Porus] or versus the Greeks in Sparta, the battle of—whatever it was in the movie 300, whichever it was. Which of these is the greater war and why? Or even simple things like "who's the stronger superhero: Hulk or Superman?" You argue with the AI and you tell the AI to take the other point of view. Now, **that teaches them not just debating but logic, researching,** go find the facts about Superman versus Hulk, find out who is the stronger superhero, and ask them to share the chats with you. From that chat, we learn how they're thinking, what they can do differently, how they can do better, etc. That could be a powerful exercise.

**Anand**: [40:44] Increasingly, I think a lot of homework is going to be along the lines of what you just mentioned, Teacher E ma'am, which is students creating something from AI and sharing. But it doesn't have to be only the output. In fact, **them sharing their inputs will probably be an even more important part because then we can learn from and correct their process.** Any other examples of how you're teaching the kindergarten children?

**Teacher E**: [41:13] Sir, really thank you for your valuable suggestion. That **communicating with the AI with the child is seriously nice.** In future, we will implement this one. And other thing in a kindergarten, nowadays the AI make our work very less. That is the true. Because every day we will spend more time to create any worksheet or any manual, because last year and all we will write on the note manually to create the every daily plan. But now, we just giving our own concept, our own idea to the AI, so it's automatically generating. So our maximum of work started to reduce nowadays. That is the only great thing for us as a kindergarten teacher. And even LSNS also AI was introduced nowadays. So we are getting the help from AI how to teach them in an easy method. How to do an activity based on number counting, sorting the things and all, we are asking the AI so that we are getting a new ideas—not a old one, a creative new ideas based on the resources we have. That is the most important. Because the resource we have, that is only we can use, right? So we will provide—we will send the details of resource what we have. **The AI will create the thing based on our resources.**

**Anand**: [42:47] You know, which is a very powerful theme. Let me share my screen and let's try out something based on what you just said. I'm going to share my desktop. Yeah, I guess it should be visible. So let me—one of the things my colleague did was he took the NCERT class—which class is this? Class 12 textbook—and he asked the models—in this case I think he was using Claude—to **check if there were any errors, effectively doing a fact check of the textbook itself.** Of course, this may not be relevant to you, but the interesting thing is it went through page by page across each of these and said, "Here I found something that might be an issue, might be an issue," etc. It actually found one where there was a real incorrect statement—very minor, but still definitively incorrect.

**Anand**: [43:51] The textbook said only broken or useless objects would have been thrown away. But in reality, it turns out—and Cambridge has published a whole series of studies—saying people often engaged in **ritual destruction of intact objects.** That is, they would place offerings before the God, and today in archaeological excavations we find a lot of intact objects. Sometimes they would also travel and when they travel, they would leave behind the items. So again, archaeological excavations often revealed intact, fully used objects. So "only broken or useless objects would have been thrown away" was absolutely incorrect. Fully functioning, intact artifacts were left behind as offerings or left behind when they had to move to other places, etc. Very small thing, but it was able to do a fact check of something this nuanced. And of course, then there were a whole bunch of other things like, "according to Greek sources, the Mauryan ruler had a standing army of 600,000 foot soldiers, 30,000 cavalry and 9,000 elephants," and it’s saying, "Look, I can't find a single source that validates this; might have been true, but if so we don't know where it came from," things like that.

**Anand**: [45:09] But this, and based on what you just said, Teacher E ma'am, gave me the idea that what I could do is take the NCERT class—let's take a junior class, grade 8 history textbook PDF—and yes, looks like we have some. Okay, this looks like the NCERT website. I will take this. And for engaging lesson—having AI create lesson plans obviously is a pretty useful and nifty trick. Let me just randomly pick chapter 3 from this. Can I download it? Yeah, I can save it as well. Let me save this. And so, which model do you normally use: ChatGPT, Claude, Gemini?

**Teacher E**: [46:03] ChatGPT.

**Anand**: [46:04] Sorry, was it ChatGPT, is it?

**Teacher E**: [46:07] Both, sir. ChatGPT commonly we use ChatGPT.

**Anand**: [46:11] ChatGPT, okay. So let's then go to ChatGPT and give it an instruction which is: **create an engaging game for the students of class 8 that will be based on the lesson in this particular chapter that I have uploaded.** It can be a cursor-based or—actually no, I'll leave it to you, you decide. Think of a nice game that will teach the concept in this chapter. And let us see what it does. I'm going to shift to work and let's just remove this sentence and I think I'm using all of the right things. Upload this chapter and okay, and then I'll also say, "Then publish it on Python." That's it.

**Anand**: [47:18] Now, this will probably take 15-20 minutes. That's okay, it's working, I'll keep chatting with you. But I wouldn't be surprised if at the end of this they are—it is able to create something that at least one out of three times we find useful to share with students. **"Go home, play this game." It's one thing to revise the chapter; it's a different way or mode of learning for them to interact.** Now, you have the skill to create a game. I don't know how to create a game myself; I'm just typing it saying "create an engaging game." You will realistically have a better idea of what kind of a game to create. You may even say, "Interview me and find out what kind of a game to create," and then create it.

**Anand**: [48:08] But that gives us another mode of interaction. Once this finishes I'll share my screen and show what it looks like. But I believe that now the sheer number of things that we can give to students as learning artifacts has exploded. You've already shared a few, like the lyrics for rhymes as a new kind of learning artifact which earlier would have required a poet; now you are enabled. Similarly, we saw how we could create songs or videos. But games are very much a possibility. These games can be collaborative. You can say, "We'll have groups of students form teams, play against other groups of students, see who gets more points." Your marks can be determined entirely based on who scores more. In fact, my course is literally designed that way. **I have students form teams, play games which are intended to teach something, and their score is literally what they score in these games.** That can be another powerful mechanism. But thank you for sharing that. You're right that AI makes the workload so much easier. When there is a tool, we may as well use it. Where it is harmful we will be careful. Where it is useful let us use it.

**Teacher E**: [49:39] Thank you, sir.

**Anand**: [49:41] Would love to hear from others. How else are you using AI?

**Teacher F**: [49:52] Happy afternoon, sir. I am Teacher F from Shree Niketan Global School. Sir, I believe like only—not only ChatGPT, Claude—it's not a limited AI version. We have so many different things we are exploring. We will be taking AI games for students which are very interactive to them, and some 3D animated videos so which they can understand and which they can explore. We are teaching like XR and VR where they can understand the digital world how it is around. So **how any teacher can decide which tool will be helpful for them? What are the main criteria we have to check?** Because I don't believe Claude will give you good images; only ChatGPT will give us. Claude will give nice PPTs. So there is difference between them, so I just want [to know] how we [can know] more and more.

**Anand**: [50:46] Very good question, Teacher F ma'am. The two ways I use and I decide which model or which AI to use is: number one, **ask AI which AI to use.** Number two, **test and see how well it does.** The first is the easiest. I would just go to maybe ChatGPT or Claude or whatever and say, "Here is my task. I want to create a presentation. I can use either Gemini or ChatGPT or Claude or Perplexity—these are the four things let's say I have subscriptions to." You give it a list of what you can access, or you could ask, "Maybe I should take a subscription to something new, you tell me. What are people saying is better suited for this?" Enough people are doing research. So **rather than you having to separately do research, you can outsource that research to AI.** That is what I do first.

**Anand**: [51:42] Whenever, for instance, recently Fable 5.1 was released as a model by Anthropic. Now, I don't know what is new in this. I'm going somewhere in a train, I don't have time to sit and read about it and all of that. So I just opened ChatGPT and said, "Search online. See what people are saying Fable 5.1 can do that is much better than other models. You also know how I'm using AI. So based on my use of AI, tell me what are specific things that I should use Fable 5.1 for." It went through a series of things and it said, "Look, for you mostly Fable 5.1 is not that helpful. Your use cases are not so advanced; it is very good at some very advanced things, you just stay where you are." Okay, that is useful for me.

**Anand**: [52:31] Or in another example, **I wanted to take my parents' wedding photo, which is in black and white, and convert it to color.** I again asked: which model will do a good job? It said, "Look, Gemini can edit images, ChatGPT can edit images, both are saying it can do a good job." So then next comes testing.

**Anant**: [52:52] Anand, can you show—can you show that thing? Because it’s very nice. Because people can't visualize that, but I've read that post of yours. If you can actually show that—and it’s a fantastic example of how to judge an AI.

**Anand**: [53:08] So here is the photo. The first time I tried it was with—yeah. The first time I tried it was in November last year. So this was the original photo and I gave it to Gemini in November and it created this image. Now, here's the thing: if it were some generic image and I told it "convert from black and white to color," it would have looked really good to me, very impressive. But here's the thing: these are my relatives. **I know their faces in and out.** So **I can tell that no, this is not my father's face.** Anyone else looking at this photo would say, "Huh, it looks close enough to this." But no—and I can't even explain it in words—I know that this has gotten the face wrong. My mother's face, it's kind of done a decent job; some of the other relatives, okay fine.

**Anand**: [54:16] But if there is something you know so intimately, so well, that at one glance you can say, "Huh, this is correct" or "this is wrong," then **you have the best benchmark for you.** Because AI will take maybe a few minutes to get the job done. If you don't know the subject very well, you will take hours to figure it out. But if you are an expert in [it]—like I'm an expert in my father's face—then you can judge very easily.

**Anand**: [54:49] So recently, when GPT Image 2.5 was released, I took the same image and passed it the conversion. Now, here it's done a much better job. My father's face is a lot closer to his; my mother's face is no worse, in fact maybe slightly better. But interesting thing is now everybody looks like they are slightly smiling. Very, very subtle. I had to go back to this and this and find out, is that really right? Because when I saw this picture, I immediately got the sense that they are happier in the color photo than in the original black and white. I—but when I had to look for evidence, I had to look very carefully. But **gut feel, I was able to spot it in a fraction of a second.** And it was correct. GPT Image 2.5, for some reason, makes people smile just a little bit more, and you have to tone it down a little bit.

**Anand**: [55:40] So these are the kinds of things that you can do, which is simply testing out a model, but preferably **testing it out in something that you know, where you can judge.** And then see if it is good, if it is bad, or ask your neighbor, somebody who's anyway an expert. Tell them, "Look, we'll try it out together. I'll send the same input to both models. Can you tell me which one looks better?" and then continue using that model. To summarize: number one, ask AI which model to use or which software to use. Second, test whichever you have access to with something that you are familiar with. Failing which, ask somebody who knows that subject well to do the test and share the results. That's usually what I do.

**Anand**: [56:33] Would love to hear more examples or questions for that matter.

**Harish**: [56:42] Anand, just—just to check, for people who are—because I really love this format where the teacher is asking the question and [you are] responding. Just for the clarity of audience: is this how you want to take the session forward? Because we have one more hour to go. I think this is beautiful, right? I mean, with whatever—go ahead, go ahead.

**Anand**: [57:02] No, what I'm realizing is see, **this format is so much better for me. I don't have to prepare.** I get to learn from the kinds of questions, what is important for the audience. I get to learn new kinds of techniques. I probably will discover a new format of presenting material. So I have a lot of benefits. If you're okay, I'm okay.

**Harish**: [57:25] No, no, I'm also okay. I think teachers also, when one teacher says [something], I think it will be relatable to other teachers as well. So I think that's the good thing. And I think now teachers—your question also was: is Anand, apart from this, had anything else to say? Correct. If you have something to say, that's what—

**Anand**: [57:41] No, no, I have my list. I am weaving my items from the list into this. That I will take care of.

**Harish**: [57:47] Got it, got it. Done. So I think for teachers, I think it's the more the platform for you to express your use cases. I think that will be—I mean, you while asking a question, you're also helping other participants also in the call. So I think more you ask questions in the next one hour, I think more we get out of this—out of this session.

**Anand**: [58:02] And share what you're doing. Please share what you're doing as well. Doesn't matter—and **what is not working for you also.** See, if something is not working for you, it's probably not working for ten other people and they must be assuming, "Oh, but it is not possible." But maybe today we can discover somebody who's been able to do it, or I may have been able to do it, or Anant or Harish has been able to do it. Discover something new. **Failures are more important than successes.**

**Harish**: [58:25] Yeah, true. So teachers, I think you can go ahead. You can unmute, you can ask. Yeah, please share your thoughts.

**Teacher E**: [58:34] Sir, sorry to once again—this is Teacher E from matriculation. Now you ask for the failures, right? I just want to tell like that. In manual, it's creating the—for large group, small group activity, rhyme—like that it's creating for 45 minutes, right? But where while **we implementing the manual, AI-generated manual, we find difficult of time management.** Because the AI couldn't able to recognize how we were connecting with the child. It takes some times, no? But it will mention as "10 minutes you can do this activity." It is not possible. That thing the AI-generating manual sometime it's getting failure in kindergarten because due to the time management. Because AI cannot connect with the physical things with us, right? Right, how we can handling the kids. So how we can create manual based on our thing?

**Anand**: [59:36] Very nice point, ma'am. I'll tell you how I'm doing things. Let me generalize this to: sometimes AI gets things wrong, not because—I mean it may not even know. Like you said, it's not able to judge how we are pacing our lessons. Similarly, I ask it to write emails and it writes emails in a certain way...

**Anand**: [60:00] ...but I send a slightly different email because this is my friend; I talk to him in a different way. So the content you gave me was okay, but the style is somewhat different. Or what sometimes happens is I ask for what I should share in one particular meeting, and it will give me a set of points. But that meeting ends up going in a completely different direction, totally unexpected.

**Anand**: [60:24] Whenever anything that I ask AI does not happen, or even when it happens, I add one more chat at the end saying, "This is what happened." For example, if I ask it for a lesson plan, I go through the lesson plan during the day, but I find that I run out of time. Then that evening I go back and read out one message saying, "**Just for your information, this is what happened: I was planning for a 40-minute session, but I was only able to cover the first 25 minutes of this and we ran out of time, maybe because the students were asking a few questions, maybe because I was taking a little more time to go deeper into this. Just FYI, no response required.**" And I send that chat, and it doesn't respond also.

**Anand**: [61:21] Next time this happens, I share it. Sometimes it gives me a full lesson plan and I'm able to cover it fully. I then share, "Today I was able to cover it fully." Not just for this sort of a thing; for almost anything, all of my chats end with what happened when I actioned it in the real world. Then **every weekend I go back and say, "Go through all of my chats. Look at all the things that worked and didn't work from what you said. Then tell me next time when I prompt you, how should I prompt you differently so that this doesn't happen?"**

**Anand**: [62:03] In this case, it would tell me, "Anand, whenever you're taking theory classes you are able to cover—this is close enough to reality by the way—whenever you're taking theory classes you're able to cover all the concepts; whenever there is no interaction you're able to cover the concepts fully in time. But when there is interaction, then it goes into a completely different direction. So **here is one prompt that you should add when you're asking for the next set of topics to cover in your lesson: add these two lines. Number one, split into a core set of topics that must be covered and an optional set of topics which you can drop if you don't have time. Second, if the conversation goes into a completely different direction, what are a few words or interjections that you can put in to bring that meeting back into the direction or topic that you want it to?**"

**Anand**: [63:03] This is what it told me. In your case, just with maybe three, but ideally about ten sessions, if you share the feedback, I suspect you will get two things. Number one, **good feedback on what you are able to cover well, what you're not able to cover well, and how you might want to pace yourself. Second, what changes you could make to your prompt.**

**Anand**: [63:31] This leads me to another thing that I wanted to cover, which is **I record all of my talks, including this one.** Anyway, I'm being recorded on Zoom, which is great, but I also have a recording copy and I'll request the school administration to please also share the Zoom recording for my reference. **Then I pass it to my models and say, "This is what I'm doing. Use this input, give me feedback. Did I make any factual mistakes?"** The sheer number of factual mistakes that I make is very humbling. I'm prone to exaggeration, and it is telling me very clearly: "Here are the ten things that you have overstated. It actually wasn't that big; it was much smaller than that." So I'm learning to control my exaggeration.

**Anand**: [64:26] It tells me what kinds of facts I tend to usually get wrong, so I'm also learning to be a little more cautious when those kinds of facts come up. For instance, model dates—when the models were released—I usually tend to get wrong. So earlier I would say, "Oh, this model that was released in February 2025." Now I at least say, "I think it was released in February 2025, but it may have been plus or minus, I don't know." So that is one kind of feedback that it tells me. It also tells me that I'm not very clued in to the emotional cues of the person who's asking the question. It says, "**Anand, you're doing a very good job of answering the factual question. But when they were asking the question, they were not looking for an answer; they were looking for comfort.**" Okay, fine, now I have to learn this also. Chalo, one more thing to add to the list.

**Anand**: [65:25] But it is useful. In short, where I'm going with this is **AI is able to use real-world information so effectively that one of the most important things a human can do these days is make sure that we have the information that we can give it.** One information is feedback. Whenever it tells us something, let us log, just document—it could be in your notes also, somewhere—log what it did correctly or wrongly. That way we can help it learn from itself and we will also know where it fails. And that therefore is an important function of failure. Second, record everything. Log successes and failures of AI; record everything, especially your classes.

**Anand**: [66:21] The way I do it is I take my phone, I turn on the recording. I have one of these collar things where your name tag or whatever [is], so I just clip my phone onto that. So the phone is usually just hanging here. I don't care if it looks ugly or whatever; I just record my talks. It makes it very easy. Of course, Zoom or Google Meet, etc., make it so much easier to record the online things. Then I take the transcript, save it. **Later on, we can always refer to it and ask it, "How could I have improved this particular lesson? Are there certain patterns of things that I'm doing especially well? Are there certain things that I'm not doing so well?"** In fact, we have done this exercise for one school, and I can show you, time permitting, what kinds of feedback it can give. Teacher J ma'am?

**Teacher J (Teacher G)**: [67:22] Hi sir, this is Teacher G. Actually, we are using AI as Chat—in ChatGPT, Claude, and KVI app for revision worksheets, for worksheets and activities, and what are the doubts we have, everything we uploaded in the AI and it gives, even though if it is not possible, it will give the answer. Somewhat it managed and gave the answers. And now I'm just working on it, which is we have some PDFs, common notes, and we need to give the revision to the children. Now exams are going for our children. So what I am thinking, my idea is **I have to upload the common notes and practice worksheet in the ChatGPT or Claude, and it has to make an app like a revision app.**

**Teacher J (Teacher G)**: [68:24] And that is my idea, I'm working on it. And half of it I'm doing on that, but I'm stuck on where it is, which means it gives the app—the Claude gives the half of the app which is HTML; HTML document it is giving. And I'm stuck on it. Afterwards it's saying I need to go on, upload it in the GitHub, and then it gives a mobile-friendly app, and then I want to be shared in the Play Store like that. But I'm stuck on that after it gives the HTML document. Is it possible—can you please help me on this?

**Anand**: [69:12] Excellent example of a solvable failure, Teacher G ma'am. So thank you for raising this. I'll just summarize: Teacher G ma'am had the AI tool, in this case Claude, build an interactive HTML revision application. But for students to be able to use it, it's saying you have to create a GitHub account, put it somewhere, etc., and now ma'am is stuck. How can we resolve this? What I usually tell participants in my workshops is: **tell Claude to create the GitHub account. Tell Claude to push it somewhere.** Or you can say, "No, I won't give you a GitHub account, where else can you submit it?" You will find that it will do it.

**Teacher J (Teacher G)**: [70:04] And now I'm using Bolt, which is usually we don't have any subscription account, and so that we are using which app is free and that AI app only we are using. So we are using a Bolt now. Yesterday only I have made it; I had an account on Bolt and I'm using on that and half of the app is ready, but it's stuck. Some bugs are coming here and there and that's why I'm asking. If HTML document it is coming, the app it is in the HTML document and afterwards I don't know where I want to be.

**Anand**: [70:50] No, what you just mentioned second is very similar to I think what maybe Teacher F ma'am had asked, which is basically **how do we know which AI tool to use for what?** And you've effectively said, "Look, I tried with Claude, didn't work. I'm shifting to Bolt.new and it is working." Very good. Tomorrow if something else comes up, we'll switch to it. These things are partly for our learning and partly therefore temporary. The cost of creation is so low. Another thing that you could try is: okay, we tried it with Claude, it didn't work. We try again with Claude, open three chats and on all three give the same prompt. Maybe one of them will click; anyway, they might be in a different style. Or give one to ChatGPT, give one to Claude, give one to Gemini; one of them will click. So that is the experimentation approach.

**Anand**: [71:58] Because the cost is low, you can run any of these in parallel without wasting our time. But the other also is tell Claude to go ahead and do the next step, which is: "Get me an account on GitHub, get me a free account." If a free account is not possible and you cannot do it with a free account, you find some free account where you can set it up. Create an account for me. Ask me any questions if you need, like what email ID or password should I put in. Ask me and it will tell it. Now then it may say, "Oh, but I cannot go control your browser." It won't, but supposing it said something like this. Then you would say, "**Okay, how can you do this? Or is there some other tool that can do it?**" And Claude will then tell you, "Yes, there is a Claude browser extension. If you install that—and here is a link, you click on it, it will install it—then I can control your browser."

**Anand**: [73:00] Then you can start doing anything. We have to fill out attendance in some old LMS thing. Tell Claude: "Open my browser, take the—here is a photo of my class, find out who all is sitting in the class and fill out the attendance." Why not? If it doesn't work, okay, we'll try again next quarter, by that time things would have improved. But the important thing is at any point, write down: "I tried this, I am stuck." Exactly as you said: "I got as far as the HTML page but I'm not able to publish it." Write it down. And then whenever you are in the mood, try pushing it, like you did Bolt.new. That's an excellent way. So now you've found a bypass. Some other time maybe Bolt.new is not your solution and you still have to get Claude to do it. Try again, push it. Works? Great. Ask it: "What else can I do today?"

**Anant**: [73:58] Anand answered it very nicely, ma'am, but I did the same thing. I don't want to give an easy answer because Anand knows the answer, but he's teaching you how to fish instead of giving the answer. But to specific answer to that which I'm tempted to say—which is I went through the same struggle—**Vercel, GitHub pages, and few others, there are a lot of free [options]**. But that is not the idea what Anand is saying. I did the same thing. Now I don't want to—I don't want to pay any money. Please give me a free account. Then it said it had to—

**Teacher J (Teacher G)**: [74:32] For that one also, that account also I'm having, sir. It is I think Vercel?

**Anant**: [74:38] V-E-R-C-E-L. But that's not the answer, I mean that's the thing, but I used GitHub gives free pages. There's another company called Cloudflare which gives free hosting and any—

**Teacher J (Teacher G)**: [74:51] Are you saying Venice? V-E-N-I-C-E?

**Anant**: [74:54] No, no, not—but I don't want to give answers like that. See, for me I'm a big fan of another company called Cloudflare. It gives a lot of free accounts, free apps you can set up. But what Anand is saying is not giving the answer but teaching how to fish. And do exactly like Anand says, you'll come up with many answers. But just to make your life slightly easy, cheating is I'm saying Cloudflare offers cloud and F-L-A-R-E offers free hosting and apps.

**Anand**: [75:24] C-L-O-U-D-F-L-A-R-E. Cloudflare.

**Teacher J (Teacher G)**: [75:28] Okay, sir. Thank you, sir. And actually last week I had seen a video which is they are using Astra six, sir, and Codex also. ChatGPT Codex. In that ChatGPT Codex they are asking a code, a prompt, which is they need to go and get the Mario game and they need to finish it. Finish that fully, finish the game fully. This is the prompt given by the user and it used the—and they didn't—actually they said it is in the file manager, the prompt—the Mario is there and the ChatGPT itself go and get the application and it is used fully and finish the whole Mario game. Even we didn't finish the Mario game whenever we are small, whenever we are young age, but the ChatGPT itself it finished the full game. So it's what we are using, it's AI era, we are in the AI era.

**Anand**: [76:30] Totally. And since you mentioned games, let me maybe take this opportunity to share my screen and see what ChatGPT came up with for the history game lesson that we asked it to create a short while ago. So it said, "I've published the Indigo Ledger" and it's created this. I could give you the link, but I don't think chat is enabled yet. So let's go through it. So this looks like an interesting simulation game. So at any point, okay, the Diwani: oh okay, the Mughal Emperor has appointed East India Company as the Diwani of Bengal, Bihar, and Odisha. What can the company do? Okay, if it's been appointed as the Diwani, then they can collect land revenue, rule only port cities, or stop trading and becoming a farming company. No, they must collect land revenue I think. And yes, evidence fits; the company became the chief financial administrator.

**Anand**: [77:41] And next case. Now I see on the right side there is a village which has grain balance, coin balance, courage balance, and evidence score is now gently building up. So I'm guessing—and it's saying this is a play together kind of game. We can discuss each case before we choose. Yeah, this I wouldn't mind getting together into a study group and seeing. Okay, company wants a regular revenue and hopes landowners will improve agriculture. Which rule? Let me now randomly select something. And oh, as a result, the grain and the coins balance came down. Now I didn't—earlier it was six and six. Now that has therefore made these villages perform worse. Okay, that is bad.

**Anand**: [78:40] Now I can do two things. One, I would obviously want it to explicitly say your grain balance has come down by one, coin balance has come down by one, shows a little more details around it to make it easier for the students to play. In other words, we can improve the game over time. The second that I want is I want to track which student is going in which direction, what are the kinds of mistakes different students are making, so how can I capture all of that? These are two questions that I would ask it. But **as we start these games we find that games are not just a means of informative learning. They are extremely powerful tools for identifying learning paths. Very granularly we can identify what decision a student is making, what thinking they have.**

**Anand**: [79:40] That kind of analytical data if you can get your hands on is really powerful. Better yet, you can even ask it to design a game so that you can say, "Look, ultimately my class has very different types of learners and I want to curate my learning towards specific groups. Now what are different concepts that students are not understanding? **I want you to design a game that will surface the various kinds of misconceptions so that in my next class I can specifically say 'Shyamala, you and Rohit don't seem to understand how tangents are different from cotangents, so let's talk about that. And you Rohit and Bharadwaj, you are not able to understand a different concept, let's teach you.'"**

**Anand**: [80:29] So we know that we can create personalized content. So we can anyway just say, "I've created one explainer for you, you go take a look at that" or we can cover it in class. But more importantly, we can find out what are the misconceptions, how many students have what kinds of misconceptions, and do that by creating questions or assessments or even games that can help surface those differences. It's a useful idea. So therefore I would say, just as Teacher G ma'am was sharing, please create and publish more and more of these HTML things. And while you're doing it also ask it, "Where can I collect the data about individual students? Where can I store it? Find me some place, put it there." And if it fails, it fails, it's okay; we're not worse off. But if it clicks, please share it with everyone. And I'm telling you it's possible. I've done it over the last few weeks, so certainly possible. For free, just within ChatGPT.

**Anand**: [81:34] Yes, would love to hear more successes, failures, questions.

**Teacher H**: [81:40] Good afternoon, sir. This is Teacher H from SNMS Matriculation. I have a one question. Now AI is very much useful for everybody making the works of others very easy also. We want to know anything also we are not asking the persons, we are not searching the persons to get the information. I don't know how to do it. Everybody started using ChatGPT and finding the solution. And since we started to implement with the students also in teaching field also, I want to know one thing: even we are not insisting—we are not insisting, we are taking the lessons using AI. Some out of the box we want to think and the children should think and we are framing the questions and we are taking the questions from AI. AI is providing one step ahead from the child grade. **Child also nowadays very smarter; we planned to give the question they should think and they should answer it. But nowadays some 50% of the children they are focusing in a correct manner; they are finding the answer, they are trying the question how to find the answer. Some children were not thinking also, just we have an AI, we will go to the AI and we will find the answer.** They started to using like that. So how how can I overcome this and I should say the real usability of AI? They should use the AI in a proper way, how I can insist the children to do that?

**Anand**: [83:25] Very good point, ma'am. And a very difficult topic as well for two reasons. Number one, a lot of what we are trying to teach children, AI does very well. Five years down the line, ten years down the line, the curriculum will probably change and people will say, "Look, AI can do this; it is not so important," just like mental multiplication became less important over time in curriculums. But we are not already there. So today we are faced with a tool that is growing more and more powerful and a curriculum that has not caught up to that level. So how do we bridge that gap is one part of the challenge. The other part of the challenge is: what should we teach them maybe instead?

**Anand**: [84:19] Let me take the second part first, which is slightly easier, but only slightly. You mentioned very rightly, ma'am, that now everybody is starting to use ChatGPT and the likes first, not necessarily asking a person. Even with Google Maps, earlier I would have—if I was lost I would end up asking a person. Now I just look at Google Maps. On trains, on buses, we don't need to turn around; we're just looking at our mobile phones. That comes with learning as well, but as a result one of the things that we know people are losing—but is becoming more important—is interpersonal interaction. **In the age of AI, how can we teach interpersonal interaction?** I don't know, so I asked AI, which is of course the ultimate oxymoron, but I did.

**Anand**: [85:07] And it gave me a few interesting [ideas] based on a conversation where I said, "Interview me, let's together discover." It gave me a few interesting ideas. In one of my recent exams, this was the question—it is a one-mark question—I told my students: "**You can donate this mark to any other student. You have to give their correct email ID. You are very welcome to exchange marks. So that way you will get back that one mark. Better yet, you can give it to two students and each student will get 0.6, or three students and each student will get 0.5. That is the cap. So your best strategy is: form a group of four. Make sure that everybody gives each other marks. Then all of you will get 1.5 marks on a one-mark question.**"

**Anand**: [85:55] Are they able to coordinate and do this in the middle of a 45-minute exam? Where I've told them they can—it's a remote exam—you can use WhatsApp, you can sit together in a room, you can do whatever you want; coordinate however you want. Surprisingly, in a class of 600 that took this particular exam, only 40 people were able to form groups of ten—I mean ten groups of four each. Shockingly low number. All the others—now they were also told, "Look, you can cheat. You can tell 100 people 'I will give you my mark, you give me your marks instead' and then don't give it to any of them. But the result will be published at the end." So it is partly teaching them social interaction, partly teaching them trust, partly teaching them negotiation, partly teaching them group building, etc., all through just one mark donation question which I thought was a very interesting suggestion that it had given me.

**Anand**: [86:54] And it worked out remarkably well. At the end of the term one of the students messaged me saying, "Sir, I have taken this course and in the last two years because it is an online course I have never interacted with any one student. This course has forced me to interact. Now we are sitting together, studying together, solving exams together. Thank you for this." I said, "Wow, okay." So **it is actually possible to use AI—well, in this case wasn't that much of a use of AI—but even in the AI era it is possible to teach as part of the curriculum some of the skills that are important even when AI is going to take over the bulk of intelligence.** And like this there are many things. **How to ask good questions**, as somebody mentioned sometime, is a very important skill.

**Anand**: [87:43] So those are examples of things which we can introduce and say: now on the one hand I need to make sure you know how to work without AI, but I also need to make sure that the things that you do are still useful when AI is there. Asking good questions: still useful when they are using AI. Forming relationships: still useful when AI is there. Multiplying two numbers: not useful when calculators are present, when AI is present, etc. But as a—that is the second part of how do we teach when students are using—can use AI. Give them things that will be relevant even when AI is being used.

**Anand**: [88:38] But let us take the other part, which is we need them to learn and understand certain concepts. They will ask AI. So if we ask a question saying "explain this concept," they will paste it into AI, they will get the result, and they will feed it back. So that obviously cannot be the way in which we ask that question. What do we want them to do? Reality is AI is there. They still have to understand that concept. Okay. Maybe we can ask them to do something that will—sorry. Pathshala Thiruvalluvar has their hand raised. I think that's the next question; I think you can [wait]. Okay, I'll finish this, may take maybe three-four minutes, please bear with me.

**Anand**: [89:32] One of the things that we can do is ask them: **can you explain it to another student?** How well are you able to explain it? Now they may put that into AI and it'll give a very good explanation to pass it to another student. Okay, great. So come into class and explain it. That's one way. So in other words, are you able to at least memorize what it has told you, which is not really different from what a textbook has told you. That's one possibility. Second possibility: record it in your own voice and share it. This is something that I did for 600 students. Then it took the transcript. Now they may be reading out from AI, but that much it is possible to tell. And then you say, "Look, I want you to explain it in your own voice."

**Teacher I**: [90:04] Sir, we need a prompt, that is the only thing right now we have understood. What is the right prompt we need to give? And also we have one more question, sir. Actually, when different teachers use...

**Harish**: [90:23] I'm sorry, Teacher I ma'am, sorry. A few people have raised their hands, if you don't mind, we'll just go in order. Sorry about that. So, okay, maybe we can... I'll just lower the hand for Pathshala Thiruvallur. Please, if you are able to... oh sorry, no, please go on. I think you are unmuted. Pathshala Thiruvallur?

**Anant**: [90:52] It's okay, we'll go in order. Yeah, Teacher J ma'am?

**Teacher J**: [91:01] Sir, actually nowadays they are using AI for CBSE answer sheet correction, right sir? How can it help teachers with speed, consistent marking, and spotting common mistakes, while the final decision stays with the teacher? **How reliable is AI at reading handwritten answers, sir?** Whatever the student has written, the teacher sees, she can identify this is the correct answer even though the handwriting is very bad also; the teacher can identify the answers. But AI is a machine, so how is it going to be identified? **Is it safe to upload students' answer scripts to AI tools? Then what about privacy?**

**Anand**: [91:47] Fair point, ma'am. Let me take this in reverse order. For privacy, different countries have different laws, and different institutions have different laws. It would be good to check with both the CBSE and the school to understand what the stance on privacy is. Supposing the school says, "For internal examinations, we are comfortable sharing the answer papers with any model," you can use any model. Or if they say, "No, we have a ChatGPT Enterprise version or a Gemini Enterprise version, and within that we have a contract with Google or OpenAI and therefore we can use it," that is also okay. But ultimately, it is the authorities within the school or within the CBSE or whatever that will have to answer the privacy question. It is good to ask.

**Anand**: [92:43] In terms of handwriting recognition quality, the models are getting better and better. Humans themselves are not perfect. So the way things are going is: let us say humans are at some level, some models are better than... many, of course, there's a range of ability for humans. I am terrible at handwriting recognition, for example. Some people are excellent at it, and over time obviously there's practice. So if we take the human range to be this much, some models are worse than the worst humans; many models today are in the human range; there are a few models that are better than the best humans even now. And the models are constantly improving.

**Anand**: [93:29] So what I can safely say is that **a year down the line you would not likely even need to ask this question**, and even today you may be asking this question because some of the models that you were using are relatively poor or are older models. That is fine. But today, if we did have to solve this problem and we don't have a model that is good enough—whatever "good enough" means—how do we deal with it? There are a few possible solutions.

**Anand**: [93:56] **Number one: ask three models.** Ask each of them to come up with the answer and then feed them to each other and say, "Now look, you decide which one is more likely to be correct." **Usually, when you do a double check or a triple check, the error rate falls down.** Models don't make the same kinds of mistakes, and different models make different mistakes. Even with two models, if you ask one model versus another model and say cross-check each other's results, they will do an excellent job. In my experience, an error rate of about 14% falls down to an error rate of about 4% just with double-checking; triple-checking reduces it to about 2%, and so on.

**Anand**: [94:41] **Second: you could do this risk-calibrated.** If it is an important exam, be careful. If it's not such an important exam, let the student come back. That's also okay. You don't necessarily need to take perfect care, just like in reality, you would not have the patience to decipher a tough handwriting; you would just give them negative marks for handwriting and move on, which is what you may allow a model to do as well.

**Anand**: [95:09] **Third: you could ask the model which words it is not sure about.** For transcribing, I do that. One of the prompts that I've given it is: "When transcribing, if you are unsure about certain words, put a question mark next to it." It does a good job of that. Similarly, while doing optical character recognition, you could ask it to put question marks against words or phrases that it's not sure of, or put in even a confidence level: 80% sure, 50% sure, very unclear, whatever; just annotate it so that when deciphering it, it will give you that level of confidence. Now at the end of it, you could also say, "Give me the overall quality of handwriting recognition on this." And what you can later on do is take all the papers, take maybe the worst 10 papers from a handwriting quality perspective, and give it to a second model. That way you would have double-checked only the ones that you need. **These are some of the strategies that we use to improve quality even with existing models**, but wait for some time—I don't know how long—and this may not be that relevant.

**Harish**: [96:23] Teacher I ma'am?

**Teacher I**: [96:28] Yes sir, good evening sir. I use AI mainly to support academic planning, monitoring, and optimization, and also to analyze the students' academic performance, sir. Actually, what I have realized is that AI gives us a starting point, but we cannot use the output directly because sometimes I could see the mistakes or the content may not match our syllabus. **So how can we prompt AI effectively to avoid these mistakes and to get more reliable results?** I think, and also I have one more question: that when different teachers are using AI for the same topic, the quality of the output is varying. **Is there any standard prompt or the framework you would recommend for the teachers to get the consistency and the results, sir?**

**Anand**: [97:27] Very good point. Let us take it in order. The mistakes are something that will continue forever, even if models become perfect, because the questions that we are asking are also starting to become increasingly subjective. What we found is that even experts are disagreeing on several topics, and even reasonably well-defined topics like, for instance, in banking: "Does this customer complaint fall into department A or department B?" 95% they agree, there is a 5% where even they say, "No, no, no, when somebody is asking for a lost PIN, it should go into the PIN-change category." Another person is saying, "No, no, no, they are not trying to change their PIN, they just lost the PIN, so what you really need to do is put them into the recovery-email queue." And it is not clear what the right answer is. So because of that, I believe that **mistakes will probably be broadened to disagreements**. I'm not saying what it did is necessarily unilaterally wrong, but I disagree with it.

**Anand**: [98:49] When there are disagreements, one very effective thing to do is log the disagreement. As I mentioned earlier, what I do is at the end of the chat, I say: "Here are the three things that I disagree with you"—or mistakes you made, if it actually did make a mistake. Now I have a log. At the end of the week, I have it go through all of my chats and find out all the mistakes that it made, or things that I disagree with. And then I say, "**Give me a prompt that I should include next time that will make sure it doesn't do this mistake.**"

**Anand**: [99:25] I don't stop there. Then I say, "Now create a benchmark out of this." That is, you take three of the questions that I asked, run it with this prompt and without this prompt. Did it really improve? Let me show you an experiment—this is a slightly advanced topic, but what the heck, let's dive in.

**Anand**: [100:21] For example, I read... so one of the problems that I have, disagreements that I have with the model, is it writes too complex. Somebody on Twitter suggested this particular prompt. They said they add this prompt to all of their chats: "**Only report to me in ASD-STE100 Simplified Technical English.**" Now this is a standard of writing in simple language. So I said, "Okay fine, I have this problem, so let me see if this is a good idea." What I did was I asked it: "Look, take a bunch of different questions that I usually ask." For example, I might ask—if models are constantly improving and I use them mostly in agents, so how can I benchmark them in an easy way? Or let's take... supposing tickets are rising 40% after we launch a chatbot but customer volume only rises by 10%, etc. Given these causes, what evidence would tell them apart? Some other prompt.

**Anand**: [101:46] So I took a bunch of prompts that I normally use and then said, "**If I don't use Simplified English, if I do use Simplified English, what is the difference in quality of the answer?**" That is, how correct is it? Does it capture the key drivers? Does it capture the mechanisms, etc.? What I found is that **when I use Simplified English, in almost every case the answer is worse**. So when the model is thinking in a very complicated way, it is able to come up with a better answer, but if I tell it to give me the answer in simple words, it is not coming up with as good an answer. Now, I want a good answer.

**Anand**: [102:30] So what I've decided is I will not add this suffix. **I will let it think by itself and then in the next chat I will say, "Now explain it to me in simple English."** This was a case of benchmarking. So I do this sometimes if it is important enough, go to the second step of benchmarking. First step is: write down, log all the mistakes that it is making, tell it to give you an improved prompt. Second: benchmark the prompt, see if it is really making a difference. Now I did this in a more sophisticated way; you could try it out in a simple way. Same question, ask it with the prompt that you now gave it. Is it giving a better result? If it is, okay, at least you have one data point saying that the prompt has improved. Maybe with three or four other data points, you can subjectively improve these prompts.

**Anand**: [103:23] So that is how you systematically improve. This is almost like writing an instruction manual. **Every chat is like a new employee coming in and saying, "Give me work." Now when giving that person work, it will be good to give them an employee handbook or a teacher manual** saying, "Look, I will tell you only two or three things, but you should read the entire manual and interpret it accordingly." So in some sense, we are building that manual for all the new employees that are coming in for our kinds of tasks.

**Anand**: [103:54] Which begs the question: now how do we share it? Which was your second part. If one teacher discovers some prompt, then how do other teachers share it? Unfortunately, there aren't very well-established mechanisms for sharing prompts consistently that I can straightaway give it to you as an answer. For now, the easiest thing is: number one, getting a group of people together and at least for that group share prompts verbally, share prompts over chat with each other, put it in some Google Drive somewhere, some Google Sheet somewhere, saying "this worked in this case, that worked in some case."

**Anand**: [104:40] But **the biggest uplift will come when you have one place where everybody dumps their learnings from AI in any format**, because then what you can do is tell ChatGPT: "Look, I am trying to create a manual. I have not yet created it, but all the material for the manual is sitting there in that Google Drive or that OneDrive or wherever. You go read everything and then answer my question." It can do a pretty good job of that. Or you could say, "Look, 40 teachers have dumped 40 lessons or things that they have learned from AI, things that have worked, things that have not worked; some are very simple, some are very advanced, some may be used only once in their lifetime, some may be used every day. You take all of that and you create an instruction manual." And we can update that instruction manual every week. How? By telling ChatGPT: "Look for anything that is new or updated, use that to update the instruction manual." And then all the teachers can be told: "There is one particular file, an instruction manual sitting here; ask it to refer to it as required."

**Anand**: [105:47] Now these things have various names: skills, whatever. All that is all advanced. The simplest thing is: **make sure there is some one place where people can dump their learnings. Second, maybe periodically collate it into something that is easier for agents to understand. Definitely tell all the teachers where they can find this information.** Today that seems to be a practical way of sharing our learnings from AI.

**Teacher I**: [106:19] Thank you, sir.

**Anand**: [106:27] We do have 10 more minutes, so I'd love to hear any more failure stories, challenges, successes.

**Anant**: [106:49] I want to... I have a regular question for teachers, Anand, which is: how many of you use paid service or are using your personal account to do all of this? Does the school offer you any account?

**Anand**: [107:04] Maybe we'll do a show of hands. Please raise your hands if you are using a paid service that you are paying for. Raise your hand if you are paying for the service, not the school.

**Anant**: [107:21] Or free, free or paid.

**Anand**: [107:23] No, no, one by one. Okay, two teachers have raised their hand for paying by themselves. Okay, let me lower all the hands and now... teachers who are using a school-paid service, that is, school is paying for the AI service, please raise your hands. That is, school has paid for it and you are using that version.

**Anand**: [107:54] Okay, no one. Now let us do the check that you mentioned, Anant. How many of you are using free versions of AI? Please raise your hands.

**Anant**: [108:08] Wow. Tick-tick-tick... 14, 15, 17... 18... 21... 25 so far. Okay, 24, sure.

**Anand**: [108:23] Fantastic. I will lower all the hands. Thank you very much. That, yeah, I guess is [inaudible].

**Anand**: [108:32] But that is interesting. **Today the best investment, in my opinion, is at least the 400-rupee ChatGPT version**, or in my case I would strongly encourage the 2,000-rupee version of either ChatGPT or Claude. Today I have a slight preference for ChatGPT over Claude, over Gemini. Gemini doesn't seem to be that good right now, but these things keep changing. The reason a paid version is helpful is it shows you how the newer models are not making mistakes. The older models often make a lot of mistakes that the newer models aren't. So if you want to get a feel for "is this just because I'm using an old model or is it because AI itself cannot really answer this question for the next few months," just make best friends with somebody who has a paid account and please use the highest model that they have—it doesn't cost any extra which model you use—and test on that the same question, and you may find that sometimes it works.

**Anand**: [109:49] Teacher K ma'am, your hand is raised. Did you have a question, ma'am? Okay, maybe the hand raised for the poll thing I guess, sure. So, I think teachers, any other thoughts, questions, reflections, anything to ask? We have a few more minutes.

**Teacher K**: [110:23] Hello sir, I'm Teacher K from Vengambakkam branch. Talking about all the uses of AI, the betterment, how can we get it better? **How safe is it, or how safe is my identity with it?** Because once I was able to see that it gave me my name, the place I work, the compartment I work with, my designation, the location of my school as well. **So how safe it is for me as an individual? Because I direct a team to use it. So when I direct a team, I should know how safe it is for my team to use it, right?**

**Anand**: [111:00] Very true, ma'am. Now here I think it is important for us to understand what constitutes safety, and that is a difficult topic I have not really fully managed to grasp. One level of safety is certain things I don't want anyone else to know. Let's say secrets between a brother and a sister that only the two of us should know. However, even that sometimes we discuss over a phone. If it's through WhatsApp, then does Meta have access to it? Turns out that WhatsApp has end-to-end privacy. But if I use a regular telephone, can the government tap into it? Answer is yes. But then you say, "Oh, it's a personal matter, I don't mind if the government is desperate enough to listen to our conversation, there is nothing there." In that context, that privacy leak is okay.

**Anand**: [112:04] Let us take bank password. I would not share my bank password even on email. I would probably only share it maybe WhatsApp or okay, call if the government gets to know my bank password then I can't do anything, but... or if the phone company gets to know my bank password because I'm sending it through my SMS. Okay, yes, that is a risk. But I still do it because it is worth that inconvenience, and there is two-factor authentication, I can approve it. So I've kind of gotten used to sharing bank password if required with the phone company or through the phone company's channel, through a channel controlled by Facebook or WhatsApp, or through a channel controlled by Google if it's on email, etc.

**Anand**: [112:49] Put another way: **privacy is largely about us deciding whether we trust the channels for the information that we are sharing**, and it is a combination of the two. There is no general umbrella of trust. So with AI, what are the channels that are involved? Apart from the Internet Service Provider, which is a reasonably... I mean, not all of the traffic gets captured, but the government can instruct an Internet Service Provider to capture that traffic, so it's important to keep that in mind. The AI companies are one organization that we have to decide how much we trust. If you're using Gemini, then you have to decide how much do you trust Google. If you're using ChatGPT, how much do you trust OpenAI or Anthropic and Claude?

**Anand**: [113:44] They all have their set of agreements and rules. Those rules generally say that their employees are not able to access all of this information, they do not share it with any external parties, etc. But we know that when there is a court order, many of these companies have to comply. So what are we protecting, whom are we protecting it from, becomes the question. And can those people get access to it? With AI, if we are concerned about information—personal information about me—should not reach other general individuals, my colleagues, strangers on the street that I don't know, etc., then the answer is that is safe. That most of these companies have done a very good job of protecting.

**Anand**: [114:42] If we are saying that information should not be in the hands of the companies themselves because they may misuse it... we think they've done a good job, but who knows? If we're saying it should not go to hackers: even big companies have gotten hacked. But we still trust them because of the convenience and the chances of some information like this leaking is reasonably small, but we still play it a little carefully. If we are saying nobody other than me should know my secrets, then you should not be sharing it with them. **For each person the boundary is different.**

**Anand**: [115:19] The shortest answer that I can give is: **if you are willing to send it via email or WhatsApp, you can send it to a paid AI model for sure.** If you don't pay for it, the only clause that they add is they might train based on your data. Even that 99% is not an issue, but if you want to be 100% safe, **if you are paying for AI, then your data is as safe as Dropbox, Google Drive, Gmail, WhatsApp, etc. Almost.**

**Teacher K**: [115:57] Thank you, sir. So there should be a borderline where we [inaudible] whatever information we give.

**Anand**: [116:05] Correct. Over time, and with different people, that boundary will change. It's okay. Maybe time for one last question and then we'll wrap up with just one thought.

**Anant**: [116:25] If not, I'll ask this question, Anand. **The "rich getting richer": if this is my learning, if you're already a good student/good learner, you tend to use these tools better.** Right? And this was probably the finding for the MIT course where when they launched the old-era OpenCourseWare and stuff: people who are already better learners learn. How do we use this for... my interest is for people who are not great learners, right? And who have not learning difficulties, but people who we consider not as smart or intelligent, who do not have the curiosity. **How do we use AI for making them a better learner?**

**Anand**: [117:16] I don't know. No seriously, normally I might follow this with some thoughts on, "Okay, maybe we could try this or maybe we could try that." I don't know at all. So thanks Anant for a question that I'm going to take back, ask AI, ponder over, explore, and I definitely will run some experiments and share what I've learned from that. Just yesterday I was asking myself a very similar... I'm not asking myself this question, one of my colleagues was saying, "Anand, what course in AI should I take?" And I was like, "I am going to shoot you. You don't take a bloody course in AI. Any course in AI is one year out of date." "No, no, no, I still want to take a course. There is one course with so much fees, another course with so much fees." "Have you tried these free things?" "No, no, paid course, and if there is a certification..." and I said, "Look, I'm going to shoot you." Which was exactly the point, I think, that I connected with when you asked: here is a person who is clearly not adept at at least modern learning techniques. **I don't know, and I want to find out. If you have thoughts, please do share. I will certainly find a bunch of things.**

**Anant**: [118:40] For me, having been a teacher eventually and worked in schools, and worked with government schools and such, things for me is anyone who's getting 75, 80, my kids are privileged, so they are smart kids and they can learn. But **I always want to look at kids who are not that well-performing, which is what we call as back-benchers, "dumb kids" and stuff. I don't believe personally anyone is; I think the world should be tuned.** Because I found certain learning methodology is tuned for certain mind and certain rigor and stuff; it doesn't tune for everyone. Today we are in a golden era where you can have personalized learning for everyone. And for me, that's why I'm excited. I'm not excited with AI per se, which I am as part of the company and building something cool, but for education, I'm excited because **even the kids where many people, their parents will give up on, and if there are some motivational teachers who can guide them to be a better learner, after some time these kids will become fantastic.** And I've seen those transformations; people call them "*ivanukku math varaadhu*" [he doesn't understand math], "*makku*" [idiot]. Those kids have changed over a period of time. And I like those stories because India is a huge population; today we need to educate a lot more people and bring them up.

**Anant**: [120:00] ...bring them up, not just make the smart kids get into IITs. See, we all went to good colleges because I believe you are smart, but I believe I had good support from my parents, I had the environment to study. If I am not that privileged, how do we enable these tools? So I’m grappling, I don’t know the answer, but I feel **the golden era for learning is because the cost of customizing a lesson for the dumbest kid is very cheap now**. I want to use that power, and I think schools should freely use it. I’d like to contribute to that. It’s just a thought, and yeah...

**Anand**: [120:42] And I would love to experiment on that. Let me do that. I’ll leave everyone with just one parting thought, which is: **AI is improving rapidly. It helps for you to use it even more than you are. Just push yourself every day.** As long as you are struggling a little bit with AI, you are fine. If AI is able to do everything that you need, or you are not able to find enough things for it to do, your learning path on AI is not as steep as it can be. **It is a good tool; it is also a bad tool. Either way, you will need to know the tool.** And every single student is using it; you will also benefit from it and from knowing the pitfalls, etc. So please use AI more. **As a rough rule of thumb, I say 50 chats with AI every day.** And if you don't know what to chat with it, ask AI! "I am supposed to chat with you 50 times a day, I don’t even know what to ask you. What can I ask you? No, no, no, this is useless, you’re giving me bad ideas. Tell me different ideas." And push it to the boundaries. Use AI more. Let's learn from it and let's learn it. Thank you, everyone. Harish, back to you.

**Harish**: [121:59] Thank you, thank you, Anand, for this wonderful session. I have taken my bit of notes because the way I am going to see or next time when I use AI, I know it’s going to be in a different perspective with whatever things that you shared more. Because here, whatever you told, **it feels more like AI becoming my sort of brainstorming partner**. Or I ask AI even to refine my own prompt. Generally, we think, "Okay, what we should learn more to prompt better?" But now you’re using the person who's right before you—the AI is there, the tool is already there before you—and you’re only going to use that same person and to refine it, to give more input. And you said the perfect analogy, right? **A new employee is coming; what is the direction that you’re going to give that employee? Every new chat in ChatGPT or whatever is a new employee, a new purpose.** And we are also getting to learn along with it. So thanks a lot for giving a different perspective. Now I just... I know it’s already time. Thank you, teachers, for spending a Saturday afternoon on this session. Thanks to Shree Niketan management for understanding this. We had to do this physically, but somehow Anand is in Mylapore, he needs to travel, so we thought let’s do this online. And if next... any chance comes, Anand, I’m assuming, will also be happy to do one of... another follow-up session or any questions you have, we can do a sort of follow-up sessions and that’s to make sure that we are in the groove and whatever. So just last few... because I want a few teachers to share how you felt. That’s what makes it more meaningful for us. Maybe two or three teachers if you can just share how it was for you? Please do.

**Anand**: [123:43] Before that, I’ll also just add one comment. **It would be great for us to even share remotely, offline, if we can create a WhatsApp group, mailing list, Google sheet, or some collaboration mechanism where we can start off by sharing what we’ve learned**, what we’re doing, some prompt practices, etc. Not for long—we’ll keep it to just one month of focused learning and discussion, and we’ll end it there. I think that can be powerful.

**Harish**: [124:14] Sure. I’ll get back on that, Anand, after a discussion with the school thing. So, to teachers: two or three teachers who want to unmute and then share your thoughts, please. Any feedback, learnings, next steps, things you’re planning to try? Yes, Teacher E ma’am.

**Teacher E**: [124:36] Sir, actually, thank you so much, sir. I asked one question regarding how to give the prompt. Now I come to know that how to download all my previous chats in my drive. And you have given me suggestions to... **how to give the prompt means, we need to analyze all the old chats and what are the mistakes we have done, and we make them to go through, and then we need to give a new instruction, right, sir?** I just understood that. That surely we will implement, sir. Thank you so much.

**Harish**: [125:20] Thank you, ma'am. Any other teachers or any thoughts? You can just unmute and then share your thoughts.

**Teacher L**: [125:32] Hello, sir. It’s Teacher L from Shree Niketan, Mannivakkam. **We really didn’t know how come these two hours flew away.** It would have been better, actually, if it is a live performance in school or somewhere else. It’s really awesome, actually. We have noted so many points, and we actually are implementing also. For sure, we will also... it will be good once sir comes to our school or somewhere, so that we can share our points face-to-face. Thank you so much, sir. It was really good. Thank you so very much.

**Harish**: [126:11] Pleasure, ma'am. One more teacher who wants to share what you feel right now after this.

**Teacher M**: [126:18] Okay, this is Teacher M here. Thank you, Harish sir and Anand sir, for this session. Definitely, our teachers have been using AI. It's not like... all the teachers in Shree Niketan, definitely, one way or the other, **even if they are not tech-savvy, we make them tech-savvy now because that is the need of the hour.** So definitely, they had been using it. **If we can have more sessions on prompts—that is where they all actually look out for help because they actually don’t know the exact prompts to be given, how to give it.** So if we can have any future session exclusively for prompts, how to give prompts and how to get it, definitely that will be of a great help for all our teachers. But today's session, definitely, we learned a lot where sir was actually helping us to prompt and get what we want. So in a way, it was definitely more of a practical experience which sir has shown us. We are looking forward to more sessions on this. Thank you, Anand sir and Harish sir. This is Teacher M from Thiruvallur, sir.

**Harish**: [127:47] Thank you, ma'am. Thanks a lot. So I think... thanks a lot. We’ll find some form of way if Anand also wants to share something. I’ll also find a way to bridge the communication thing. I’ll speak to the management. But before closing, in the meantime if Anand or Anant wants to share... I just want you to switch on your video so that I can click a group picture. Sort of... it will be nice if you can just turn on the video for whoever it’s possible. I know you have to leave the school and get back to home on a Saturday, but once again, thanks for spending and probably I think once you start... whatever Anand says, probably like when I next... **even he gave a goal: 50 chats per day as a goal**. So probably if you can start doing it daily, and we’re also learning it. And probably if we can have a collective circle among teachers itself, **you can create a prompt library by yourself as teachers**. I think you can eventually create a good resource on school level, campus-wide as well. But I’ll also get some insights from Anand if he also wants to share some resources; I’ll find a way to bridge a gap. And I’ll take a group picture if people have switched on. I’ll just go to multiple tabs. Just a smiling face—I know it’s hard, you’re going to leave the school—but just share a good expression of yours. I’ll just click a picture and... yes, thank you, teachers. I’ll just go to the next slide as well. Okay. Wonderful. Thank you, thank you, teachers.

**Anand**: [129:22] Thank you, all.

**Harish**: [129:23] Thank you all for coming and hope you get to practice. Yes, thank you. Thanks a lot. Anand, Anant, anything to add? All good?

**Anand**: [129:31] No, all good. Thank you.

**Harish**: [129:33] Thank you. Thank you, teachers. Take care, take care. Yeah, bye.