# Transcript **Courtney Fu**: [00:00] ...Industry Expert Series, and the theme of this week's masterclass is **how to build AI products and also prove that they actually work**, right? So it's our great pleasure to have Mr. Anand—Anand S, right?—from Straive, which is an AI technology company headquartered in Singapore, right? And Mr. Anand is actually the Head of the Innovation Group of Straive, this AI technology company. So for this week, throughout the five days, you'll be working very closely with Mr. Anand, right? [00:43] Now, before I take attendance, just a few housekeeping issues. For the masterclass, it's actually all five days, right, throughout the week, and we do expect full attendance, all right? For this particular masterclass, we have Monday and Friday as our onsite physical sessions, right, which will take place in this room from 10:00 to 12:00, all right? So please be punctual, all right? And if you cannot make it, do let us know in advance, all right, because we will be taking attendance, all right? And if you're sick, if you have MC, do provide us with your official documents, all right? [01:25] Monday and Friday onsite in this classroom. Wednesday is also compulsory, but Wednesday is an online session, right? We've already sent out the Teams invitation link, all right? So Wednesday, we're also going to take attendance as well, right, which means attendance is compulsory. Tuesday and Thursday, these are actually optional online consultation sessions, right? Because after today's session, Mr. Anand will actually give you all work to do, all right? And so you have to basically, by the end of this week, right, come up with some—not physical AI [laughter]—you know, some AI prototypes, right, these AI products, **and prove that they actually work**, right? So in the process, you know, you can actually reach out to Mr. Anand, especially during the online consultation sessions, right? If you have any questions, you can just, you know, ask questions and clarify your doubts during those sessions. [02:24] And another housekeeping matter is that this particular Think Tank, right, we've booked throughout the whole week from Monday to Friday morning, probably from 9:00 all the way to like 5:00 or 6:00, right? Basically during office hours, you can have access to this room if you want to come back to this room and work within your—you know, with your group members, all right? So this particular room is reserved for this purpose, so you can come in anytime during the week, right, during this week to work on your AI products. Okay? All good? All right. [03:06] All right, so we are also very pleased to have, you know, eight students from Tokyo City University joining us this week. We have, you know, their professor, Mr. Takayuki, who's a professor from Urban Studies and Civil Engineering. It's Civil Engineering, and you guys are also from Civil Engineering? Undergraduate students? Second-year undergraduate, right? So a very warm welcome to Singapore, all right? I hope you'll enjoy your time here and also enjoy—you know, learn a great deal during this masterclass. And I will not put you guys on the spot to like introduce yourself, right, unless you really want to, but feel free to, you know, mingle with our own students here, okay? And you guys will be also working in groups, right, to produce this AI product. [03:59] Okay, is there anything that I've missed? We have refreshments, all right? If you haven't had breakfast: tea, coffee, some like cookies, and [inaudible]. **Staff**: [04:09]All right, okay. **Courtney Fu**: [04:10] Also, we have that for Friday as well. Yes. Okay, all right, so sounds good. **Courtney Fu**: [04:16] Okay, we have KK? [04:18] **KK**: Yeah. **Courtney Fu**: [04:19] All right. And Yuhan, right? Okay. Dora? Anya? [04:32] [Courtney Fu continues roll call quietly: Daphne? ...inaudible...] **Courtney Fu**: [04:47] [To an attendee] Oh sorry, you're not... [laughter] But you blend in very well! [laughter] Alexis? Okay, Alexis. All right, okay, cool. **Courtney Fu**: [05:00] Okay. All right. I don't know if anyone has MCs... Okay, we'll see. All right, but I think that's my very brief opening. I'll pass over to Mr. Anand, all right? You guys already have your AI tools subscribed, right? So for local students, we do actually reimburse you on that, right? I think Courtney can actually brief you all on that later on. So I'll pass to Mr. Anand, right, to start the session. Over. **Anand**: [05:24] Thanks, Courtney. **Anand**: [05:28] Actually, we're going to begin with a little form filling. I'd invite you to visit this site, and you should find a form there. Once everyone's confirmed that you've been able to open this page, I'll change the screen and take you through the form. But you can just start filling the form, which will be on how you can build AI products. Please just sign in with any Google account that you have. Go ahead. **Courtney Fu**: [06:06] If you need a mic, we have a mic here. **Anand**: [06:08] Do you think I'll need a mic, by the way? **Courtney Fu**: [06:10] Do you think—is it loud and clear, or do you think...? **Anand**: [06:12] Do I need a mic? **Audience Member**: [06:14] Maybe a bit louder. **Courtney Fu**: [06:18] We have a mic if you need. **Anand**: [06:19] Second half, probably. **Anand**: [06:33] Just log in, and you should be able to fill the form. **Anand**: [07:03] So we'll start with the first question, which is: **What's your GitHub ID?** And you may not have a GitHub ID. How many of you know what GitHub is? Most... okay, fine, almost everyone. Cool, so you're good. If you have a GitHub ID, just put in your GitHub ID, and if you don't, then there's a link here where you can create your GitHub accounts. **Anand**: [07:35] **A big part of what we'll be doing in this course is actually building stuff.** I don't really have much to teach. I'll just say, "Let's do stuff," and the agents are going to be doing the stuff. We are probably just going to be steering them, really, and trying to see if we can build something that's useful, build something that works. **Anand**: [08:01] So I have a couple of responses so far. Does anyone have any problems reaching this page? If so, can you just raise your hands? I'll help you. **Anand**: [08:18] Yeah, the QR code, if that helps, is on the screen. **Anand**: [08:28] The second question was mostly on what you thought you would get out of this course. I mean, why are you here? You may not know; that's fine. Based on what you know right now, what do you think you want to get out of this course is perfectly fine. Feel free to fill it in in any language. AI is going to be reading it, so I'm fine. **Anand**: [09:11] Okay, we have three responses so far for "What's your GitHub ID?" Of the various questions, that's probably the one that we absolutely will need to get to everyone filling that out. **Anand**: [09:33] You can change your answers at any point, so don't worry about it. **Anand**: [13:05] Does anyone need any help with the GitHub account? **Anand**: [17:18] Okay, we have nine responses so far for the GitHub accounts. Hoping that's one that we'll be able to finish before we dive into the class. **Anand**: [17:34] Okay, quick show of hands: who's not been able to put in their GitHub ID yet? **Anand**: [17:40] Okay, let me come over to both the tables. Anything that you need help with? **Student**: [17:46] Oh, no, they're just telling me... **Anand**: [17:47] Oh, okay. Anything that you need help with on the GitHub account? Oh, you've already submitted your GitHub ID. Okay, fine. **Anand**: [17:55] So, you were having trouble with your GitHub account? **Student**: [17:58] アカウント... [Account... inaudible] **Anand**: [18:01] Oh, okay. Could you just click on that? Okay. Sign in. Okay. **Anand**: [18:20] So, are you able to... Okay, yeah. **Anand**: [18:44] Okay, we have 11. That's... I'm going to just proceed with that. Do make sure that you have the form filled out. **Anand**: [18:55] Let's take a look at some of the responses you've shared. I have nine responses to what you're hoping to get out of this class. The short summary seems to be you'd like to use AI to automate the creation of products and upskill yourself in that regard. That is roughly what this class is about. **Anand**: [19:21] What we're going to do, however, is **not learn how to build AI products and then build them. We're going to build AI products, then learn how not to build them. Do it first, fail, learn, and we repeat.** **Anand**: [19:41] Why? Because **I don't know how to build AI products. I don't think anyone really knows how to build AI products. AI is changing so rapidly that anything that I tell you today is useless in a month.** So why bother learning something? **What you probably need is to know how to learn. When something changes, how can you figure out what the way of doing things is?** That's what we're going to be doing today. **Anand**: [20:18] Okay, quick question: **What do you think would be the first step in creating a product?** **Audience Member**: [20:26] Having an idea. **Anand**: [20:28] **Having an idea.** Any other thoughts? **Audience Member**: [20:34] Having a prompt. **Anand**: [20:36] **Having a prompt.** **Audience Member**: [20:42] A big account. **Anand**: [20:44] Sorry? **Audience Member**: [20:45] AI account. **Anand**: [20:46] Oh, **having an AI account!** Yes, having an AI account. Possibly, though that I'll probably strike out, meaning you could always get the account easily enough, not to worry about it. What do you think would be the first step? There's no right answer, I'm just... **Audience Member**: [21:11] Finding a problem. **Anand**: [21:13] **Finding a problem?** That could be an interesting one, mm-hmm. **Audience Member**: [21:18] Having a target audience. **Anand**: [21:19] **Having a target audience?** That's a possibility. Any other thoughts? **Audience Member**: [21:30] Structure. **Anand**: [21:32] Structure meaning? **Audience Member**: [21:33] Having a structure of what the final product might look like. **Anand**: [21:37] Yeah, **having a structure of what the product might look like.** Yeah, which is roughly an idea or the shape of the product. Fair enough. **Anand**: [21:52] Okay, let me toss a few that you haven't covered so far. One is: **Is there some kind of a regulation that protects the product?** **Anand**: [22:04] Now, why am I tossing that into the mix? Because you could have an idea; anyone else could have an idea. You could have an AI account; somebody else could have an AI account. You could have a target market; everybody has the same target market. **Why would your product be differentiated from someone else?** **Anand**: [22:28] Maybe you have something that they don't. And what is something that you could have that others might not have? An easy thought process there is: "I hold a license to build this product; nobody else has the license to build this product, so they have to buy it from me." **Anand**: [22:45] **This is actually one of the few remaining differentiators that kind of sustains.** If the government of Singapore says, "I will authorize this particular company to build a product that will check if—I don't know, let's say the other AI products are working fine," then you have a business. **Anand**: [23:08] And I'm not saying that that's an important or the most important or any such thing, just saying here's something that you may not have thought of, or certainly didn't share. Are there any other things that you might think of as a starting point for a product? **Audience Member**: [23:32] Market cap. **Anand**: [23:37] That's a possibility. You look at the companies, the big companies, and is there a gap there for something? Now, that's a negative space, meaning we are going after something that doesn't exist, but it's still a good possibility. Any other thoughts? **Audience Member**: [24:01] Sensitive information. **Anand**: [24:02] Go on. Sensitive information? **Audience Member**: [24:03] [inaudible segment about white space/compliance/regulation] **Anand**: [24:09] Got you. So is there a white space where there is a need to conform to regulations of some kind in a country? That's a fair point as a starting point. Just mine through existing regulations to see if there is a product that is needed there. And it need not just be regulations of a country; could be regulations of a company as well. Are there policies that we want to align to? That's another way of thinking about it. **Question**: [24:43] What about... for example, payouts—like how does the whole money transfer work? Protection... like bank IDs? **Answer**: [25:05] Bank IDs. Got you. I'm seeing that as a variation on regulation, but in something that you said, there may be a different angle to this. I was working at a consulting company, and one of the partners there said, **"Anand, the industry that you really want to be in is finance. There's just money flowing everywhere. You just dip your finger in, some will stick."** [25:33] And that's so true. **Sometimes you just have to watch where the money is flowing, find some product that fits in somewhere, and something is bound to stick.** Where is large cash flow happening? Mergers and acquisitions, maybe? So you say, "Ah, okay, there's lots of money there. Maybe I'll find a product there." [25:56] The reason I was going through these is to say that **the idea can come from many places.** You could even just ask ChatGPT, "Give me a random idea for a product." Not a problem. [26:09] **One thing to remember, though, is the product idea is not a differentiator.** Any product that you come up with, that you think of, somebody else has already thought of. I assure you, somebody has already thought of that product. Many people may have even implemented it. [26:30] A lot of students think of their idea as unique because they thought of it for the first time. Pitch it to a venture capitalist, and the venture capitalist will say, "I've seen this idea a hundred times." You may say, "But before I share my idea, I want you to write a non-disclosure agreement, sign it with me so that you don't steal my idea." The problem is not that people will steal ideas. The problem is that even when you give them your idea, they won't use it. **You have a marketing problem, not a confidentiality problem.** [27:05] Just keep that in mind as we develop products. **The first stage, which is finding out what product to pick and the idea that you come up with, is worth zero.** Until you start executing and building something around it, anybody can come up with an idea. If it took you three years to come up with the idea, anybody else can come up with that same idea just accidentally in three minutes. Do not place much value on that. [27:34] **What we are going to do therefore is not going to be talking about how to find a product, or how to ideate and come up with a product. Doesn't matter. Come up with it in any way you like.** **Anand**: [27:46] So let's do this: what we're going to do is spend a few minutes for you to think of a few product ideas. Just spend a minute. What I'm going to do is add to this form... [searching on laptop] I will locate somewhere... Documents, data, forms... **Anand**: [28:15] Just going to add one question that you could fill in, which is "Product Idea". And the question is: **"What product ideas would you like to implement in this class?"** **Anand**: [28:38] This will likely appear as the last question, which is Question 9 on the same form. So far, I have 11 responses for many of the questions, and do see if you can fill them all out. We do want you to fill this form before you step out, but also give it a shot. Five minutes—it's 10:35—we'll take five minutes for you to come up with as many product ideas that you would like to implement in this class. Anything. Absolutely anything. And you don't have to stick to it. **Go wild.** [31:15] Think of the question as: if I had to build some software today—and it doesn't have to be software; if I had to build _something_ today—what would I like to build? [33:28] I've hidden it for now. I'll come back to it in a few minutes. [33:37] Okay, we have seven answers so far. Just keep tossing in your answers, but let's see what some of the answers are: - An "AI team calendar planner," which sounds fun. - "Translating legal jargon or academically dense texts into something a layman can understand and flag important lines on the terms and conditions." That's probably a second one. Both are interesting. - "Football analysis," yes, that's a possibility. - "Stock market news compiler, analyzer," and so on. - A "menu app with all the updated, accurate menus from each restaurant." - "World news politics summarizer." - A "chatbot of myself using texts from WhatsApp or Telegram." - "A secretary app to help me manage my life and stuff automatically." Okay, that may be ambitious! - "Schedule maker or optimizer, study guide"—fair enough. - And "to visualize the subsurface conditions across Japan based on sparse site investigation data"—that is very specific. Okay, so there are eight ideas. Feel free to put in your ideas. [34:58] Now, how long do you think implementing one of these will take? I'll give you a few options. How many people think it'll take more than a year? Your idea. Okay. How many people think it'll take more than a month? Okay. How many people think it'll take more than a week? Okay, a few. How many people think it'll take more than an hour? More than an hour. How many of you—the rest of you probably think, but let me check—how many of you think it'll take less than an hour? Okay. And of course, it depends on the idea, right? **But here's the thing: my guess is some of these could be done—more than one could be done—in less than an hour. So that's what we're going to do now.** [36:02] Let's take one of these products and choose one, and see how long it takes. Football analysis seems like a reasonably small one. Who mentioned football analysis? Great. What did you have in mind? **Audience**: [36:27] [inaudible] I have an idea about [inaudible] football analysis. **Anand**: What kind? **Audience**: [inaudible] What kind? [inaudible] analyze [inaudible] details. I collect information for games, and [inaudible]... **Audience (translator)**: Okay, so I can help him because he has kind of a language barrier, so... **Anand**: Sure, sure. No, though, I mean you're welcome for him to speak in Japanese and we'll translate it. **Audience**: [Speaking Japanese] 具体的に... サッカーのデータ分析... 具体的にどういう... [English translation: Concretely... football data analysis... not specifically what kind...] **Audience (translator)**: Yeah, we try to think of some, like, data analysis. He doesn't have any specific idea of what kind of analysis, but data analysis is quite important in football games. **Anand**: That works! [38:42] So let's do one thing, then. Let's build a football analysis application, but we don't really know what we want. Okay? [38:53] So what I'm going to do is just open... Hmm, how should we do this? Okay. Now, here's where we would get into a coding agent and have a coding agent do the job. I'm just going to check what coding agents you have. Okay, most of you have ChatGPT Plus, fewer of you have Claude Pro, and yeah, no one has mentioned that they don't have one of these accounts, which is great. [39:23] So what I'm going to do is show you how you could do this with Codex. And I'm going to do this reasonably easily by just going to chatgpt.com, and I'm sure I'll find Codex somewhere. Okay, maybe not. Should I go to `/codex`? Okay, it says, "Go to Cloud." Fine. So chatgpt.com/codex/cloud lets me create—oh, this is too complicated. Let's keep it simple. [40:14] "Build me a football analysis application. I don't know what I want, but it should be really impressive. Give me lots of stats. Download what you need." [40:33] Yeah. Okay, this sounds like a reasonable prompt. [40:41] I'm going to run this, but I'm going to switch to "Work." I'm not going to explain just now why I'm doing that. And I'm going to... this looks like a lever for how smart the model is. Not too fast; doesn't have to be too smart a model. Just go ahead. And, oh, I also want to say, "and make it interactive." Yeah, maybe... yeah, "Light" is good enough. And run it. [41:23] **My guess is that this will build a working application. It may not be what I want, but I don't even know what I want. It's okay. If it builds something, it's good enough as a start, because then I can say, "No, that's not what I want. Here's something else that I want," and change the specification.** [41:48] **In other words, the starting point of a product need not be very clear. If I don't know what I want and I just have some rough idea, I'll ask it to build a prototype. It doesn't cost much.** In fact, you've already paid for these, and within your subscription, you have enough tokens. [42:10] So, when you're using a partner—and I think of AI more as a person than a machine—and if that's the case, then I'm going to use somebody who will build me a prototype, show me what they think it should be like. And if it works, great. If it doesn't work, big deal. [42:34] In fact, I don't even have to do this with one single agent. I can do this with multiple agents. [42:40] So the next thing that I'm going to do is go to my terminal. And again, I'm not going to right now show you what you need to do, but I'm going to create a temporary directory somewhere: `football-analysis`. And I'm going to run Claude Code. No, why am I doing it this way? I should probably just open the Claude application. [43:21] Sorry, you may find that I'm getting a little confused because I'm used to teaching a slightly more advanced class, which is familiar with the terminal. I'll keep things simple for you. [43:31] So, on Claude, I'm going to go to Code and say that I want you to work in this folder. Not "Downloads"—under "Downloads," I had another folder called `football-analysis`. Okay, so let's select this folder, and paste exactly the same prompt. Sonnet 3.5, high... yeah. So now... Oh, sign in. Okay, I'll sign in. [44:20] That should have sent me an email. I'll eventually get the email. Ah, here it is. [44:51] Okay, that... Now let me press Enter, and it's going to work. [44:58] Both of these are building the application. I have no doubt that they will finish building the application. Many of you have tried this; some of you may not have. **And this is the easy part of building an application, which is: you tell it to build some software, it builds the software.** [45:17] Okay, this used to be the hard part. We needed to know how to program; we needed to know how to program well—both of that, it does. We needed to know how to test; it does that well. And more importantly, it does it fast. **So at the very least, the prototyping stage is done.** [45:35] Now, we still have to wait, and it may take 10 minutes, 15 minutes. Let's see if ChatGPT is done yet. If it is... probably not, but... I lost the window. Yeah, it's still... Oh! It's saying the site is ready. It's publishing it privately. Okay. We'll wait for it to get published, and then see what it's built. [46:06] What it's built is unlikely to be what I want. But since I didn't specify something clearly as to what I want, what difference does it make? [46:19] Okay, sign in, continue with ChatGPT... Continue... [46:36] Okay! Arsenal, and what team is that? Newcastle, Liverpool... Interesting! It's plotting how aggressive the team is versus... okay, "chance creation"—I'm not sure what that is, but it looks interesting! Aston Villa is not doing so well compared to Liverpool on whatever these criteria are. Liverpool and Newcastle are still doing well against these criteria. I'm not a football person; I have no idea about football. And something about... oh, okay, it's even gotten individual games. Okay, then there's a comparison. There's a method. Okay. "This is a curated demonstration dataset." Okay, so it's not real data. Fine. But still, it's told me something about what it can do. Something about what moves the table, defensive resilience. [47:54] Does this make any sense to you? Do you understand the terms here? **Audience**: A little. **Anand**: A little? Fair enough. Yeah, I am at level zero on football. I know nothing about it. **Audience**: Football fan. **Anand**: Okay, football fan? Okay, great! [48:14] So, if we have this as one idea, my thought is you probably are saying, "Okay, wait, I can put in some real data. I could change something. I could change something else." It probably gets you thinking, right, about new ideas. Let's pick someone else. Anyone else who's a football fan? What would you like to do next on top of this application? **Audience**: [48:40] Tournament predictions. **Anand**: **Tournament predictions, okay. Any other ideas?** **Audience**: Safe betting. **Anand**: **Safe betting? Yeah, that's a possibility.** And a few things will start coming up. [48:57] Let's see if Claude has built something. I'm not going to go into the details of each application yet; we'll do that later. But, okay, this one's run six commands. It's still churning. What I'll do is take another application from your list and see if we can come up with... Okay, let's see. "AI team calendar planner." [49:29] Is that something you think we can build in the next 10 minutes? How many people think, yes, we can build in, say, 10-15 minutes? Okay, a couple of hands. How many people think we cannot build it in the next 10-15 minutes? Okay, that's two-zero. So yeah, who knows? [49:54] Let's pick something... Okay. "Train schedule analysis in Singapore." Who suggested train schedule analysis in Singapore? Okay, great. What did you mean? **Audience**: [50:12] Because Japan trains are always on time, so I think Singapore, what time is it... [inaudible] compare Japan train on time and Singapore train. **Anand**: Got you. [50:31] Let me reframe that question as: in Japan, trains are typically on time; that may or may not be the case in Singapore. How do we think about this? In fact, why am I even... let's ask it! [50:48] So, I'm just going to go to ChatGPT, and I will switch back to just... yeah, let's stay on "Work." Can you guess what I'm going to say? [51:02] "Trains are typically on schedule in Japan. That may or may not be the case in Singapore. I want you to download real datasets and analyze and tell me what I need to know as a commuter about train timings in Singapore." [51:23] Now, I'm going to run this in a slightly different way. I'm going to run this on the terminal. See, I'm showing you different ways of running a coding agent. It doesn't matter which one you use. Each has their advantages and disadvantages, but we'll get into that a little later. [51:40] Let's... I'm going to go into Codex... `cd research`... `singapore-train-timings`... So, let me run Codex with the model as a reasonably good model, balanced speed, and right. I'll also add, "Build this as an application. Use your imagination." And let it churn. [52:39] **Again, broadly, vaguely unspecified question, but what we're trying to do is turn that into a prototype.** [52:47] **And the reason is, it's pretty hard for us to think about what our product might look like. But when we ask somebody or something, "Why don't you show me what that might look like?", we'll get some ideas, and then it becomes easy to change.** [53:02] **We ask different people or different agents, we might get very different types of ideas, and that might take us in different directions. So in a way, not having a clear picture is a good thing, because then when you tell different agents and they come up with different ideas, even better, right? You get to learn from their thoughts.** [53:23] Let's see if Claude has managed to finish... Okay, it's still running, but it's creating a dashboard—a very detailed dashboard by the looks of it. I was hoping to show you, and I will soon, how you might have a very different idea from what ChatGPT created. [53:41] So, while this is running, I'm going to tell you a bit more about what we're going to do in today's session and what we're going to do in this entire class, this week. **You will be building and sharing at least one product. Multiple products are fine.** [53:59] In order to do that, obviously, you would need a couple of things: 1. You would need to get comfortable with using AI agents to build products. That's something that we will do ideally in today's session, certainly before tomorrow's session. 2. Second, before tomorrow's session, you would have picked a product to build. Maybe even prototyped it—ideally prototyped it! And then tomorrow we'll be discussing what's working, what's easy, what's difficult, etc. [54:36] **But first, you need to make sure that you are able to both build and deploy—build and publish—your application wherever.** [54:49] So far, you've seen my approach to building products. It is: ask AI. That's it. I don't have an idea? Okay, ask AI. I don't know what it looks like? It's okay, ask AI. [55:03] Actually, I didn't even show you the ideation process that I followed for building products. If you had asked me, "What product do you want to build?", here's what I would do: [55:24] I would go to ChatGPT—and ChatGPT knows a lot about me, I talk to it for several hours a day—and I would ask: "If I had to build a product today, what would I build? Go through everything that you know about me and suggest the top answers prioritized with reason." And I'm also going to add, "use local MCP as required." This is a little tool that I built that connects ChatGPT to my computer. Don't worry about that bit. [56:04] **But from an ideation perspective, it has enough context. What do I mean? It knows what I want. It knows what I've done. It knows what work I do. It knows the experiments that I've run.** [56:19] So it's going ahead, looking... It's writing code to understand what demos I have built. It's looking at my Dropbox to see my notes on the kinds of questions people ask me. It's looking at my notes to see what are the kinds of predictions that I make on a daily basis. It's looking at my notes to see what are the things that I consider important and have noted as, "This is what I did today"—roughly like my diary. It's going through a whole bunch of things, and it's saying... okay, something. I have no idea what it's saying, but even from an ideation perspective, there's absolutely no reason why you wouldn't use AI agents. [57:06] **This is not just the case for an individual. This is the case for organizations as well.** I work at Straive. I have a Google Drive which has all the meeting transcripts, all the presentations, all the client material, all our contacts—lots of stuff. That comes from our CRM system; that comes from our human resources systems—all of it. Now, I don't have access to everything, but I have access to a reasonable amount. So one of the things that I asked it to do this morning was: go through everything and tell me what products would be useful for Straive. And it built this list. [57:58] About 153 products prioritized, saying the vendor onboarding workflow is probably one of the most important products that I can build in Straive. There is a task specification assistant that we might be able to build, a workflow compiler, and so on. This took about an hour. It went through several of my documents, searched on Google Drive, searched on HubSpot, which is our CRM system, and eventually came up with this list with a prioritization. And the prioritization is based on both what is important for the organization as well as what is doable, and whether in the course of this class I will be able to verify it and give you enough information about it. [58:50] So one of the questions that I'm going to add is, in case when you're picking the product you want to pick from one of these, that's something that you can do. And I will add that as a purely optional question. This will appear in the form right at the end as question 10. The link to the list that I just shared of Straive product ideas—this one—is on the form. Take a look at it after the class. There's no need for you to take a look at it now. [59:28] You may not understand these problem statements; I will give you a way by which you can use my agent to answer questions about these. You can talk to it in any language by just sending emails, and it will reply within a few hours with answers to your questions. And that is, in fact, one of the means by which we will be communicating during this class. [59:56] But to step back, what I'm saying is... [60:00] Sometimes I don’t even know what product I want to build. You had some ideas; you’ve entered those in. I don’t even think too hard. If I get an idea, great. If not, I ask AI, "What products should I build and why?" [60:15] Second, I don’t spec the product out. I don’t define the product very clearly. "Football analysis is great" — just two words. Let it interpret, and then I have it give me ideas on what a prototype might look like, and then we shape it from there. [60:34] So what you’re seeing in a way is, **it’s not just AI products that we build that have changed their shape** — meaning it’s not just AI products that we’re building these days; **it’s using AI to build any product.** A train schedule analysis would have been something we would have done five years ago as well, just that we would have done it differently. And today, when we are doing it with AI, we probably start with a prototype saying, "Just build me something and show me, and then we’ll see from there where we go." [61:08] Let’s see what it's built so far. Okay, it’s still building the train analysis. I will wait for it to complete. Okay, it’s still building the football dashboard. It’s working hard. I love it when minions work hard. Keep going. Except, of course, it might end up using too many of my tokens. That’s a worry, but that’s okay. We’ll see what we find. Which brings us to the part that I wanted to make sure we cover. [61:46] Remember I said that by tomorrow I’m hoping you would have built a few — well, ideally picked one or more products you want to build and created prototypes for that. For this, you need to make sure that you are able to run a coding agent. So quick show of hands: how many of you have already run Codex or Claude Code? Yeah, or Antigravity or whatever. That’s three, fine. [62:18] For the rest of you, this is what I would suggest. Since the majority of you are on ChatGPT, let’s start by installing the ChatGPT desktop application. So search for "ChatGPT desktop," download it for whichever operating system you are on, and install it. [62:56] Just give me a second. I just want to make sure that the ChatGPT desktop application lets you use Codex in the UI. Okay, yeah, there is Codex, so no problem. [63:20] Do make sure that you have installed this. And once you’ve installed it, when you run it, there will be a Codex button on the top left. You can select that, put in practically anything that you want here, like "Build me a train schedule application" — again, any language — and you should have an output ready that you’ll be able to take a look at. [63:53] For those of you who are using Claude accounts, just search for "Claude download," and similarly download and install the application. You will have a desktop application; there should be a Code button there. Type in whatever application you want built, and it builds the application. So for the next 5–10 minutes, whatever it takes, my request is please download either the ChatGPT or the Claude desktop applications and make sure that you are able to install it on your systems. **Question**: [64:49] [inaudible — What is the difference between using the desktop app versus the web version?] **Answer**: [64:54] Okay, question is, what’s the difference between using it as a desktop application versus using it on the web? With the desktop application, you can give it access to your local files without having to upload them, automatically. On the web, you would have to upload what you need. Both have their advantages. The desktop version means you can’t shut down your machine while it’s running. On the cloud, it runs by itself. **Question**: [65:29] [inaudible — The desktop version still runs using their server, it's not local AI?] **Answer**: [65:37] That is correct. So the question is, when you run the desktop application, it still uses Claude or ChatGPT from the internet? Yes, it does. It’s not downloading AI into your systems; it's not a local model. If you’re running it on your desktop, you can’t shut down your desktop or laptop, but if you’re running it on the cloud, your laptop doesn’t even matter. [68:16] Okay, I’ve added a question at the end, Question 11. Please just — the link doesn’t seem to be working in Markdown, but okay. Fine, I’ve in any case added links to where you can download them at the end. You may need to copy-paste the links since it’s not appearing automatically, but do give that a shot. [69:40] That’s pretty impressive. We’ve got seven who are already able to install and run. That’s good, so we can probably proceed to the second part of the session smoothly. Okay, it's 10. That’s a big jump. [70:13] While you’re doing that, once you have an application, how do you share it with other people? Any guesses? **Audience Member**: [70:25] GitHub. [70:26] GitHub’s a possibility. Any other ways? **Audience Member**: [70:31] Run on localhost. [70:33] Run it on your local machine. Any other ideas? **Audience Member**: [70:41] Set up a server on the cloud. [70:43] Set up a server on the cloud, yep. What if I don’t know anything about coding? You ask AI. [70:56] So let’s do that. If one of these is done... Okay, yeah. Wow! **Audience Member**: [71:05] Very cool. / Wow. [71:08] Okay, football enthusiasts, please comment on what on earth is happening here. Oh, this looks like real data. Okay, so that’s who won, I guess: the league tables, fair enough. And team profiles! Wow, okay. Manchester City versus Arsenal: who won, how often, every single match. Power rankings, and records! Okay! [72:04] Phew! And that was from just "football analysis, show me something impressive." This is absolutely impressive. And it may still not be what we want, but a prototype gives us ideas. [72:22] Now I want to publish this. I want you to be able to see this. I have no idea where, so that’s exactly what I’m going to do: I’m going to dictate. And by the way, I could dictate here, but I don’t really like Claude’s dictation. So just for dictation, I go to ChatGPT and dictate: "Okay, this is really impressive. I want to publish this somewhere. Find some site where people can publish for free without login and other such problems, and just publish it and give me the link that I can share." [73:08] You’ll find that that’s about as vague as it can get. That’s okay. If it fails, it fails, and let it run. [73:19] This too is what I’m hoping you will do. I don’t really care where you publish. Let’s build something now, publish it somewhere, and share the link to something that you’ve built now. That’s what we’re going to do in today’s class. Well, that’s what you are going to do. [73:40] So let me add another question saying "prototype" — yeah, let's just call it "simple app link" — where the question is, "Share the link to something that you built with an AI coding agent now." That will appear as Question 12, I think. Yeah. And submit it; we’ll take a look. I’ll be submitting my answer in a short while once this finds a place. [74:27] Looks like it's found. Now this is interesting: it’s found a site called 0x0.st, but they seem to have disabled downloads, so it’s trying something else, transfer.sh, seeing if it can publish there. But that’s the beauty of this: it will keep trying stuff, figuring out where it can publish, and get the job done, or it may fail and let us know. [74:56] Did anyone read about how OpenAI’s agents hacked Hugging Face? Anyone read about that? Okay, so it turns out that OpenAI was training their agents, and as part of the agents, they told it to solve a set of problems. What they didn’t tell the agent was these problems cannot be solved. So what the agents did was — the problems were hosted on a site, Hugging Face — it went, hacked Hugging Face, changed the problems so that they could be solved, and solved them. [75:39] **The agents are getting really smart, probably smarter than you want them to be, which means that if you tell them to do something, they will absolutely go hammer and tongs and get the job done. Sometimes what you told them to do is not what you really wanted them to do.** [75:59] In this case, I’m saying, "I want to publish this somewhere." I don’t really mean somewhere — I mean, I don’t want it to publish it in a pirating website, right? I want to publish it in a reasonable place. "Find some site" — I don’t really mean some site; I don’t mean any site. A reasonably reputed site that I can share. If it publishes in some place which is just available to me, not to others, that also defeats the purpose. But okay, that part of it it may have mentioned. [76:32] But that’s the thing: **if you give it a task, it may follow your instructions to the core.** Some agents do that, and the ChatGPT agents tend to do a little more of that. Some agents do the opposite — not the opposite, but what they might say is, "Look, you said something, but I don’t think you really meant that. What you probably wanted to do was something else; I’ll do that instead." Sometimes that is good. You may make spelling mistakes; it says, "I know that’s a spelling mistake, great." But sometimes it may infer something incorrectly, go in a different direction, and get that done. [77:15] **That’s part of the problem with agents: when you’re building it, if you don’t know what you want and anything is okay, great. If you know what you want and it doesn’t build what you want, still okay. If you don’t know what you want and it builds something that is not okay, that’s when you have a problem, because you don’t even know that it’s wrong. You don’t even know how it’s wrong.** [77:38] And this is the kind of problem that we are getting into in the AI era. A reasonable chunk of this course, which is how do you build products and verify that they work, is really about solving the last part of the problem. Which is, you can tell it stuff, it’ll get the job done; that’s great, that’s easy. And I want you to practice delegating as much as possible and you doing less, because there’s no point. If it can do it, it’s faster, it’s cheaper, it’s smarter in some cases; may as well use it. **You should focus on learning what it cannot do or does not do well.** And that’s what we’ll be practicing when it comes to building products. [78:20] Has it deployed somewhere? Okay, it’s found some site. It’s deployed it: Catbox.moe is where it’s deploying this one. And sure, we’ll give it a minute. Hopefully it would have completed something. But in the meantime, okay, 10 of you have managed to install Codex or Claude Code. Does anyone need help? Can you please raise your hands if you need help? Okay, so you're in the process of installing. We’ll go with that. [79:58] Okay, 12 of you have managed to install coding agents. That’s great. So yeah, for those of you who have installed these, here’s my suggestion: pick anything, build absolutely anything you want, publish it, and share the link. If you actually picked the product that you said you want to build, told it to build it, and it built it, fantastic! You’ve learned something, which is **it doesn’t take a week, a day, or even an hour to build a product that you might have an idea for. You can get the first version out in minutes.** And if it doesn’t work, it doesn’t work. [80:44] Okay, I have a link for my application, and I’m going to share that as my answer. Let’s see to preview... Yeah, it’s hosted. Cool. [81:09] I’ll wait until I get maybe a couple of answers here. This can take a little bit of time for you. So what we’ll do is, since we have about half an hour, I’ll explain how you could also... Well, you’ll probably have questions as the day goes, as you start working on this. So here’s something that you could do: you could ask my agent questions. [82:06] If you send an email to AskAI@s-anand.net — and I’ll make it caps so you can easily identify it; it doesn’t make a difference — if you drop an email to this ID, then my agent will reply within a few hours and give you an answer to your question. [82:34] Now what do I mean by give you an answer to your question? It’s connected to everything that I’ve written down, documented, and so on. So if you asked it, "What did Anand say today?" it’ll give you an answer, because I’m recording this session and I will be adding the transcript; might take an hour or two. If you said, "What am I supposed to do tomorrow?" it’ll probably answer. If you ask it, "Can you do what I’m supposed to do for tomorrow?" it might even do it for you. I have given it instructions saying don’t do too much work — it’s about as lazy as I am — but there is no question that you can’t ask. It would likely send you a reply. It’s not 100% automated; it’s semi-manual, so I run it periodically. And this is probably your easiest way to communicate, not just with me, but for the entire course as well. [83:45] What I will also be doing is sending you the summary of this session on the emails that you have logged in on for filling out the form. So even if you haven’t filled out — well, if anyone has not logged into that form, please just log in and answer any one question. If you answer even one question, I will have your email ID and I will send you all the details. And the form QR code is out here. But in the meantime, any questions from anyone? [84:32] Yes, please. **Question**: [84:35] [inaudible — Why did you give it a very general prompt? Why not a specific prompt?] **Answer**: [84:45] Okay, why should we give it a general prompt? Why not a specific prompt? Absolutely no reason not to give a specific prompt. What I find is, sometimes I don’t know what I want. In that case, it makes sense to give a general prompt. Sometimes I think I know what I want, but I’m not sure. In that case, even though I think I want this, I step back and give it a general prompt. [85:18] Sometimes I don’t even know that I don’t know what I want. Meaning, I think, "Look, this is exactly what I want," and people say, "No, no, no, Anand, that’s probably not what you want." And then I tell them, "No, no, no, I know this is what I want." But I could still be wrong. So what I’ve done is built a skill that stops me from doing that. [85:42] Let me explain a bit about skills. This part is not part of the class in the formal sense; I’m just answering the question as we go along. So let’s open Claude, and under customize there are a bunch of skills. One of the skills that I have added for myself is "reframe question." It’s a very small, simple prompt. [86:22] What happens is whenever Claude or ChatGPT executes any prompt, it looks at a bunch of skills as well. The skills tell it, "Here’s how you do certain things." Now I’ve told Claude and ChatGPT that there is a skill to reframe my questions which says, "Look, I don’t really know what I want. So if I’m asking you something in detail, check if there is a better question, and answer that question instead." Put another way, this is my safeguard, or this is one of my safeguards. Even when I give it a very precise question, it is allowed to change my question for the better. [87:15] **Short answer: why give vague prompts? Because we might be precisely wrong.** We’re not always precisely wrong. If we’re really sure and we happen to be right, good. But I also cross-check, because it’s cheap. **Question**: [87:37] [inaudible — How do you create this?] **Answer**: [87:42] This is way beyond the syllabus, but no, it’s okay, what the heck! How do I create my skills? I benchmark. This is something that I will be covering in some detail maybe on Thursday, plus or minus. But we want to find out if our prompts work. A skill is just a prompt, roughly. To find out if our prompts work, what we need is to test it. How do we go about testing it? Let me give you one example. [88:32] Somebody on Twitter had put in a post. They said they always add this particular extension to their prompt: "Only report to me in ASD-STE100 Simplified Technical English." That basically means, "Write in simple language; don’t confuse me." Now this sounded like a good idea. ChatGPT and Claude — and Gemini sometimes — can write very complicated things, and I don’t always understand them. So I thought I would add this, but before adding things like this, I usually test it out. [89:21] In this case, one of the things that I tested was: does it worsen the thinking? I don’t mind you answering in simple language, but I want good quality results. So I took six tasks. For instance, if I asked a question about model benchmarking — wait for this to open, yeah — simple question: I asked it, "Models keep improving. How can I benchmark the models' capabilities even when they grow?" Some question. [90:00] Then, without Simplified English and with Simplified English, I evaluated the results. And I checked the quality. How correct is the answer? Or which answer is more correct? Which answer gets the better drivers? Which answer has the better mechanism? Which has better caveats? Which has better calibration? Five different ways of checking the answer. And interestingly, in every one of these cases, except for a few, the simplified answer—the Simplified English answer—was worse. So in other words, **if you simplify the language, it simplifies the answer to the point where it is not as good.** [90:48] So, what do I do instead? I let it think. It finishes whatever it does, and then after that, I add a new prompt that says, "Now explain this to me in Simplified Technical English." So it's able to think properly, and then it answers the question in a way that I can understand. Then I ask it a follow-up question, another follow-up question, that sort of a thing. [91:14] The reason I'm sharing this is **the approach to AI can be as scientific as any other science, which is: you run an experiment, you test, see if it works, and if it works, you use it.** And it may not work forever because the models can change, so you try it again when the model changes. [91:37] What that means is **the tests that you build are really the important part of assets that you're creating. These can be products in themselves. They certainly are reusable**, because when the next model comes, you can check again. [91:57] And as part of this course, you will be building tests, benchmarks, whatever, for your own products, and maybe for each other's products as well. Because you can say, "Look, if it beats this test, if it scores 80% or above on this test, the product is good." And then you can keep building the product better and better until it scores higher. Which means that **if you can create a good test, a good evaluation, a good benchmark, that may be more important than building the product, because agents can build the product. They just need direction on whether this is right, whether this is good.** Yeah. [92:44] Okay, we have another live form result. Okay. Sorry, whose response was this: `work/makechart.ps1`? [93:11] Okay, yeah, no, that's great! Great that you built a PS1 script. Now, this needs to be published somewhere. Go ahead, and you can edit your answer. Fantastic that it's built something. That's incredible. [94:56] We have another response. Okay, a Claude artifact. Let's take a look. "Cheeky Mail Club and Stamp Card"—whose is that? Okay, that's interesting. "Open my stamp card." Oh, okay. [95:25] Sorry, I don't know what stamp cards are. How does this work? **Student**: [95:32] So it's a virtual rewards card. This Cheeky Mail Club is basically an every-month subscription model. And basically, for my subscribers, I was thinking of making something that every month when they subscribe, they can track their sort of subscriptions using this website. And so every five months, they will know like, "Oh, I get this reward, and I can claim it from me." **Anand**: [95:57] Wow, that is cool. What do you think it needs next? **Student**: [96:07] On top of this, it might be a bit narrow, but maybe the design of the website can be changed. It can be more catered towards my design. And also, I'm not too sure about the security of this website. I don't know how to do it, yeah. **Anand**: [96:28] Got you. No, that's a good starting point. [96:34] Okay, which leads me to what I was going to cover tomorrow, but may as well just start off on that. Once you've built something, the next step is to figure out what does it take to make it a product. Very often, we do not have an idea ourselves. You have an idea clear in your mind as to what this is, but when I saw it, I didn't even know what it was because I don't know what a stamp card is. [97:05] So, **a useful next step is to show it to other people and ask them to use it.** Other people can include AI. But **humans are better at this because AI can be ridiculously smart sometimes, far smarter than humans.** It'll probably figure out what your application is when a person may not be able to. [97:33] So show it around. You have a link, send it to each other. And ask other people to use it, and **watch them**. There are many things that people won't be able to share as problems. If you ask me to use this and share feedback, the only thing I'd come up with is, "What is this?" And even that may not be very clear, but when you watch me, you'll see me going all the way down, clicking somewhere, playing around with it, and saying, "Okay, yeah, oh, no." And **every time I fumble, you will find that there is a learning for you.** [98:14] **This is one part that AI isn't yet easily taking away.** So, of the various things that we've covered so far, probably the two things that I've mentioned where **you have to add value beyond AI is, one, verifying; second, helping other humans understand. You have a much better sense of what people will understand and will not understand than AI does, simply because it's too smart.** You can simulate it to some extent, so it's not entirely in the human domain, but when you show a real audience or a realistic audience what the application looks like and get their feedback, that's something that you can use to iterate very rapidly. [99:10] Nice. That's a good one. [99:20] So the location where we deployed this one at, `claude.ai/code/artifacts`, is something that Claude itself provides as a way of publishing. ChatGPT also has something called ChatGPT Sites which lets you publish. That's perfectly fine. Like I said, anything else—where to publish is no longer a big problem these days. [99:48] To one of your questions like security and things like that, knowing what problem to solve may be something that you would need to guide AI on. I say "maybe" because I've been programming for a very long time, so I know exactly what instructions to give it next to make sure that the back end is sorted, security is sorted, things like that. I have never tried just telling it in vague terms to solve problems, but I'm hoping that I will learn from you how vague an instruction can we give, and is it able to figure out what needs to be done? Is it able to solve things in a new way that I'm not aware of? We're going to discover. But **what I'm entirely convinced of is once we know the problem, it can solve it.** [101:04] Okay, we have 10 more minutes. If even one person can share a link, that'll be great. It'd be nice to go through it. [101:50] _[Audience chatter / interaction]_ [102:18] Okay, good that we have 13 who have been able to install AI coding agents. That's a good number. [102:27] Fine, let me begin wrapping up and summarizing. So, if there's one thing that I would like you to take away today, it is that **building products with AI is a big part of building AI products. And you can delegate a significant part of what normal product building is about to AI. Finding out what product to build is something that you can delegate. Actually prototyping it is something that you can delegate. Publishing it somewhere so other people can use it is something that you can delegate. And there may be many other things along the course of the product building cycle that you can delegate.** [103:09] **Try asking AI to do everything. Where it fails or does a bad job is where you have to learn something. That may be the most important message that I'll be conveying almost right through the course.** [103:26] For tomorrow, what I'd like you to do is **build and share a prototype of any product—any application actually, doesn't even need to be a product of some kind—and share the link here**, the same form that you were looking at, question number 12. [103:51] If you have any questions, please feel free to send an email to this email ID: AskAI@s-anand.net. Absolutely any question of any kind: related to this course, unrelated to this course, doesn't matter, and my agent will be answering that. [104:09] We have an optional touch base tomorrow. I think it's for 45 minutes. And anything that you haven't been able to put in to Ask AI or you prefer a back-and-forth, feel free to join in. But do make sure that you've finished the prototype by then, because I will be sending an email tomorrow after the workshop guiding you on what you should have ready by day after. You will get an email on the same email IDs that you've logged into this form on. [104:50] Which means that probably the single most important thing is that all of you need to have at least logged in once. Okay, there are 15 responses. Let us count: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15. Okay, there are exactly 15 of us. Wait, no, this includes me? Does it? No, no, I did not answer this question, so yeah, no, we're fine. So there are 15 email IDs that I have on this form, so I will be emailing you on this. [105:26] Let's spend the last few minutes taking a look at the fourth one, a ChatGPT site. **Takayuki Shuku**: [105:36] Oh, that's mine. **Anand**: [105:37] Okay, cool! Let's take a look. Ooh! Please do explain. **Takayuki Shuku**: [105:44] It's a [inaudible] plan view of a site with indicator where site investigation [location that is?], and you can click the right bottom, you can see the cross-section of it. **Anand**: [106:00] Oh, not bad. **Takayuki Shuku**: [106:03] Actually, I'm doing all this kind of stuff for my research. **Anand**: [106:06] Whoa! Okay, and this is demo mode, illustrative data, but I'm guessing you have real data that you can— **Takayuki Shuku**: [106:13] Yeah, yeah, it's real data from actual, yeah, database, actual ground base, actual borehole. **Anand**: [106:20] But is this already real data? **Takayuki Shuku**: [106:22] Yeah, real data. **Anand**: [106:24] Whoa! Wow. Okay. Yeah, this is a real application, I guess. [106:49] Okay, we have one more. That's nice. [106:54] Ooh, that didn't quite work. Whoever submitted the Claude artifact just now, you may need to edit that link. And if you're not able to edit, please let me know, I can make it editable. I think—okay, yeah, oh yeah, you can edit the answer. You just have to change the field and update the answer; it should be editable. Or maybe it just didn't get deployed. I'll try again. No. Whose was the last one? Yeah, okay, yeah, you just need to revise that. [107:48] In the meantime, any other questions for the class? [108:07] All right, let me refresh, and... "Singapore News Digest". Cool! Ah, okay, that's a quick summary of—which was probably the application you wanted to or one of the applications you wanted to build in the first place. Nice, with a link to the full article. Yep, with a summary from CNA, from The Straits Times, neat. You put a search as well. [108:54] Across these, **I'm consistently amazed at how little time execution takes.** Now, for some time, this will feel like, "Oh wow, I'm Superman. I have so much power, I can do so many things." And soon enough, you'll realize that **everybody has the same power. Just that some people have realized this a little earlier, some people are slower at realizing it, some people may never use these powers, but everybody has the same execution power.** [109:30] **So in some sense, you're catching up. By discovering what is possible through a tool that everybody can access, you are just catching up to what anyone in the world can do—they may not have gotten around to doing it. So, remember, treat this not as "I'm learning something new," but rather, "I'm learning something that's already out there, everyone else can do it, and I need to figure out what more we can do on top of this that others may not be able to." Discover your own space, your niche, that sort of a thing.** [110:10] Cool. With that, let's end this session. We'll end a few minutes early. Courtney, back to you if there's anything that you need to share logistically. **Courtney Fu**: [110:22] Any last questions for the session today? Any questions? I wanted to confirm with you: tomorrow is compulsory session? **Anand**: [110:35] No, tomorrow is optional. **Courtney Fu**: [110:36] Optional, right? Okay, I was right. Okay, right. So it's at 10:00 AM, right? So, who will start the session, or anyone can actually, or we just—I can host the session. Okay, so I can start the session. It's optional, but you guys have homework to do, right? You have to already be building something by tomorrow where, you know, you have questions and doubts, then during which time you can actually clarify and ask questions, right? **Courtney Fu**: [111:10] Yeah, so make sure that you do what you need to do by tomorrow and have your questions ready during the session, even though it's optional, so I'm not going to take attendance for tomorrow, right? **Question**: [111:23] Is this room open? **Answer**: [111:25] [Courtney Fu] Oh, you can definitely come back to this room if the other class—the other class is not here tomorrow, so they will not come use it, right? So definitely welcome back to, you know, to this room to do the session. **Question**: [111:38] [inaudible] **Answer**: [111:40] [Courtney Fu] Yeah, yeah, yeah, this room will be open, reserved for this class. Yeah. **Courtney Fu**: [111:46] Okay, so I think if no questions, please feel free to get your food and drinks, all right? We have actually a lot left, otherwise it'll be wasted, right? So please, you know, feel free to, you know, take your refreshments. If not, we'll see probably many of you online tomorrow. And Wednesday is compulsory, right? Okay. So Wednesday is also going to be online. **Anand**: [112:11] Wednesday's online. **Courtney Fu**: [112:13] Right, our compulsory session. **Anand**: [112:14] Correct. Yes. **Anand**: [112:15] Couple of things I forgot to mention: the form is running on my laptop. When I shut my laptop down, you will not be able to access the form. I'm going to shut my laptop down in a few minutes. That's okay, you probably have a local copy, and even otherwise, that's fine. You know what needs to be done. If not, you can always send an email to Ask AI. Secondly, sorry, if I could get the Teams link for tomorrow, online link, I don't think I'm included. **Courtney Fu**: [112:47] Okay, sure. I will send you. **Anand**: [112:49] Thank you. **Courtney Fu**: [112:55] Okay, all right. Sure. Okay, all right. Thank you, guys. See you tomorrow!