 Saya Tuyak. Saya basically started as a developer. Now I'm an AI apprentice at AI Singapore. So if you haven't heard about AI Singapore, it's actually a national government initiative to train you guys to do how to do AI technologies and machine learning and deep learning sort of thing. So if you haven't heard of it, you can check it out. Just type AI apprentice program and you will find it. It's a nine months full-time program. They pay you to learn stuff, so it's cool. So this is me. I'm currently an apprentice at AI Singapore. One of the side things that I do, I'm also an even manager at Geese Hacking. Geese Hacking actually is a tech community. We sort of embrace tech, innovation sort of thing. If you are a junior developer that want to learn new stuff, you can check out our events. We also have free events to train you online. Mobile development, work development, all sorts of stuff. So this is my backstory. I study IT from Republic Poly, solid go to Commitment Science, and also after I graduated, I worked for RPA Analyst. It's basically automation of business processes. So you sort of automate critical issuaries, digitising sort of things into... We are basically writing scripts to do this kind of thing. So it's sort of like a developer job, but you're not building application, you're writing scripts. So from there, I progressed into AI apprentice, where I learned about machine learning, artificial intelligence, and de-learning. So yeah, these slides are really based on my experience. So if you have different experience, the experience might differ from how I experience it as a developer that go into AI. So take it with a pinch of salt. So there are quite a few difference that I can get from being an AI practitioner versus developer. The first thing is jobs code. How does it differ? So as a developer, you are just writing rules for the program to follow, whereby if you are AI practitioner, you are sort of programming the machines to learn the rule by itself. So you don't write rules. For example, you are trying to classify between cats and dogs. If you are a developer, you will write cat is something like small, cute little things with like a pointy ears. So if you are a dog, it's like a long term, longer nose sort of thing. You're writing all sorts of rules to classify all these pictures. But if you are trying an AI approach, you probably just keep feeding the algorithms with all the pictures. Oh, this is cat. This is cat. This is cat. It recognizes that the picture is cat. So it's a little bit different. I have drawn out this diagram. So what you do as a developer is basically you have input and you have a program. You process it. You try to compute it and then you have result. So for example, you try to call API call. You write API, you have input, API key, then you try to write API call. You press button, you get back the result. That's developer. But for AI, you have input and you have output. You don't know what to write. How do you come from this input to output? So you just feed entire thing that input output into the computation which is like the program. Algorithm will generate a program for you. So you don't have to write like specific rule what to do and what not to do. Then the top process. The next thing is top process. If you are a developer, you probably think about apps. Let's say you want to do some project, right? They will say, oh, what do you want to do? Are you want to do web app? Move all app? Are you going to build on like or Kotlin, Android, C-sharp, C++ but as an AI apprentice or the AI practitioner, you have to think about the data. What kind of data you have? You cannot just build like like very good machine learning model if you don't have any data at all. So you need to have like input and output to get the process, the program by itself. So you need a lot of data. Like maybe you are trying to classify whether it is dot or cats. Maybe try to get identify like criticat fraud by using a lot of transitions. Or maybe try to get like try to recognize the faces. Like, you know, if you try out like image recognition, you try to detect who you are like Facebook you try to detect who you are, who your friends are, recommend you text or sort of things. So all these require data. That's why Facebook let you like upload photos for free or maybe Google upload photos for free. This is how you're paying, by uploading your data and like tagging your friends. So remember if you remember like 2008, something, there's no such thing. So you just have to tag your friend manually. Like you will go to your face and tag them, tag them, tag them, right? It's basically you're training their model. So once they get it, now they are using it to like recommend you. Oh, is it your friend? Is it your friend? Yeah. So it's you that, you know, train their model and you're working for that intentionally. As a programming skills, right, like we are developers so you need to have a lot and like you need to like in depth sort of thing. You need to learn like Python, C++, we've been learning a lot. I've been from IT all the way to combine science. I learn programming every day, you know, do recordings. I have to learn in the like multi-training, synchronizing, you know, ABI call, asynchronous, so many things, right? But if you are AI apprentice, your focus is a bit different. So you won't be learning in depth into how to program things because it will be automatically program, right? So you have to, you will be learning more towards understanding the context which is like math, like the formula, how you're going to interpolate from like the input to output. Like you have this sort of data, for example, you're trying to predict HGV, housing price in 10 years' time. So you have to know what kind of input that you can put in, then what kind of output you're expecting, then you try to come up with like math formula behind it to understand it and if you don't have math background, you sort of can't relate what's happening with the data and how the program is going to give you this kind of result. So you don't need very in-depth programming skills. You need a handful like maybe Python, R, SQL, no SQL, maybe API calls sort of thing. But the more intensive things you need is to understand the algorithms and the math behind it. So the next one is math. Like how did you have to go, right? So as a developer, you don't really need to do a lot of math. Maybe like brilliant algebra for some operations. But as an apprentice or AI program, you have to do like statistics, probability, you know, even like integration, differentiation, all sorts of things have to be important. You have to be familiar with all the symbols that they write, you know, all the Greek letters. Like when I see it, I just want to throw your people away that kind of level math. So, yeah, if you understand it, like when you get into it, right, you understand it, you read it like a code. Like when you try to read, for example, you try to read as that function, right? It is the same, like if you understand it, you can read the codes. Like, if you are a developer, your project manager or your clients will just give you like, oh, just use this data, like a dummy data, right? Maybe 4 number or 1, 2, 3, 4, 5, 6. Then you just use that to try to program, you know, this input is value or not, try to use rejects to like validate 5 number, 6 number, 6 number run alphabetical order, right? So, you don't really need like real life data to produce something which is doable, but doing AI application you really need a real life data because what you need is input and output, right? It has to be real to get a real program that can do execution. So, if you don't have a real life data, too bad your AI program will just predict wrong stuff all the time. Oh, then this is like one of the main things that a lot of people are talking about is reproducibility. What it means basically is try to reproduce whatever you have done, right? Like one input how often you can try to reproduce that? You can always reproduce 3 button with one input, all the time. You don't really need any specifications or any data format or anything, but for AI applications the requirements of the data like how many columns do you have, how many rows do you have, what are the data when it was taken all sorts of things are very important for the programs to predict accurately. So the tolerance of having around data is very low for AI application. So the data is like the key things that you have to be care of. So if your data is dirty doesn't make sense your AI doesn't make sense as well. So for the accuracy as a developer you have to be really like 100% sure that your buttons will work. Let's say your program e-commands website your checkout button, right? You cannot say checkout button 80% of the time. Your boss probably will fire you for saying that so your checkout button all sort of lock in buttons have to work 100% of the time. But for AI applications the accuracy doesn't really dead. No, not really straight. F2P comfort level for the business to make decisions. So you can say I think HGP price will highly go up by 2% in the future and no one knows in the future so they're like I think based on these sort of formulas I think it's okay. We accept it that 82% so this kind of comfort level that you can play around so you can really get 100% accurate models just to be clear so anything above 85% that's quite doable already. So yeah, the most $1 million questions what AI replace developer? What do you guys think? Yeah, I think it's probably not a lot of people but there's like 200 article talking about it. So I think it's probably not because developer we have some foundations that I don't think AI can replace some logics that in our head is happening. Whenever we progress something there's some unspoken rules or spoken things in our heads happening that AI can never replace. So yeah the next one this code is in C it was trained on Linux repository. It can write this kind of code. So it's only trained I think 2 days. Do you think you can write this kind of code in 2 days? Yeah, it isn't legit if you take a look at it. Yeah it has like for loop it has some comments in there it isn't legit written by some experienced programmer but the good thing is you can compile that so you can push it to give hard report and if you want to try it out there's the GitHub profile as well and you can go and check it out. He has written like entire code and entire programs the AI programs to you know get into this kind of states. So if you are free maybe you can just take a look keep training your model entity you can code like you then you can just drink coffee every day just let it runs like I've asked you to do some fisheye right just let it runs just go for coffee or play games whatever man So if you are interested to get started the good thing about AI everything is free you don't have to pay money to learn it there's a lot of good youtube videos that you can find there's Coursera course which is machine learning to intro this is like go to machine learning intro whatever you are not sure about meh formula or whatever the machine learning intro given by the professor Andrew Ang is very good and deep learning intro is there as well and the medium is actually like a step overflow for AI practitioners so everything that you want to do for example I want to generate just now the code actually probably someone has already done it the guides are there how you set it up how you run it what kind of things that you should expect to do all these are there and the cargo is like a challenge thing so what they do they sort of upload the data because data is a more important thing they upload the data they give you a promise statement you try to solve it using the machine so if you can like beat other people in accuracy or maybe like simplicity wise they will reward you can be like 1 million dollar if you just keep trying out yeah the prices are too high you just take a look it's amazing yeah and data camera is the intro causes which is free if you want to try out the tracks you have to pay yeah but if you are just like willing to try it out you can just drop then an email saying that you are developer you are interested in this they are very nice they give you 3 months access and you can finish it in like 2 months so yeah and if you haven't heard there is AI Singapore approach which is like government funded agency we have a lot of programs like apprentice program there is AI for E AI for everyone program which is like runs in a day half a day around 9 to like 3 or 4 yeah we sort of like give you what you should be expecting AI then there is if you are like industry then there is AI for industry they train you on like data cam data science track so you have someone to like get you along if you have any question you can ask also do a question so if you check out right actually it's all government funded you just have to pay like certain like 20% I think I'm not sure about the cost yeah but AI for E is free AI for I you need to pay a little bit more yeah that's it and yeah if you have any question I'm happy to take either here or at the back yeah if you don't want to talk to me in person you can always terima kasih