Transcript of 미국 빅테크 일자리가 한국으로 왔다!
홍정모
0:00Since the talent couldn't go there, the0:02 jobs came here. Yes. Big Tech jobs have0:04 been flooding into Korea. When I0:07 analyzed the job postings, it turns out0:09 these are roles they couldn't even fill0:11 in the U.S., so they're coming over.0:13 They are all high-level positions.0:15 They're really hiring. Yes. I'm telling0:17 you, they are really hiring. The jobs0:18 are coming to us. New Korean recruits0:20 are outstanding. Since Koreans in the0:22 U.S. perform well, they obviously know0:24 that. A somewhat absurd situation has0:25 occurred where the jobs are coming to0:27 us. I'm a backend developer with 70:28 years of experience. Changing fields0:30 isn't easy, but web backend roles0:32 building systems with AI—that’s0:34 where the highest pay is right now. But0:38 the burden of accumulating knowledge0:41 one by one or memorizing things has0:43 been greatly reduced. You must never0:46 compromise on sharpening your brain's0:48 thinking circuits. You just need the0:50 circuits in your head. Actually, it’s0:53 because Korea has a firm grip on the0:55 manufacturing industry. But we have to0:57 protect that. That is the mission of0:59 your generation. You must protect it no1:01 matter what. Now, today's topic.1:03 Today's topic is that U.S. Big Tech1:04 jobs are coming to Korea. I will talk1:06 about how you should prepare for that.1:09 I investigated about 500 job postings.1:13 As I looked into it, I found there are1:14 patterns. There are commonalities as1:16 well. I'll start by talking about the1:19 areas where I see opportunities. Big1:23 Tech companies are actually hiring in1:25 Korea. Honestly, at first, I thought1:30 they were just trying to push sales or1:33 marketing roles here. But when I looked1:38 into it, they were hiring engineers and1:40 researchers. It wasn't that. They are1:43 really high-quality jobs, senior1:45 engineering roles based in Korea—not1:48 even remote work—so the jobs have1:50 come to us. So, Korea is overflowing1:54 with talent. Overflowing with talent.1:57 We have tons of talent, which is why1:58 we've been exporting them. To the U.S.2:00 Yes. Since the talent couldn't go there2:03, the jobs came here. Yes. Big Tech2:05 jobs have been flooding into Korea. So2:08 I’ve been tracking and investigating2:10 this for months. It’s not just a2:12 temporary thing, either. And there is a2:14 reason for this. So I don't think it's2:16 a short-term trend. I expect it to last2:18 for quite a long time. Geopolitical2:21 factors don't change that quickly,2:23 after all. It’s an opportunity. An2:25 opportunity. Big Tech is coming here to2:28 treat you well. The reason they are2:31 doing this is, in fact, because Korea2:33 has a firm grip on the manufacturing2:34 industry. The infrastructure is2:36 stronger than you think—it's very2:37 powerful. But we have to keep it that2:39 way. That is the mission of your2:41 generation. You must uphold that. You2:43 have to do that. The reason I emphasize2:45 C ++, graphics, physics, and the2:47 fundamentals of LLMs and RAM is because2:50 that is your mission. You must uphold2:52 that. You must uphold it no matter what2:54. Without this chaos, we would just2:57 have to stand in line. This means3:00 people with a score of 100 get into3:02 good places first, followed by those3:04 with 90, forcing us to live a life of3:07 standing in a single line. Unless you3:11 possess exceptional abilities to3:13 guarantee a score of 100, these changes3:16 in the world and technological3:18 advancements are actually a form of3:20 positive chaos. Therefore, when we go3:23 through that process and things start3:25 to settle, it is actually a fantastic3:27 opportunity. These are just big tech3:29 companies. I’ve gathered big tech3:31 companies here. Looking here, Riot3:33 Games is also hiring in Korea. Wow,3:36 there are so many. After analyzing3:37 these job postings, it’s clear they3:39 aren't outsourcing just to save money.3:41 They are genuinely looking because they3:42 need people. These are positions they3:44 couldn't even fill in the U.S., so they3:46 are looking abroad. The jobs I have3:48 gathered here are all high-level3:50 positions. NVIDIA autonomous driving3:52 system software. C ++, Python, Linux,3:54 and CPU/GPU architecture—this means3:57 you have to use both the CPU and GPU3:59 together. So, things like heterogeneous4:01 parallel processing are fundamental.4:03 Surprisingly, there aren't actually4:05 that many people with a solid4:07 foundation. Amazingly, that is how the4:09 world works. In some sense, once you4:11 cross that barrier to entry, a new4:13 world opens up unexpectedly. That’s4:16 the feeling I get. Linux kernel4:18 engineer. It requires strong C4:20 programming skills. Yes. You can be a4:22 Tesla employee with C language skills,4:24 and you can even work in Seoul. Living4:26 in the U.S. might suit some, but4:29 honestly, Seoul is better for leisure.4:33 And with your family and relatives here4:36, you can work in Korea with pretty4:39 decent pay. And then, you get "Tesla"4:42 stamped on your resume. Over here is a4:45 Field Deployment Engineer. There was4:48 some talk that there would be a lot of4:49 jobs in this area. Looking at the4:51 languages, they mention Python and4:53 JavaScript, so think about it. In the4:57 past, there were very few cases that4:59 required both JavaScript and Python.5:02 But now, they require both. It’s5:03 written in small print, but it includes5:05 mentions of LLMs and generative models.5:08 To handle neural networks, deep5:10 learning, or things like LLMs, you also5:12 need to know Python. Then, Teradyne is5:16 a semiconductor test equipment company.5:20 But from my perspective, looking at5:21 these job postings, I really hope that5:23 semiconductor engineers, especially5:25 those working in software, get the5:26 recognition they deserve. That's what I5:30 think. It's Qualcomm, and Qualcomm is5:32 famous, right? Qualcomm is famous, and5:34 I’ve seen many people move to even5:36 better companies after working there.5:38 Qualcomm new grad. Qualcomm new grad.5:41 What could this mean? Python and C/C ++5:44. While other companies are poaching5:46 experienced workers, Qualcomm has been5:48 looking for experienced hires too.5:51 Nevertheless, new grad positions have5:52 appeared. What does that tell you?5:54 There’s a real shortage. Jobs are5:56 coming back. But the same jobs won't5:58 return exactly as they were. I’ve6:00 been saying that they come back in a6:01 different form. So, please, study6:03 according to global standards. I didn't6:06 expect Big Tech to start hiring in6:08 Korea like this. New grad positions are6:10 starting to appear. New grad. Once jobs6:13 start to spread, they spread quickly.6:16 As the hype around AI begins to subside6:19, I believe companies will go back to6:22 hiring again. The reason I’m6:25 analyzing these job postings is that6:27 they are the clearest evidence for you6:30 to predict where the opportunities will6:32 emerge when they come flooding back.6:35 Entry-level embedded software engineer,6:37 new grad, commercial—it seems anyone6:40 can apply. They really are hiring. Yes,6:42 I’m telling you, they really are6:43 hiring. Jobs are coming to you. New6:45 Korean graduates are exceptional. Since6:48 Koreans in the US are doing well, they6:49 would naturally know that. An6:50 incredible situation has occurred where6:52 jobs are coming to us. Lam Research is6:54 one of the world's top 5 semiconductor6:56 equipment companies, a US firm that6:57 supplies etching and deposition6:59 equipment to Samsung Electronics and SK7:01 Hynix factories. And they are now7:03 requiring more advanced skills, like7:05 multi-threading. You are all studying,7:07 right? So, I compiled statistics on the7:09 languages. Python is overwhelmingly7:11 high. But the Python mentioned here7:14 refers to Python as a professional7:16 programming language. And interestingly7:19, C and C ++ are even more common.7:23 It’s not that Java is bad, but rather7:25 that there is already an abundance of7:26 that talent, which is why they come7:28 abroad to find the talent they lack. US7:31 Big Tech is coming here specifically to7:33 hire talent that is in short supply. C,7:37 C ++, and professional-level Python are7:40 in extremely short supply. I think this7:44 clearly demonstrates that. Then there's7:46 CUDA, and I think that’s how the7:48 order goes. The Ministry of Employment7:51 and Labor had a survey on why companies7:53 couldn't find the people they needed.7:56 Uh, 25.8%said they couldn't hire7:58 because there were no applicants with8:00 the required experience, and8:02 specifically in IT R&D and engineering8:04 technical jobs, the vacancy rate for8:06 roles they wanted to fill but couldn't8:08 was 16.2%, which is a whopping 2.58:10 times the average of other occupations.8:13 So, the hiring side is desperately8:15 looking. But if the applicants 'skills8:18 don't match, they just can't be hired.8:22 Actually, I’m generally known as8:24 someone who teaches the basics or8:26 difficult subjects, and you might think8:29 of me that way, but what I pursue is8:31 actually avoiding competition. If you8:35 find a good niche market point and8:37 strike hard, a vacancy temporarily8:39 opens up like a vacuum. That’s how8:42 you should push forward. Areas with8:44 barriers to entry are advantageous. Uh,8:47 barriers to entry—it's especially8:48 easy to study nowadays. It’s very8:50 different from the past. So, when I say8:52 now is an opportunity, it was a time of8:54 chaos before. It's heading toward a8:56 stable phase, so the future is8:57 predictable now. Because we are in such9:00 an era, it’s honestly a really great9:02 time for you all to live smart. So, you9:06 just need some proof that your9:07 foundation is solid, and then you need9:09 to be ahead of the curve. Things like9:12 CUDA. And accelerated computing. I see9:14 this as a very good opportunity right9:16 now. The Korean government's investment9:19 direction, and then US trends arrive in9:22 Korea with a slight delay. So, in the9:25 current situation, big tech companies9:27 are hiring in Korea, and they want to9:29 know if you understand the9:31 characteristics of LLMs well. Many9:33 people think that C ++ is difficult. Uh9:36, C ++ proficiency—C ++ is just a9:39 symbol. I’ve said this many, many9:42 times. C ++ proficiency is not about9:45 memorizing syntax and knowing where to9:47 use which grammar. It’s not about9:49 understanding inheritance or things9:51 like that either. It’s about whether9:53 you know data structures and algorithms9:56 well. I mean, we use C ++ because9:58 it’s the language that can implement9:59 data structures and algorithms most10:01 efficiently. If you want to be good at10:03 C ++, you have to do data structures10:05 and algorithms. Next, NVIDIA physics10:07 simulation engineer. Uh, I guess they10:09 really are short on people. Well, there10:11 are a lot of interesting things. Right,10:13 AMD is also hiring in Korea. But AMD10:15 seems to be hiring for gaming roles.10:17 But something very interesting is that10:19 Korean is required. Wow, Korean—are10:22 you all good at Korean? Yes. If you are10:25 good at Korean, you can work at AMD in10:27 Korea. Rendering engineers' market10:29 value seems likely to go up. Next, in10:32 terms of the study order, I recommend10:35 you quickly learn DirectX 3D 11, then10:38 study Vulkan, and then move on to10:40 DirectX 3D 12. A Deep Learning10:44 Applications Engineer is a very10:45 high-level engineer. So, it is10:48 important to study LLMs themselves.10:52 Even if you are just using an LLM as a10:54 tool, understanding the nature of the10:56 LLM itself is extremely important.10:58 After all, an LLM is the culmination of11:00 all the machine learning technology11:02 that humanity currently possesses. Top11:05 talent is focused on it, and all the11:08 electricity and GPUs in the world are11:10 being used to develop these LLMs. From11:12 a study perspective, what is the best11:14 software technology humanity has? It is11:16 currently the LLM. Therefore, studying11:18 this LLM is a very meaningful endeavor11:21 for you all. Furthermore, it is11:23 directly linked to jobs. It seemed like11:26 AI would do everything, but when you11:27 actually use it, there are many cases11:29 where it makes things harder for me.11:32 It’s good at rambling on and on about11:35 things, but when you ask, "So what?",11:39 it can’t answer. But on the flip side11:41, there are things that have become11:42 easier. So, if you definitely delegate11:44 the things AI is good at to the AI,11:46 productivity does indeed go up.11:49 Therefore, I think it is important to11:50 distinguish these things well. The11:53 easiest way to distinguish them is by11:54 studying the principles of LLMs; then11:56 you will see what it will be good at.11:58 And what it won't be good at. It just11:59 becomes clearly visible. So there is an12:01 aspect of needing to study LLMs, and12:03 then there is what we call AX. In Korea12:06, those doing things like AX need to12:08 have a precise understanding of the12:10 nature of AI and the nature of these12:12 LLMs. I think there is an aspect of12:13 needing to understand how to handle12:15 them. Uh, quantitative tasks, simple12:17 repetitive tasks, and tasks that are12:19 just agonizing to do—those have12:21 decreased a lot. They have decreased a12:24 lot, but what you must do instead is12:26 never compromise on your fundamental12:28 intelligence. The era of piling up12:31 knowledge in your head is over. But you12:34 must never compromise on clearing the12:35 circuits in your brain for thinking. In12:37 the past, if you said you were good at12:39 C ++, or good at Java, you first had to12:41 memorize the syntax. But now, you don't12:44 even need to do much typing. Physically12:47, at least, it has become much more12:49 comfortable. Once you actually start12:51 working, you only need to evaluate12:53 whether it was made well. Because the12:57 difficulty of creating a good program12:59 and the difficulty of evaluating if it13:02 was made well are completely different.13:05 You know those 1,000-piece or 10,000-13:07piece puzzles? Putting those puzzles13:10 together is extremely difficult.13:12 However, regardless of whether a puzzle13:14 has ten thousand, a hundred thousand,13:16 or a million pieces, checking if it’s13:18 put together correctly happens in an13:20 instant. Now, does this story come from13:22 somewhere extraordinary? Not at all. It13:24 comes from algorithms. This is an13:26 example used when teaching algorithms.13:28 So, if you don't know this story or13:30 this example, you haven't actually13:31 studied algorithms. Therefore,13:32 there’s no need to increase your13:34 proficiency in areas that AI can handle13:36 for you. You just need to have the13:37 right mental framework. If you get a13:40 feel for the easy things and expand13:41 little by little, you can study very13:43 efficiently. It’s true that it’s13:46 become easier to study. It’s also13:48 true that it’s become easier because13:49 of AI. If you set your direction well,13:52 you can achieve high results very13:55 comfortably. It even drew out how the13:59 500 job postings I analyzed today14:03 connect to what I teach. My approach to14:07 teaching LLMs has an aspect of studying14:09 them as a case study for the latest14:11 machine learning and deep learning14:13 technologies. Three years after I14:17 shouted to study based on global14:19 standards, Big Tech companies have come14:21 to Korea and fulfilled my prediction.14:24 The fact that they are operating14:26 engineer and researcher positions in14:28 Korea is, from your perspective,14:30 absolutely a good thing. Since some14:33 people ask about the order of building14:35 an LLM from scratch, I'll organize it14:37 for you. In Part 1, Principles of14:39 Statistical Language Models, we start14:41 with what a basic language model is. In14:44 Part 2, we move on to examples of14:45 entering the language model field using14:47 neural networks. Part 3 finally moves14:49 on to Transformers, exploring how14:51 modern LLMs began. And I have to say,14:53 even by my own standards, I taught the14:54 principles really well. You couldn’t14:56 visualize this any better. The core14:58 point is that that’s what I’m good15:00 at. Teaching difficult things easily. I15:01 really do teach difficult things in a15:03 simple way. So, I pick out the most15:05 important aspects of Transformers and15:07 teach them to you precisely. After that15:09, there’s theoretical content on15:11 in-context learning, and that is very15:14 important. So, please make sure to15:16 study that. Next, what I’m covering15:18 in Part 4 now is training15:20 conversational LLMs through fine-tuning15:22. Doing fine-tuning. It’s excellent15:24 for studying the basics. Then Part 5 is15:26 building a RAG system, which is15:28 directly connected to what is referred15:30 to as AX in Korea. Yes. And for the15:33 final part, there is building agents.15:38 My lectures are structured with fun15:39 examples, but in the latter part, I15:41 cover high-efficiency computing like15:43 multi-threading and multi-processing. I15:47 incorporated things like rendering,15:49 thinking it would eventually become15:51 such a hot technology. I had faith.15:54 That technology would eventually head15:55 in this direction. Yes, I had that15:57 faith. So that’s what's included in15:58 there. What I recommend is to watch the16:00 Python crash course, check out the C ++16:02 core summary lecture, and if you feel,16:04 "Oh, I think I can do this much using16:06 AI," then you should go straight to16:08 data structures and algorithms. But16:10 here, you absolutely have to use your16:12 brain. Focusing on intellectual effort.16:14 If you study the basics of data16:15 structures and algorithms, especially16:17 the algorithm lectures taught at16:18 universities. As for my lectures, I16:22 took the Stanford curriculum and broke16:24 it down to explain it more easily. Then16:27 there is the algorithms part, and Part16:29 2 includes some more practical elements16:32. My "First Steps in Deep Learning"16:34 course is currently free. So, just try16:36 skimming through it once. If you look16:38 through it and think, "Deep learning is16:40 doable," then you can move on to16:41 building an LLM from scratch. The "16:43Building an LLM from Scratch" course16:44 has a lot of videos. I've put a lot of16:47 effort into summarizing them into16:49 videos, so even if you just watch them16:51 for fun, they contain good talking16:53 points for interviews and content that16:56 can help you sound like an LLM expert16:58 anywhere. There are quizzes, too, and17:01 it’s quite fun. Then there is my CUDA17:03 course, where the focus is on17:04 heterogeneous computing, which means17:06 parallel computing across different17:07 types of processors. And that is17:10 exactly what the market requires right17:12 now. Suppose you’re working with17:14 robots. You need to boost performance17:15 in those robots. Maximizing hardware17:18 capability ultimately comes down to17:20 keeping both the CPU and GPU busy and17:22 configuring them to maintain an optimal17:24 state, which is what this part covers.17:27 Next, in terms of graphics, algorithms17:29 that handle 3D space are inevitably17:32 best developed in the graphics field.17:34 It’s here that you realize, "Ah, this17:35 is what programming with math is like."17:37 And "This is what it means to handle 3D17:39 space." For studying this area,17:41 graphics is, and will always be, the17:43 absolute best. For those who want to go17:48 deeper, you can look into Vulkan and17:50 Gaussian Splatting. Especially for17:53 those who want to study in professional17:55 depth, these are topics that are hard17:56 to find elsewhere. If you've tried SPH17:58 simulation, you’ve probably gotten a18:00 good grasp of a few things. Like what18:02 it means to perform simulations on a18:03 computer. And how to handle 3D space.18:05 Then, how it connects to graphics, and18:08 since we extract polygons to render SPH18:11, right? That technology is actually18:14 called "reconstruction" when building18:16 things like digital twins. It is18:19 directly related to that 3D18:21 reconstruction technology. So, studying18:23 that is very meaningful. I actually saw18:26 a job posting from Naver Labs. The18:29 first thing they were hiring for was18:31 mesh-based 3D reconstruction. Postings18:34 for that keep appearing. So I realized,18:36 the openings are definitely there. But18:38 right now in Korea, while LLMs are18:41 being applied and released as services18:44 that any company can use, the real18:48 battleground is Physical AI. Physical18:52 AI needs to operate in a 3D space, and18:54 what is the most developed field for18:56 dealing with 3D space? It’s computer18:59 graphics. So, approaching it from that19:01 angle is really great. The technology19:03 used in game engines is speed-oriented.19:05 Surprisingly, though, we’re in a19:07 situation where we have no choice but19:09 to use the physics engine technology19:11 found in game engines. Why? Because of19:13 the speed. So there are some19:15 limitations we hit. But new19:17 technologies are emerging now, even for19:18 simulation precision. To do things like19:21 reinforcement learning, speed is what19:23 matters. Reinforcement learning really19:25 needs to be fast. It means you have to19:27 do it a lot. Surprisingly, you can't19:28 really use expensive simulation19:30 technologies. So, we end up borrowing19:33 game technology for now, but studying19:35 physics simulation for games first, and19:37 then moving on to technologies used in19:39 robotics and similar fields is very19:41 advantageous. You mentioned two things,19:45 and both of them are areas with plenty19:49 of job openings. Honestly, job19:51 prospects in other areas aren't very19:53 good. You have to choose between making19:57 robots or building systems for a19:59 company using LLMs; the two are very20:02 different. Considering the hardware20:06 side of things, I don't think it'll be20:08 easy for a data engineer to suddenly20:10 jump into Physical AI. However, for20:13 those who usually do graphics, I highly20:15 recommend heading into the Physical AI20:17 field. Graphics experts are usually20:20 interested in simulations and things20:22 like that, so I think the AI engineer20:24 path would be better than Physical AI20:26 for them. For now, I have a lecture20:29 series uploaded for free on my YouTube20:31 channel. "First Steps in Deep Learning.20:33" Just take a look at that casually.20:35 Yes. You should watch that first and20:37 then make your decision. If you are20:39 proficient in Python, I think it would20:41 be better to look into the LLM side20:43 from the perspective of a data engineer20:45. I'm a job seeker preparing for the20:47 gaming industry. Should I start20:48 studying AI now as well? Well, in the20:50 gaming industry, art and tech are20:52 separated, aren't they? If you are20:54 headed toward the tech side—meaning20:56 if you aren't purely in art—I20:57 strongly recommend studying AI. Game21:00 companies these days are even21:01 struggling desperately to transform21:02 into AI companies. What I recommend is21:04 this. What I recommend is that you21:07 build a robot simulator. Assuming21:10 you've already finished the basics. But21:13 looking at you now, it seems like you21:14 haven't studied data structures and21:16 algorithms. You must do it, no matter21:17 what. Yes. If you're twenty, you must21:19 do it, no matter what. If you're in21:21 your twenties, I always tell you to21:22 just do it. You will never lose out.21:24 Yes. If you're doing physics and things21:26 like that, you have to be good at21:28 coding. Even if you go to graduate21:29 school, being good at coding gives you21:31 an automatic advantage. It will never,21:32 ever mess up your life. My graphics21:34 course is not a course for memorizing21:36 APIs. It's a course that teaches the21:38 principles of graphics, but you have to21:40 use an API after all. So I teach it21:42 using DirectX 11. It makes it easier to21:46 connect to DirectX 12, allows you to21:48 study what an API actually is, and21:50 fundamentally, it's a graphics course.21:54 That math program is really good; even21:57 I think I made it well. Yes. I'm a22:01 backend developer with 7 years of22:02 experience. They say changing22:04 industries isn't easy, but web backend22:06 developers using AI to build systems22:08 are the most highly valued right now.22:11 Practitioners are talking among22:13 themselves, and they're having a hard22:15 time because there's no one good at it.22:17 Among themselves. That talk will start22:19 to break out now. It doesn't work as22:20 well as you think. But they didn't22:21 study the underlying principles. So22:23 they start by building the system first22:24. They buy GPUs, build a system, set up22:27 the backend, and think it'll work, but22:29 the "how" is missing. They thought it22:32 would work, but the "how" doesn't work.22:35 Those who fill that gap by studying22:37 will be treated very well. Why, if22:39 you're a backend developer with 7 years22:40 of experience, you've done well in your22:41 company, right? And you're likely good22:42 at backend. But if you know a bit about22:44 AI. That's an opportunity. This is a22:46 really great opportunity. There are far22:48 more developers on the East Coast now.22:51 They say there are more developers on22:53 the East Coast than in Silicon Valley.22:56 The reason is that places like Wall22:57 Street and banks need to use AI for22:59 their web backends. They need to adopt23:02 AI, but there are no people. There23:04 weren't many engineers on the East23:05 Coast before. That's why I always say23:07 this. Korea just waits and watches, and23:09 then once the US starts doing it in23:11 earnest, they chase after it like crazy23:13. So if you study, there will23:14 definitely be opportunities. The reason23:17 is that companies need to adopt AI, but23:20 there are no people, so if you think23:22 that LLM-based services, which we call23:25 AI, will be used in every company—23:27finance, law, accounting, etc.—then23:30 it's the right choice to study. Next,23:33 for those who worked in back-end and23:35 handled large services, if you have23:37 that experience and also know a bit of23:39 AI. And have built a service that23:41 incorporates AI. I think you’d be23:43 compensated quite well, as businesses23:45 have now decided that AI is essential.23:47 There is no need to obsess over23:49 knowledge. You need to be better at23:51 handling knowledge that is complexly23:53 interconnected. And algorithms are what23:56 do that. I thought people would23:58 naturally study data structures and24:00 algorithms. But surprisingly, they24:03 don't. I was actually quite shocked by24:05 that. Grammar exists to implement24:08 something. It’s a convenience feature24:10 meant to make things easier. But if you24:12 don't know the intent behind those24:13 convenience features, it just looks24:15 complex. That is why I tell you to24:16 study data structures and algorithms.24:17 Paradoxically, since you don't need to24:19 memorize that grammar anymore, the24:21 generation that maintained their24:23 competitiveness by memorizing grammar24:25 is disappearing. There is no need for24:26 that anymore. From the perspective of a24:28 student, things have gotten much better24:29. Conversely, if that was your core24:31 identity, you will be pushed out. When24:33 I created my Python course, I was very24:34 careful to ensure that the difficulty24:36 of the introductory practice problems24:37 was something you truly had to solve24:39 with your own brain. Exercises that24:42 purely cultivate your own mental24:43 faculties are truly important. Whether24:46 or not you have that will determine the24:49 course of your life. It's certain. As24:52 AI becomes more powerful, your24:54 fundamental cognitive abilities will24:57 need to work even harder. It will be25:01 even more intense than in the past.25:03 Things that you could get by with25:04 through memorization are, in fact,25:06 being swept away. So, things like being25:08 a "human encyclopedia" are no longer25:09 needed. If you are incredibly good at25:12 data structures, algorithms, and using25:14 AI for coding, you will be successful.25:16 Yes. There are barriers to entry, but25:19 once you cross them, you will do well.25:23 Yes. And those barriers to entry have25:26 changed. It has shifted from25:29 memorization to understanding and25:31 building neural circuits in your brain,25:33 so once those circuits open up, you25:35 instantly cross that barrier. Then,25:38 it’s about how well you can control25:41 AI. Also, can you avoid being swayed25:43 when the AI talks nonsense? Can you25:45 judge what the AI has created quickly25:48 and accurately? Everything is shifting25:50 toward this. For instance, if you look25:53 at snowboarding, experts glide down25:56 smoothly, effortlessly, and handle it25:58 with such ease. But beginners struggle,26:02 asking why they keep falling, and it's26:04 extremely painful if you don't grasp26:06 the technique. So, I truly hope you all26:08 study in a way that leads you to26:09 success and ease. If you are in a STEM26:11 field, I absolutely recommend studying26:14 Python programming. Without exception,26:17 regardless of the field, Python is a26:20 must for all STEM areas; people in26:23 group chats get confused whenever a new26:26 model is released, wondering what it is26:28. It sometimes feels like chasing after26:31 girl groups. When you look at the flood26:35 of messages in group chats, it feels26:37 like people talking about which girl26:40 group released a new song, what the26:42 style is, or what dance they're doing.26:46 But if you chase after things that way,26:48 there is no end to it. Just feeling26:49 burdened by "technical debt" and26:51 struggling. Getting all stressed out.26:52 Thinking, "I feel like I need to know26:54 this," there is no need to be like that26:55. There's no need, just study. You just26:57 need to study the basics. A significant26:59 portion gets resolved easily. For27:01 example, many people suffer because of27:04 RAG, right? But the technologies used27:07 in RAG are actually just traditional27:10 search technologies with an LLM27:12 attached. For those who worked on27:16 traditional search, they had no27:17 interest in it until the LLM came out,27:20 and now they ask, "Why does the LLM27:22 talk such nonsense during searches?""27:25Why is the accuracy so low?" If you27:27 approach it like that, you start27:28 wondering, "Should I use this library27:30 or LangChain?""Should I use LangGraph27:32 or something else that just came out?"27:34—that’s when you get confused. Yes.27:36 That is why studying the basics is so27:37 important. You said it's not easy for a27:40 newcomer to get into such positions,27:42 and you need to judge that coldly right27:45 now. There are currently a lot of27:47 experienced workers. So, if there is no27:50 change, you have to get in line. If you27:53 want to enter backend development as a27:54 newcomer, you have to be better than27:56 those with experience. It’s not easy27:58 to be better than experienced backend27:59 developers. Because, for example, if28:01 it’s Java Spring, versions are28:03 tangled and complex, and you need to28:04 have that knowledge. To be able to do28:06 that better than someone with28:07 experience. That is impossible. Even if28:10 you're smart and good at algorithms, in28:13 fields where the know-how itself builds28:16 a barrier to entry, a newcomer cannot28:19 just break in like that. If both the28:24 experienced person and the newcomer are28:26 backend, would the experienced dev say,28:27 "I’ll do AI, so you take over the28:29 backend"? Would that happen? That’s28:32 not easy. It might be good in theory,28:34 but it doesn’t work that way. It is28:36 human nature to continue doing what28:38 they were already doing. Even within a28:40 company, if I want to do other work, I28:42 have to move to a different company.28:44 Moving companies is easier. Than28:45 staying within the same company. "I was28:46 originally a backend dev, but I’ll28:48 focus on AI now." If you tell me to28:49 hire a junior and put them in charge of28:50 the backend, do you think I'd hire them28:51? No, that’s not it; I want someone28:55 who can complement me, someone whose28:57 skill set can balance mine. Oh, if I29:01 feel like I can work with them, then29:02 I’ll hire them. Korean society is29:05 quite dynamic. The new generation wants29:08 to take on new roles. That’s why the29:11 government is pouring money into29:12 projects. They’re saying, "Hey, stop29:13 doing what you were doing and try this29:14 instead.""Do high-tech.""Do something29:16 difficult.""Do something that uses GPUs29:17.""Do AI.""Do something that can become29:20 a key export industry." They are29:22 driving us in that direction. They’re29:23 pushing us intentionally, but some29:24 think, "Oh, I don't think I can do that29:26.""I’ll just stick to what I was29:28 doing." Is that even possible? Again, I29:30 want to emphasize that this is only29:32 because of the current situation. In a29:34 normal situation, once you’re settled29:36 in a company, you’d hold your29:38 position, hire some juniors, train them29:39 to grow the team, and get promoted—29:41that’s how it’s supposed to be. Yes29:43. I think it would be wise for you all29:46 to cope with this smartly. It would be29:49 a good time to buckle down and study29:50 during the holidays. My lectures are29:53 well-structured for just browsing29:55 through the videos, so it would be good29:57 to watch them during the holidays, and29:59 I hope you spend your time in a way30:01 that helps your growth. Well, let's30:03 wrap up here for today, and I look30:05 forward to seeing you all again next30:06 time. Thank you.
5,269 words · 816 lines
![[선데이 라이브] 당신도 AI 시대에 살아남을 수 있다!](https://i.ytimg.com/vi/D95UC9U5ShA/hqdefault.jpg)
