YT YouTube2Text

YouTube2Text → Transcripts → Yeni Bir Meslek Doğuyor...

Yeni Bir Meslek Doğuyor... — Transcript

by Can Deger · 2,816 words · 422 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Now, my dear friend, I am going to ask
0:02you a question. Maybe you are 18 years
0:04old today, but if you were 18 today and
0:07starting your tech career from scratch,
0:09you are just starting out. Which career
0:12path would you choose for yourself?
0:14When you want to be involved in
0:16technology and answer this question
0:18honestly, I wondered about something. I
0:21asked myself, for example, would I be a
0:24software developer, a cybersecurity
0:26expert, a cloud specialist, or
0:28something else? But I noticed something
0:31else here. Rather than just software,
0:34cybersecurity, or other topics, I
0:36discovered a more interesting cycle in
0:39the tech world. Though others have
0:41certainly mentioned this before me, I
0:43would say it became clearer to me. That
0:46is a more accurate way to put it. And
0:49at the end of this, I realized that
0:51another career path, a profession,
0:53perhaps one that will leave its mark on
0:56the coming years, could emerge. I have
1:00mapped out a roadmap in my head for how
1:01you can do this. If you are interested,
1:04I wanted to show you what you should
1:06look for and how this cycle repeats
1:08itself. I will explain what this cycle
1:11is, what this profession is, and how
1:13you can reach this profession shortly.
1:15If you are ready, let's begin, my dear.
1:18Dear friends, by the way, before we
1:19continue with the video, I want to talk
1:21to you about a small solution. As you
1:23know, the AI field is moving incredibly
1:24fast. ChatGPT is releasing a new model.
1:26The next week it's Claude, then Gemini,
1:28then something else. After a point, it
1:29becomes difficult to decide which task
1:30to do with which one. It becomes hard
1:32to choose. It becomes hard to make a
1:33decision. That is why I have been using
1:35TLM Abacus lately. It is very useful
1:37for me. I want to talk about it.
1:39Because you don't have to deal with
1:40opening single applications. GPT-4o,
1:41Claude, Gemini, Groq, DeepSeek R1, they
1:43are all in the same place. When a new
1:45model comes out, you don't wait too
1:46long. It is added quickly. I do
1:47research in one place. I create a draft
1:48in another place. Sometimes I prepare a
1:50presentation. I generate images. I
1:51create videos. I even give it an idea I
1:53have and tell it to turn this into a
1:54working website. It even does that. The
1:56AI agents within it can also complete
1:58time-consuming tasks on their own. They
2:00also have a feature I really like.
2:01RouteLLM. I don't have to struggle with
2:03which model to choose. It chooses the
2:04most appropriate model itself based on
2:05my request. If I want to, of course, I
2:07can manually switch to the model I want
2:08. The most logical part is the price.
2:10Normally, we need to get separate
2:11subscriptions for most of these models.
2:12Here, they are all on a single platform
2:14and it is only 10 dollars a month. If
2:15you, like me, constantly switch between
2:17different tools during the day, I
2:19definitely recommend you try TLM Abacus
2:21. Take a look, I hope you like it. Take
2:23care of yourselves. I am returning to
2:24the video. Let's go. Now, when we look
2:27at the last 30 years, I want to take
2:30you back to 1996, 1995. Just think
2:33about it, friends. The internet
2:34revolution took place. The internet
2:36came to Turkey in '93 if I remember
2:37correctly. It arrived in '92 or '93,
2:39something like that. After that,
2:40everyone started creating pages on the
2:42internet by slowly getting
2:43www.something. ...addressing.tr domains
2:49. They moved their own websites to the
2:52internet and did various things. And
2:54then, attacks against these websites
2:57started to occur. The disruption of
3:01this interconnected internet happened,
3:03and once a problem arose, it naturally
3:06brought forth a concept, a solution.
3:09These developments. What happened? The
3:12concept of network security emerged.
3:13Network security experts came along,
3:15firewalls were introduced, DNS
3:16protections arrived, and so on. This
3:17and that happened. Those things arrived
3:19. Then, between 2006 and 2008, the
3:23concept of Web 2.0 emerged. After that,
3:27the writing of many web applications,
3:30the deep integration of mobile phones
3:32into our lives, and the emergence of
3:35mobile app concepts created the
3:37application security problem, and the
3:39profession of application security
3:41expert was born. Then cloud emerged in
3:45the 2010s. Actually, it emerged after
3:472006, but its popularization and
3:49widespread use happened after 2010. The
3:51cloud concept, cloud security, this
3:54cloud idea, everyone moving to the
3:56cloud, shifting to cloud solutions on
3:58Azure, AWS, and GCP. Then, remember the
4:01secrets, keys, various things, and
4:03forgotten S3 buckets. Suddenly, the
4:06concept of cloud security emerged. And
4:08then, what did you see happened?
4:11Kubernetes emerged, DevOps emerged, and
4:14so on. Our way of doing business began
4:16to change. Then, what did we see, my
4:18friend? We started to see the DevOps
4:21concept, and these Kubernetes and
4:22container security concepts in the
4:24environment. What is the common
4:26denominator here, my friend? When you
4:27look at it, what do you see in this
4:29picture? What do you see within these
4:3130 years from back in the 95s until
4:33today? A technology emerges. The
4:35technology is accepted. It comes into
4:37general widespread use. Then, attacks
4:38related to this emerge. For instance,
4:40network security, good friends appeared
4:42. There were DNS poisonings, domain
4:44hijackings, etc., at the end of the 90s
4:46, that's how systems were hacked. And
4:48then, in web applications, XSS appeared
4:50, SQL injections appeared, and things
4:53like that came out. Then in the cloud,
4:55IAM appeared, zero trust appeared, and
4:57those kinds of things arrived. Then
4:58with DevSecOps, pipeline security
5:00arrived, deployment security came along
5:02, and supply chain attacks happened,
5:04etc. Things like this occurred. What
5:05happened, my friend? Technology grew.
5:07Technology became widespread. Attacks
5:09related to the technology emerged, and
5:10the technology gave birth to a
5:12profession regarding its security. It
5:14birthed a profession. Right? This
5:17repeated itself in a loop. When you
5:20look at it like the internet of 30
5:22years ago today, what is the growing
5:24technology, my friend? Artificial
5:26intelligence, AI, right? Then this
5:29concept of AI security will emerge. If
5:33we say this will be the profession of
5:35the future, would we be very wrong? I
5:37think the picture we see when we look
5:39at this process tells us exactly that.
5:41But there is an important thing I want
5:43to explain here. The misconception
5:45everyone falls into is this. If you set
5:47out saying, "I will be an AI security
5:50expert," crash and burn immediately.
5:52You would fail right away. So, my
5:54friend, listen very carefully to what I
5:55am about to say. Because from now on,
5:57if you choose this career and want to
6:00progress along this path, I will give
6:02you the keys to some very important
6:04things. First and most importantly,
6:07buddy, the concept of AI security is
6:09not prompt injection or indirect prompt
6:11injection. Okay? These are just the tip
6:14of the iceberg in AI. There is a much
6:15larger situation at hand. You shouldn't
6:18evaluate the concept of AI security so
6:19simplistically. Today, agents have
6:21emerged alongside AI. There are these
6:24agents performing tasks on your
6:25computer on their own, and there is MCP
6:27. There are the learning patterns of
6:29the LLMs running in the background.
6:30There is the poisoning of those
6:32learning patterns. There is them
6:33acquiring and sharing misinformation.
6:35There is the creation of "slop."
6:37There’s endless stuff. There is them
6:38acting according to their own logic.
6:39Companies are now internalizing AI.
6:42They are embedding AI into their
6:44companies. Well, brother, if the AI
6:46within this company sends financial
6:47reports to the wrong person, for
6:48instance, it will start running the
6:50bank. If you make companies send money
6:53from bank accounts to the wrong person,
6:54or if you cause them to do other things
6:56from the outside and be affected by
6:58them, look, I’m talking about a much
7:00bigger picture, a much greater
7:02capability. Do you get what I mean? I
7:04mean, when you look at it in depth,
7:05many points and components need to come
7:07together and act in relation to each
7:09other. What does this require, brother?
7:11Learning many different and varied
7:14disciplines. So, rather than AI
7:18Security being a simple heading or a
7:20simple discipline on its own, can we
7:22say it’s a compilation of everything
7:24we’ve learned in the last 30 years—
7:27network security, cloud security, web
7:29application security, and so on—as a
7:31discipline where we learn a known and
7:34applicable version of all of them? I
7:38think we can. I think if a concept like
7:41AI Security, AI security engineering,
7:43is emerging today and will influence
7:46the next 10 years—which I find very
7:48likely—Then it needs to be evaluated
7:50on a very wide scale. But you might say
7:54, "Brother, do I really have to learn
7:56all of this, 30 years of experience,
7:57these highly capable tools, and whatnot
7:59?" Isn't that a bit difficult? "I'll
8:02just take a course, I'll read.""Let me
8:05just get a certificate or something.""
8:07Can I just become an AI security expert
8:09like that, brother?" I don't think so.
8:12Because, for example, if I were to
8:14become an AI security expert today, I
8:16wouldn't go chasing after learning it
8:18directly, brother. Whoa, here comes the
8:21roadmap. Look, I’m preparing you,
8:23buddy, don't forget this favor. It's
8:26not a standalone field; as I said,
8:28it’s a 30-year multidisciplinary
8:29structure. If I were starting from
8:31scratch, I definitely wouldn't go
8:32chasing certifications. Once we’ve
8:34taken that in and put it in our pocket,
8:36I would go after understanding the
8:37system, brother. Instead of hoarding
8:39training videos, I would learn how an
8:41LLM actually works. What is a
8:42Transformer? What is a context window?
8:44What is embedding? What is the purpose
8:46of function calling? Why did RAG emerge
8:48? Why is MCP being talked about so much
8:50? Look, I would ask myself these
8:51questions. Because that’s where the
8:53action happens at this layer. Okay? I
8:54would understand these first. Because
8:56that’s where the attacks will be.
8:57Because this is the entrance gate. But
8:58I need to know what's going on behind
8:59the scenes. I need to know the cloud,
9:01man. I need to know AWS, Azure, Google
9:03Cloud. I mean, something running
9:05locally on my notebook is no use to me.
9:07Then threat modeling. Man, I've been
9:09saying it for years; if you can't
9:11predict or imagine how a system will be
9:12attacked, you can't protect that system
9:14. That's why threat modeling is
9:16definitely one of the most important
9:17things. And finally, there's practice.
9:19I mean, you can't do this job without
9:21breaking things. Just watching training
9:23videos and stuff won't work, man. You
9:25have to break things. You have to
9:26really smash it up. So, let's say, what
9:28would I do if I were in your shoes? How
9:31would I do this job, man? I'd install
9:32Docker on my laptop. I'd put Ollama on
9:34top of it. I'd put WebUI on top of that
9:36. I'd make a simple RAG. All right?
9:38Then I'd try to break it. I'd try
9:40prompt injection. And, well, let's say
9:42I'd try prompt fuzzing. I'd look at it
9:44with Garak. Look, take notes on these.
9:46Pause the video. Look, search for all
9:47of these on the internet. Yeah, we're
9:49not just talking nonsense here. Then
9:50there's that thing, like, there used to
9:52be Damn Vulnerable Web Application.
9:54There's Damn Vulnerable LLMs now, for
9:56example. I'd install that; if I
9:58couldn't do anything else, I'd try to
10:00break that. Oh, there's also that.
10:01PortSwigger has AI labs. LLM labs and
10:04AI labs. Man, I'd dive right into those
10:06, for instance. And I would put all
10:07these studies of mine on GitHub. I
10:09would fill up the GitHub. Because when
10:11I go to a job interview, you know,
10:12GitHub is very important. They even try
10:13to lynch me saying, "Man, look at
10:15Erwin's GitHub, there's nothing written
10:17there," trying to hit me from there, my
10:18poor brothers. That's not a problem. I
10:20don't have a need for it, my dear. You
10:23have the need. When I'm hiring someone,
10:25I look at their GitHub. I look at their
10:27certificates too, but I definitely look
10:29at their GitHub. Why don't I do it? I
10:31don't know, no one asked me, but they
10:33might ask you. There's a possibility
10:35they'll ask. Why leave an open door?
10:37Right? That's why I'm saying it.
10:38Otherwise, this isn't something I'm
10:40telling you to take as an example.
10:41Could I explain it? Why don't I do it?
10:42No one has asked me until today, man.
10:44If they ask, I'll put it up too. If
10:46they don't ask, let it stay like this
10:47for me. Yeah, you can say that too, huh
10:49. Don't put this. No one asked me.
10:50Who's going to look at it? There are
10:51already tons of weird job postings in
10:52Turkey. None of them are real job
10:54postings. We're already applying to
10:55jobs through postings. No one even
10:56cares about us. You're right about that
10:57too. I can't say anything, but, I mean,
10:59what would I do? We're 18 years old,
11:00right? That's our topic, right? That's
11:01the scenario, right? That's why I'm
11:03explaining it accordingly. Now take
11:04what you will from it. I don't get
11:05involved in the rest, Can. That's why I
11:07think a profile is still important. Or
11:09a blog post, or, let's say, build a
11:11community. I mean, create a simple
11:13community. Go to your university or
11:14somewhere else and talk about the
11:16project you did. Give a 10-minute
11:17Lightning Talk. I mean, record a video
11:19of it and put it on the internet. Like,
11:21this guy is active, he’s done some
11:23things, etc. Instead of just watching
11:24courses, for instance, what was that
11:26guy's name? Andrej Karpathy, man, take
11:29a look at what this guy writes. Look at
11:31his articles, look at what he explains.
11:33The man is the king of this field.
11:34Master Karpathy explains very logical
11:36things. Alright? Fantastic. Put that in
11:39your pocket for starters. Check out
11:41Hugging Face. Hugging Face, for example
11:43, is an important place. You must look
11:45at those LLMs. It’s not just OWASP
11:47LLM anymore. It’s not just the
11:48standard one anymore. When we talk
11:49about security, again, there's the
11:51OWASP Prompt and Garak. Let me mention
11:54those again. Microsoft has a thing
11:57called PyRIT. It’s spelled P-Y-R-I-T.
12:00Then Rize, Istanbul, Turkey. Just look
12:03it up like that, uh. PyRIT is a great
12:07project related to AI security testing
12:10and red teaming. Take a look at it.
12:12It’s a nice framework. I like it, for
12:14instance. Take a look at that. Besides
12:15that, PortSwigger's AI labs, as I
12:17mentioned before, are definitely a
12:19place you should check out. You can
12:21study those. TryHackMe and Hack The Box
12:24have labs for this. Try them out, get
12:26some experience. Honestly, AI Security
12:29really seems like an important
12:30profession for the future, my friend.
12:33I’m not just pulling this out of thin
12:34air. I mean, looking at the last 30
12:36years, everything I’ve seen is that
12:38technology develops, then spreads.
12:40After it spreads, a problem statement
12:42emerges, and then the relevant security
12:44discipline appears to solve it. And
12:47these people have always earned good
12:50money throughout their lives. So, there
12:53will be money in this. So, it feels
12:56like this field will have a future. No
12:58matter what, there was no "internet
13:00security" career in '95. There was no "
13:02Cloud Security" in 2005. There was no "
13:04Kubernetes Security" in 2015. And today
13:05, AI Security is at that same point.
13:08The important thing is, maybe 10 years
13:10from now, we will still call AI
13:12Security the most current, the best
13:15profession, and when we say that, we
13:17will need people to secure it as the
13:19technology evolves. At this point,
13:23asking yourself not "how can I learn AI
13:25security," but "how early am I in this,
13:28where do I stand?" is perhaps the best
13:31move you can make. I mean, I want to
13:32draw your attention to this. Because
13:34sometimes in life, being in the right
13:36place at the right time, and perhaps
13:38having watched this video, could be a
13:40very important turning point. Now, I
13:42have a message for you in the future,
13:44my friend. After watching this, if you
13:46change your career or decide to pivot
13:48here and end up becoming successful,
13:50come back to this video and leave me a
13:53message from the future. Let’s see
13:55what happened. Will it work out? I
13:56might be completely wrong, but it feels
13:58like this field is going to grow like
14:00this. What do you think? I’m waiting
14:02in the comments. Let’s go then. I
14:03love you all very much. May God protect
14:05you. Can is out. See you. My YouTube
14:07family. Subscribe, guys!

About this transcript

This page contains the full transcript of Yeni Bir Meslek Doğuyor... by Can Deger, generated from the public captions YouTube serves with the video. The transcript has 2,816 words across 422 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.