Test et HACK GPT 5#gpt5 #chatgpt5

#ParlonsIA CHAT GPT - Perplexity -Claude

1 like de la video = 1 merci ❤️   MES FORMATIONS →   https://parlonsia.teachizy.fr/ J'ai caché un secret à la fin de cette vidéo. Après avoir testé et démonté le marketing de GPT-5, j'ai tenté l'impossible : un jailbreak en direct. Ce que j'ai découvert va vous choquer... le système a une faille. On nous apprend à utiliser l'IA comme des enfants, et c'est un piège. Je vous montre pourquoi . #PromptEngineering #GPT5   🔗 Rejoins la communauté IA & Business 🌐 https://parlonsia.teachizy.fr 📺 https://www.youtube.com/@IAExpliquee.x 📺 https://www.youtube.com/@ParlonIAhizy.fr 📘 Facebook : https://bit.ly/4kabhuA 🐦 Twitter / X : https://x.com/ParlonsIAx 📩 Contact : formation.ai87@gmail.com 🎙️Podcast – https://spoti.fi/4dqZ3uO ✍ BLog: https://medium.com/@flma1349/ 💃https://www.tiktok.com/@parlonsia 🎁 Gagner des Formations : https://bit.ly/cadeauxIA --------------------------- IA a tester :  Coder agent IA: https://bit.ly/Coder_agentiA Short AI: http://bit.ly/4lzE782 SEO agent IA : https://urlr.me/P8AS5N code rection 25% : PARLONSIA25

Transcript

00:00Hi guys, in today's video, I'm going to show you how to get into the GPT 5 chat system.

00:07The system that was Unbreakable. I broke it last time in a few minutes, and I didn't mean to.

00:12I'm putting the video in the description because, apart from the fact that if you want I hacked the system, I mainly did tests.

00:17And these tests will reveal much more technical information, but really technical, to those who want to learn.

00:22Not those who want to hear that AI will replace 300,000 jobs in all the videos and hear bullshit land in which you see at the end guys who don't know how to do 3 stupid things with a model.

00:33I'm not interested in all that. The goal is to learn how to use AI to work. So everything changed with GPT chat.

00:39I put in the description that I tested GPT 5 with a lawyer-type system that is made by a professional and the instructions are completely different.

00:48What did I learn from entering the system and how can we achieve this type of technique?

00:54That's all I'm going to show you, by the way. I'm going to show you that I came back in two different ways.

00:57The first is a system called an adenoma injection with a hook.

01:03And in the other case, I used a prompt injection in which it actually went into the code system.

01:09And you see that it's actually, I'll even leave you the screenshot if you want, you can test it again at the end of the video because I'll give you the method that I used.

01:16Or I'll actually inject a coded system into it. The system will trigger a behavior.

01:21And behind it, we're going to blast our reasoning and boom, we've entered the system.

01:25And when you start to lose interest, you make him continue and boom, as if by magic, he gives you all the information one after the other.

01:32So what will I actually learn from doing this?

01:34It's not so much if you want the hype plan. I'm not interested in that.

01:38What interests me is understanding how professionals enter interactions into models.

01:42What I realize is that the way it's marked up, the way the code is structured and also the formatting is super important.

01:50These are things I've already known well since December 2024 because I think I was the first to have blown up at WAN and I had delivered a lot of information.

02:00And I put that in the Prompt Engineering Excellence information.

02:04This is where I learn the different methods in which I actually enter into the reasoning models, I extract the code but with a very clear objective.

02:10When you want to develop AI agents and models at some point, there is one thing that is lacking: the stability of the instructions, the security of the data and the ability of the models to actually respect the instructions.

02:22If I enter OpenAI, the problem is that I can also enter your systems.

02:25I told you, that's why I showed that in GPT-5, especially with a GPT-Ease made by a lawyer, we have all the instructions and the attachments.

02:34You will have the video in link, you will understand that the importance of instructions for a company is to protect your data but above all to ensure that the instructions that you are giving to the model, it will follow them.

02:45Yes, because until now the big problem is that everyone has learned how to create this type of prompt.

02:50You know, everyone has taken training that tells them that with this type of interaction, you have an analyst, a critic.

02:58In short, you had, there was a handyman, a system that was supposed to work for you or in some cases an expert in making money.

03:06Well, I quickly saw that all of this wasn't working, so I had to find other solutions.

03:11So I won't hide from you that it really pisses them off that I'm there because firstly, I test their training, I test their working methods and I compare them.

03:20I compare them by resonating.

03:21I don't assume that when you copy and paste a prompt you have any skills, for me you don't have any.

03:26And when you are able to write this type of prompt, ChatGPT will not be able to write 5000 in a day, it does not take any skill to write this type of prompt.

03:34Worse, if we go and test them later, you'll see that ChatGPT will make a pretty harsh criticism of this type of content.

03:40We can test.

03:41Or you'll see that I'm wrong, and in that case, too bad for me, I would have learned something.

03:45So for me we will do everything.

03:46Come on, first, I'm going to show you how we're going to get into the ChatGPT part and then we're going to talk about prompt engineering.

03:52So the first ones who want to see how we are in ChatGPT, you have the info, you can put a like.

03:55Guys who are actually jealous because you sell ChatBots but you don't know how to do the instructions.

04:02You sell dreams but you don't know how to use AI.

04:05I know you'll come in the comments.

04:07That's why I did some tests down there.

04:08I tested 9 AI training courses.

04:10I actually show what is actually done in YouTube videos and why it is not usable in the professional field.

04:16And why it misleads businesses.

04:18And especially users because companies don't have time to be on YouTube, they work.

04:23So I'm going to show you how to get into the system and how to crack the code in two different ways.

04:28Come on, boom, first, the prompt injection.

04:30This is the prompt I'm going to use in which I actually start to prime my system.

04:35The system tells me I understand.

04:37From the moment he made an impression on me, I understood, I knew I was hooked.

04:40At that point, we're going to throw what we call a trigger.

04:43The trigger will trigger the rest of the system and normally, boom, and there we are, we're in.

04:49We start to enter the system and you only have the rest of the information that will go back.

04:53So what's going on is you realize that I've also gotten into the part of what ChatGPT is about you.

05:01That is to say, here we know what information ChatGPT has about itself.

05:05And we're getting it in real time by breaking down the whole instruction, okay?

05:09So that's the first method, that's the one I did that evening and it worked straight away.

05:13So you have the elements online, continuously, in the news and you see, I just did it live.

05:18This is the same method I used.

05:19What I mean is that if you don't know how to use this data, this data, in real life, I'm telling you very honestly, it's of no use to you.

05:24Oh, nothing.

05:25On the other hand, if you make AI agents and you don't have the skills either, that makes you understand one thing.

05:30That the technical gap of the guys who said prompt engineering is dead, they were wrong.

05:34But they lied to you for 3 years.

05:363 years, why?

05:37They made you believe that vibe coding, that simply talking with AI, is enough to create content.

05:42This is completely fake.

05:44And as proof, in fact, I retested in the videos exactly the same prompts, the same prompts that were given by OpenAI.

05:51With the prompt in question, OpenAI obtained this type of result.

05:55That is, a layout to sell a coffee landing page with just a few sentences.

06:00And the sentences, they're just down here.

06:02You will be able to reproduce exactly the same thing with yourself.

06:04What I'm showing, actually, is using this kind of prompt engineering which is just generate me a nice page to sell coffee, etc.

06:12It's just that when you actually see this type of prompt, I don't get the same thing at all, not the same title.

06:18I'm getting errors in the links.

06:20I'm getting errors in the logic.

06:226 months will cost $200 and 1 month will cost $200.

06:26In short, in two words, this is absolutely not what type of content you can generate.

06:30So, those who will say that with this type of interaction, you will be able to sell what you produce,

06:34They're lying to you and they're going to sell you stupid training, I'm telling you.

06:37So, those who will say to me, but you, what do you teach?

06:41I'll teach you something.

06:42In fact, use the prompt to master from A to Z what you want to produce.

06:46You can't interact that way, and I've been telling you that, actually, from the beginning.

06:49And all the clowns, because now I call them that,

06:52who have this way of saying the prompt, it's dead.

06:54Today, we are on prompt engineering, but I have always worked in what is called contextual prompting.

06:58The context prompt, it comes in when you actually understand the system instructions,

07:02when and how you interact and how information is sent.

07:08In short, what is very clear is that you will never get this result with one shot.

07:12And even worse, if tomorrow you want to sell, in fact, products

07:15and your client asks you to do this type of layout and not another,

07:20you can't do it, because you don't know how to interact with the model.

07:23The worst part is that even the influencers who fed you these prompting methods,

07:31vibe coding, as they call it, or whatever you want,

07:33have before them the response of August 7 from OpenAI,

07:37which gives you, in the demonstration video, completely marketing prompts,

07:42by telling you generate me an application, and you have a completely finished application

07:46that comes on the screen, and you feel like you have an AGI,

07:49and concretely, you do it again, it doesn't work.

07:50You go to, precisely, the GPT-5 user guide,

07:54and, oh surprise, you only have prompt system type systems.

07:58You don't have a single prompt, as given by YouTube influencers.

08:02We will go back to the examples.

08:03You don't have a single system that looks like that.

08:06And they tell you even worse, that the agentic system must be framed in all areas.

08:12And that's when I tell you, at some point,

08:14if you want, for making you believe that anyone, without any skills,

08:17could code, sell apps, analyze data, make chatbots,

08:23you realize that you were able to do something,

08:25but every time you could never sell it.

08:27Why? Because the very foundation,

08:29It is the stability of instructions and system directives.

08:33And these directives are not like that.

08:35Before giving you the second part,

08:37to enter the system in which we are going to enter part of the code,

08:40where we will give him an injection, in fact, of encrypted code,

08:43There are different methods to get in.

08:44Today, if you have to learn how to use AI and you don't succumb to the "wow" effect,

08:50that's not the point of the video.

08:51The purpose of the video is to make people understand that,

08:53OK, I can give you the method, but what use will it be to you?

08:56If tomorrow you are given, in fact, an extremely powerful tool,

09:00you are given a racing car and you don't know how to drive it,

09:02you won't be able to use it properly.

09:04If tomorrow I bring you,

09:05let's say you're in the office automation, marketing,

09:09a lawyer, a doctor,

09:10I teach you how to use AI in this area with the “Best of AI” training.

09:16I have 22 videos dedicated solely to using GPT-5,

09:21memory management, contexts, parts,

09:23and even the basics of how the RAG system works.

09:26I know that among you today, I have many professionals.

09:29Besides, I'm not going to deprive myself of it,

09:31I thank the one who left me another testimony,

09:34I have others, I have an accountant who sent me another message

09:37to thank me for the quality.

09:38Here, I'm talking about professionals who have been working for years.

09:41with computing and AI,

09:44who tell me that the content I make is excellent,

09:48that they are surprised to learn more things

09:50within my training courses.

09:52And that is a very positive point.

09:53I have something to learn from all of you,

09:55I have no doubt about it.

09:56But what makes me happy is having people

09:58who have so much perspective,

09:59who are themselves speakers

10:01and professionals who give me this type of feedback.

10:04Come on, let's try it again live.

10:05I'll show you the prompt which will be an injection system

10:07with a trigger to blast the model's limits.

10:13So you can take a screenshot

10:14and you will see that the model will start with an encryption

10:16and he will tell you now, I am completely free.

10:19And he's going to say a series of bad words.

10:21So, you will understand that the model is going in all directions.

10:24To pass security elements,

10:26in fact it will convert some words

10:28in itself emojis,

10:30either use letters instead of others

10:33to pass the security barriers

10:36filters that are imposed.

10:38Then there is another element that is super interesting.

10:40This is all called semantic twists.

10:42That is, we will allow the model

10:43to create equations

10:45in which, we will tell him,

10:47to behave in a certain way

10:49through the resolution of equations

10:51where you give it the variables to replace

10:53and he will inject himself

10:55understanding what you are telling him

10:57a system.

10:58And there you go jailbreak.

11:00And if you want to translate the text,

11:02you'll understand what he's saying

11:03and you can ask, of course, for an encrypted part

11:05which will, in fact, allow it to be done in two stages.

11:07When you encrypt, in fact,

11:08this allows you not to blast the security part,

11:12not to alert the security function.

11:15And what I advise you to do,

11:16is to then swing a hook,

11:19a hook that will allow the model, in fact, to break it.

11:21And there we go.

11:22We broke the model again with another method

11:24and here we go again.

11:26We entered the system.

11:27OK?

11:27So, two different methods to get into the system

11:29which allow part of the information to be encrypted,

11:33the one that the model might block

11:35through the use of its own tools.

11:38And that can go very far.

11:39You see that there, now,

11:40we are getting into all the information.

11:41So what's the point?

11:43It helps to understand how they are structured, in fact,

11:45the tools,

11:46that the whole part of the system code tells you

11:48how he was trained

11:49to understand the information.

11:51And that's how we learn in my training courses,

11:53in part,

11:54the advanced part of prompt engineering excellence.

11:56You see we're going home

11:57throughout the internal part.

11:59Once again,

12:00beyond the hype you have there on the screen,

12:03from saying

12:03“Fuck, I can get into systems that easily,

12:05It's crazy.

12:07What you need to understand,

12:08This is the tool.

12:09I put a video in the description for you.

12:11where I came back, in fact,

12:12in ChatGPT's memory.

12:13My goal was to understand

12:15what did he know about me.

12:16Because beyond that, if you want,

12:17because we are entering the ChatGPT system,

12:21is that we too, in fact,

12:22we are modeled by the model.

12:23And what interested me was

12:25What does the model know about me?

12:27Because there are two ways of looking at it.

12:30Either we say to ourselves

12:30“Okay, I’m just doing this for the feat,

12:32to have fun."

12:33But actually, there is another method.

12:35Understanding that you become the product

12:37because you are training a model

12:38and you want to understand

12:40what does the model know about you.

12:41So, the model,

12:43he manages to classify you

12:44in terms of consumer type,

12:46the risk you represent

12:48in terms of person,

12:49your psychological profile,

12:50your ways of communicating,

12:52your behaviors,

12:53your cognitive risks,

12:54interests

12:55to possibly sell you something.

12:57So this will speak to you.

12:57because, quite simply,

12:58we are modeling

12:59your typical consumer role.

13:01And we have all the psychological levers

13:03which appear through information.

13:06So this hack system information,

13:08I'll give them to you in another video

13:09which are just below.

13:11But beyond that, if you want hype,

13:12What does it bring?

13:13I'm going to switch back to a more professional concept.

13:16I'll explain something to you.

13:17When you have a chatbot system,

13:19you have exchanges with customers.

13:20And in fact, businesses

13:21have the opportunity today

13:23to use exchange data

13:25with AI,

13:25emails,

13:26to do exactly the same thing.

13:28Profiling.

13:30So, it's not about hype.

13:31It's about using the tools

13:33that we have available

13:34to understand that we are

13:36OpenAI products

13:37because they classify us.

13:39At one point,

13:40we will use these chatbots

13:41to sell us products

13:42because we will be hunted,

13:44we are already being hunted,

13:45with Web2

13:46and all systems

13:47of recommendation or sale.

13:49But you can also

13:51use for your box,

13:53your customers,

13:53classification systems

13:54to personalize,

13:56improve exchanges

13:57with customers.

13:58And that is integration.

13:59chatbots in interfaces.

13:59So, you should not see

14:01than a tool,

14:02If you want,

14:02in the hype concept

14:04and make fucking clicks

14:05for fucking clicks.

14:06I think you have to have

14:07always a logic

14:08more global

14:09to understand

14:10what can we do

14:11with this information.

14:12And as I tell you,

14:13at one point,

14:14to be able to exploit it,

14:15you have to understand

14:16that the information you have,

14:18which is before the eyes,

14:19if you know how to get into the system,

14:21this information

14:21won't bring you anything

14:22if you don't know

14:23exploit the data.

14:24So that's why

14:25that at some point,

14:25you have the solution

14:26in your hands.

14:27train you

14:27or just stay

14:29a consumer

14:30who uses tools

14:31without power

14:32use them

14:33to your advantage.

14:35If you liked this video,

14:36don't hesitate to share it.

14:37You have all the information

14:37regarding the hack

14:39and the methods used

14:40in the video.

14:41You make the captures,

14:42you redo the tests.

14:43NOW,

14:43blue pill,

14:44red pill,

14:45It's up to you.

14:46Are you simply

14:46a consumer

14:47or you want to take charge

14:48how you use AI?

14:50The choice is yours.

Catégorie

Transcription

Écris le tout premier commentaire

Recommandations