- il y a 5 semaines
- #chatgpt
- #opeani
- #samaltman
1 like de la video = 1 merci ❤️ MES FORMATIONS → https://parlonsia.teachizy.fr/
Sam Altman nous a prévenus : ChatGPT est une bombe à retardement. Je vous montre le hack pour entrer dans sa mémoire et révéler tout ce que l'IA sait sur vous. Une vérité choquante sur les risques pour votre vie pro et perso. #chatgpt #opeani #samaltman #chatgpt 5 ##gpt
🔗 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
Sam Altman nous a prévenus : ChatGPT est une bombe à retardement. Je vous montre le hack pour entrer dans sa mémoire et révéler tout ce que l'IA sait sur vous. Une vérité choquante sur les risques pour votre vie pro et perso. #chatgpt #opeani #samaltman #chatgpt 5 ##gpt
🔗 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
Catégorie
🤖
TechnologieTranscription
00:00Hello, corticosteroids, today we are going to launch a nuclear bomb because here we have no filter, I tell you clearly.
00:06Today we're going to talk about Anis Ayari with the Defend Intelligence channel.
00:11I'll also mention a few other channels in passing that I'll quickly show you on machine learning and AI news.
00:17And it will be the opportunity to take a tour of Casper's, Monsieur Phi's, Mamie Twink's or even AmiStory's.
00:24What do all these chains have in common? You might ask yourself the question.
00:28It's the ability to talk about just about any topic without being censored.
00:32And in this specific case I was censored at the request of Anis Ayari.
00:36So who is Mr. Anis Ayari? Which channel does he support?
00:39So I made a video where I tested live the same prompts that were used by OpenAI, the same prompts.
00:47Where I demonstrated that OpenAI had given us a great marketing course in which in fact the prompts do not work and that the GPT-5 model did not work.
00:54Within two hours of OpenAI doing this test, I came out with my own independent test and incredibly I was one of the only ones going in that direction.
01:03The only one at that time on the web and practically in France, absolutely certain.
01:08It could have been a media sensation if the networks were not trying to suppress the truth, which is what is happening.
01:15That is to say, we are completely driven into a purely marketing structure by hundreds of thousands of influencers.
01:22And by the way there is a video that is played on repeat.
01:25But it's not the only one, we'll come back to that.
01:27The video in question was the GPT-5 video from Defend Intelligence's channel, which kept popping up in my feed as if I absolutely had to watch it.
01:37And it's done, I watched it but it's not the only one.
01:39I have to tell you that Monsieur Phi's video must have appeared at least 30 times in my feed.
01:45At one point I thought to myself, well, they want me to watch her, there's no choice.
01:48And so the video of expert Luc Julia came across to me.
01:51Ah oh Luc Julia, and yes the founder of the series who appears on all the channels about whom we have a lot to say.
01:58But the question that really got me was how come this person managed to remove my content in which I showed very clearly.
02:06We are telling you that artificial intelligence, based on completely fragmented instructions - when I say fragmented, I mean the structure of the accounts, which resembles nothing, nothing - allowed us to obtain results.
02:20And in common critical sense, because the goal here is not to attack people, it is to show what does not work on content that is extremely viral.
02:28The problem with viral content today, and we'll get to that, is whether we can trust these platforms that throw pages and pages of content at us.
02:38by practically forcing us to see them and by hiding other content which in my opinion is much more relevant.
02:44And analyzing yourself if you want content and saying boom, we're making you skip the video, it can simply appear to be censorship.
02:52So what is the goal of this censorship?
02:54So we're going to do exactly the same tests with the prompts, I'm going to show you what happens when using this type of prompt.
02:59And in the video, we will check if there are the 9 minutes that he claims being a copy of his content.
03:05I discover that in fact, he is already invited to talk shows, programs such as Le Crayon,
03:10that this is not the first time that we will find interviews with him, where he is presented as a pillar of the world of machine learning.
03:17Afterwards, I don't know, I didn't really find anything substantial.
03:21That's when I discovered that Annie Sayari is an AI engineer and I warned her on her page, I did some tests,
03:28Here is the link to my video, go take a look, I don't have the same results as you.
03:32He responds by saying roughly, you're promoting my page.
03:34Honestly, when I get into what we would call artificial intelligence and machine learning,
03:40I don't watch content whose objective is to code AI to create dating sites,
03:46finally to chat on dating sites, to cheat in a poker tournament with AI.
03:50You see, that's not the kind of subject that interests me.
03:53That's fine, I think, for an audience that's 15 years old, that's 13 years old,
03:56where we tell them, hey, it's great to see an AI playing Monopoly,
03:58or even see AI on Tinder.
04:00Yes, why not?
04:02What I like, in terms of artificial intelligence and content,
04:05and I invite you to subscribe if machine learning interests you,
04:07for example, it is content from Louis-François Bouchard,
04:10a Canadian who, in France, and that is a purely French problem,
04:14Look, in France, what's the matter?
04:15He has 5000 subscribers.
04:17You put it in the US version, it soon has 70,000 subscribers.
04:20So, yes, we can be interested in machine learning,
04:24make content, but the problem we really have in France,
04:27It's the level, the overall level and the interest we have in these contents.
04:32Do you want to train?
04:33I have done training for people who have no skills,
04:36nor in code, but who want to use artificial intelligence
04:39effectively in the professional world.
04:41Not everyone is asked to become a developer.
04:43That's why there are channels that also do more news analysis,
04:47It's one of the channels that I follow and appreciate.
04:49And you see, we are not on channels that have 200,000 subscribers,
04:53which get 200,000 views per video.
04:55Indeed, we are targeting niche targets.
04:58The question is, do we have the right to analyze their content?
05:02This is a bit of a question that arises today,
05:04because we're supposed to take these people at their word.
05:06And the worst part is that their videos are being highlighted,
05:10because that brings in advertising.
05:12And advertising makes money.
05:14So, I don't know the level of independence of these people with regard to the products,
05:17although they may say they have no interest.
05:20They still had, he had, GPT-5 cat three weeks in advance.
05:24Does a guy who has been doing email for ten years,
05:26Is he able to write prompts like this?
05:29When I saw that, I thought it was a joke at first, in terms of prompts.
05:34I said, that's not possible.
05:35Generates a 3-2 game where the guy walks down the street,
05:38whose goal is to headbutt.
05:40These are benchmarks.
05:41These are presented as benchmarks.
05:43Another benchmark, make a Donkey Kong platform game
05:46where we have a donut and we have to avoid donuts.
05:48Brief.
05:49So, I think that today, we need to know how to use it.
05:51because we have a job market that has changed.
05:53So what did I do?
05:55I used the same prompts in my demo,
05:57I ran the code and, oh surprise, it doesn't work.
06:01For what ?
06:01But I explained it in the video.
06:03I spent my time bringing you technical advice
06:05to understand prompt engineering.
06:06And what do they tell me?
06:08They tell me, ah, you have completed my content.
06:11Attention, attention.
06:11And the craziest thing is, look what I received.
06:14from the platform.
06:16Like, wait, and I publish content on machine learning
06:19and watch content related to adult content
06:24on which we are doing tutorials on AI.
06:27So, we tested the prompts.
06:29So here they are, the prompts.
06:30It doesn't work.
06:31Let's play the headbutt game again.
06:33It doesn't work.
06:33The website, landing page for water sports specialists,
06:37it doesn't work.
06:38Ah, this one worked.
06:39The landing page.
06:40And what did I explain?
06:42Having a page that fits into a web page
06:44does not allow you to sell a web page today.
06:48A web page is a server storage,
06:51an update, security, SEO,
06:53the back-end, security, capchas,
06:56all the connections that need to be made,
06:57all personalized content.
06:59In short, the concept of I make a sentence at random
07:03written in a language roughly
07:04does not correspond in any way to a product that can be sold.
07:07And telling you this, if you will,
07:09that's enough to alert, to say.
07:11Instead of saying,
07:12ah, maybe I should have made prompts
07:14on par with a guy who does machine learning.
07:16No, no, no, no.
07:16We'll sweep this under the rug.
07:18We will request the removal of the video,
07:20but based on what?
07:21Justify me during an exchange
07:24that these prompts are functional,
07:26that we can use it in the professional world.
07:27Okay, fine, are they machine learning specialists?
07:30Show us some machine learning content
07:32on which we learn something.
07:34What interests me,
07:34What can I do with my business?
07:36And I think there are a lot of entrepreneurs
07:37who would adopt AI
07:38if we could show them how we work with AI
07:41rather than telling them every four mornings
07:43We're going to steal your data with AI.
07:45Or, I made a drone
07:46who can hunt humans.
07:47You understand what that means.
07:48Normally these are things
07:49which shouldn't even appear.
07:51I remember a guy from OpenAI
07:53who had loaded a system
07:56in Python in GPT chat
07:58so he can shoot people.
08:00He was fired from OpenAI.
08:01But those who wanted to work on GPT-5
08:03and who needed to understand
08:04how GPT-5 worked
08:06and that there was a very big problem
08:08product launch,
08:09I brought them a crucial point
08:11that the entire part of the code
08:13inside GPT-5,
08:15especially on the prompt,
08:16and that was the whole basis anyway
08:17of the analysis,
08:19was brought into the game
08:20official documentation
08:22OpenAI Cookbook
08:23which speaks precisely about methods
08:26of GPT-5 prompting.
08:27So, the prompt,
08:28differently than what is said
08:30the video in question,
08:31has absolutely no use
08:33of classical words and structures,
08:35but of well-defined structures.
08:37They are explained by OpenAI.
08:38And these structures,
08:39I identified them
08:41since December 2024.
08:42Because I came back
08:43in systems,
08:45I've seen dozens of them
08:47and I modeled them,
08:48I put them in training
08:49for the attention of professionals
08:51or people like lawyers,
08:52doctors who need
08:53today of a level
08:54high reliability
08:55and not to ask
08:56to an AI code me a game
08:58of a guy who gives
08:58headbutts
08:59to people on the street.
09:01I'm sorry about that,
09:02This is not prompt engineering.
09:04That, for me,
09:05these are titles
09:05to do rank view
09:07on social networks,
09:09on a young audience.
09:10Well, that's my point of view.
09:11The structure of my training courses
09:13are entirely based
09:14on this structure.
09:15And here you have an example
09:16of code that is given
09:17by OpenAI which explains to you
09:18than to code,
09:20you have a structure
09:21where you have to define
09:21all parameters,
09:23prototyping,
09:25the outline,
09:26the guideline,
09:26spacing,
09:27the padding,
09:27the overstate,
09:28the entire structure of the framework
09:29with the elements
09:30that you want to use,
09:31the type of code
09:32and the entirety
09:33of the content of your folders.
09:35So.
09:35So if you find
09:36that this type of prompt,
09:37it's simple stuff,
09:38It's basic stuff,
09:38if you want,
09:39we can discuss it again.
09:40But it is not
09:41what was shown to us
09:42in the video.
09:43And what is a stain,
09:44This is when I will be justified
09:45and take me down
09:45in my training,
09:46you've never bought one
09:47of my training.
09:48But I assure you
09:48that you would learn things
09:49like me,
09:49I would learn things
09:50of you in machine learning.
09:51It's obvious.
09:52But on the other hand,
09:53your video on your prompts
09:54is not robust.
09:55So.
09:55Quite simply.
09:56Because we will come back,
09:57we will analyze
09:58is it that for 9 minutes,
09:59we took a copy
10:00identical to your content
10:02that we threw away
10:02to make views.
10:04This is not the case.
10:05Afterwards, surprisingly,
10:06as I said,
10:06is it inside,
10:07there was content
10:08adult character?
10:09No, there weren't any.
10:10So, it's still
10:11very embarrassing to have
10:12this kind of thing
10:12who arrives at the same time.
10:13So, what will interest me
10:14NOW,
10:14it's to really see
10:15if he could have asked
10:17that the content jumps.
10:20So what you see
10:20there now,
10:21it's simply
10:22the screenshot
10:22of a page.
10:23And I assure you,
10:24we have the right to finish
10:24from a screenshot
10:25pages.
10:26And it began
10:27around 23 seconds.
10:28So here, 26,
10:29it starts by indicating
10:30we are able today
10:31to ask GPT-5
10:32with normal sentences
10:33forget prompt engineering.
10:35And I demonstrated to you
10:36that it is false.
10:40It's never been so technical.
10:42For me, it's been since December.
10:432024 that I try
10:44to tell you guys
10:44and that it simplifies as much as possible
10:46and simplify,
10:47This is not the right solution.
10:49And that when we move forward
10:50in the video,
10:51I only talk
10:52all along
10:53on content analysis
10:55on which I indicate
10:57that we will test the prompts
10:58and that I do
10:59from prompt analysis
11:01and that I say
11:02that we are in the process
11:02to make fair use,
11:03that we identify
11:04the account and the content.
11:06And all along,
11:07I am doing
11:08examples
11:09for pedagogy,
11:10to explain
11:10how we should
11:11practice analysis.
11:13At 6 minutes 57,
11:15we are in the process
11:15to take the prompts
11:16and we test them
11:17and we see
11:17what doesn't work
11:18on the sequence of codes.
11:20And it's there
11:21that we try to code
11:22and there is nothing
11:23who walks.
11:23And I say
11:23these methods
11:25are not applicable,
11:26you have to use
11:27other methods
11:28and information
11:29that is on the page
11:30from OpenAI
11:30will guide you
11:31to allow you
11:32to find
11:32best practices.
11:34There you go, that's it
11:34which was given
11:35and that's it
11:35which was censored.
11:37So, if we take
11:38the legislative part,
11:39Article 122-5 of the CP
11:41requires that the name
11:42from the author
11:42and the source
11:42be clearly indicated,
11:43it's done.
11:44It was the case,
11:45so it was clearly stated.
11:46That it can be used
11:48in the frame
11:48justified, critical,
11:49controversy,
11:50educational, scientific
11:51and information about the work
11:52in which it is incorporated,
11:53it was the case.
11:54We were in the process
11:54to do analysis
11:55educational, scientific
11:56and information,
11:57so all the boxes
11:57are checked.
11:59And there,
11:59we're going to loop back
12:00on a set
12:02of contents
12:02to show you
12:03actually what's happening.
12:04Why people
12:06who have, we will say,
12:07200,000 followers,
12:08because it is not
12:09the only influencer
12:09who has an approach
12:12extremely marketing
12:13general public?
12:14But how is it done?
12:15that in fact,
12:15chains
12:16like Mr. Phi
12:17to 375,000 subscribers
12:18can, themselves,
12:20do exactly
12:21what I did
12:21without any problem?
12:22And above all,
12:23when the content
12:24in the end indicates
12:26that Luc Julia,
12:27a public figure,
12:28so by attacking
12:30directly a person,
12:31not a content,
12:32not a statement,
12:33my title was not
12:34Anis Yanari
12:36This is complete nonsense
12:37in its prompt engineering,
12:38No,
12:39it was GPT-5
12:40why everyone
12:40makes a mistake in using it?
12:42It was pedagogy,
12:43it wasn't the same thing.
12:44There,
12:45we have a direct attack
12:46of people
12:46to make a high ranking,
12:48many views,
12:48203,000 views in 9 days,
12:50This indicates a bit of the impact
12:51that it can have
12:52also on these reasons
12:53professional.
12:54But let's be interested
12:55to the content.
12:56What will interest me
12:57in the content,
12:58is that in the end,
12:59we have exactly
13:00the same things.
13:01Content excerpts
13:02in which we have
13:03interventions,
13:04content excerpts
13:05on which we have
13:07the name at the very bottom
13:08left
13:08of the page in question
13:10or the page logo.
13:12SO,
13:12it was exactly
13:13the same
13:13that I did.
13:14When there are interventions,
13:16there is the source,
13:17there is information
13:17and then,
13:18we discuss it
13:19and you see that
13:20all along,
13:21these are sequences
13:22on which
13:22Next,
13:23we will debate.
13:24SO,
13:24why do they have the right
13:25to be able to do it
13:26and why we,
13:27we don't have the right?
13:28Yes,
13:28because it is not
13:29an isolated case.
13:30Me,
13:30I looked
13:30from time to time
13:31videos
13:32who were still
13:32put forward a lot
13:33by Amistory
13:34and in its content,
13:36we're going to look at it.
13:36So these are
13:37societal surveys
13:39who do more
13:40of a million views
13:41and there is one
13:42Moreover,
13:42it's TikTok
13:43or even
13:44the lie
13:45of entrepreneurship
13:46social networks
13:46where we show
13:47that we have influencers
13:49who promise you
13:50to earn a lot of money
13:51without any skills
13:52and know that,
13:53it's part of the stuff
13:53that I immediately
13:55identified
13:56like content
13:57dangerous
13:57for people.
13:58I think the platforms
13:59have a huge role
14:01in the promotion
14:02of these contents
14:02and in fact
14:03that people
14:04were manipulated
14:06by offers
14:06promotional
14:07and there are some besides
14:08a video about Casper
14:10and Casper,
14:11167,000 subscribers
14:12has 90,000 views
14:13on the video.
14:14How is it made up?
14:15the video?
14:15Look,
14:16these are elements
14:17on which
14:17we don't even have
14:19The source
14:19getting started videos,
14:20we don't even have
14:21identification
14:22but its content
14:22is not skipped.
14:23Do we have
14:24the rights
14:24quotes
14:25people ?
14:25Do we have
14:26identification
14:26content?
14:27No,
14:27we have a video
14:28which starts
14:29with content
14:30which was used
14:31and there,
14:31It's no problem.
14:32What's the difference?
14:33How is it done?
14:33that they have the right
14:34to make content
14:35on which they do
14:36of the analysis
14:37of influencers?
14:38It's what ?
14:38This is the number
14:38of subscribers
14:39who makes the difference?
14:40Actually,
14:40we are not equal
14:41before the law.
14:41That's the truth.
14:42The platform
14:43does not allow
14:43to be equal
14:44before the law.
14:44SO,
14:45it's very good
14:45that we can say
14:46that there is danger
14:47with these promotional offers
14:49but why are they
14:49the only ones
14:50to be able to do it?
14:51That's the question.
14:52With the same methods
14:53and sometimes without using
14:54the sources
14:55and the codes
14:56which are normally
14:57mandatory.
14:58And how is it done?
14:58that there are YouTubers
14:59who are protected?
15:00That is to say that we cannot
15:00not to talk about them.
15:01How is that done?
15:03Because you see
15:03when we're going to talk
15:04for example, certain influencers
15:05that we see all the time
15:06and that we see all the time
15:07and that we see all the time
15:07and that there are only
15:07if you talk about them
15:09surprisingly
15:09your account,
15:10your content is skipped.
15:11Good,
15:12here it's more difficult
15:13she has 340,000 subscribers
15:14so she won't be affected.
15:15But I assure you
15:16I have another one from a YouTuber
15:17who plays the victim
15:18constantly on social media
15:20but you see that here
15:21in the friends section
15:21For example
15:22she clearly states the sources
15:23but it's content
15:24which starts purely
15:25taking quotes
15:27and all the great part
15:29of its production
15:30is based on content
15:32already existing
15:32on the networks
15:33that she takes back,
15:34that she commands,
15:34that she analyzes.
15:35That's exactly what we did.
15:36on the estate
15:36in artificial intelligence
15:37with the difference
15:38that no one was attacked
15:39identified person
15:40and on which
15:41the sources were cited
15:43but we talked
15:43working method
15:44So you see the problem?
15:47When I went
15:47on the forum part
15:48suddenly
15:49we learn
15:50that there will be
15:51a new video
15:51and yes my video
15:52certainly bothered
15:53even if she did
15:53than 1000 views
15:54and then
15:54I saw that there was
15:55a problem
15:56because suddenly
15:56the algorithm was spat out
15:58you saw
15:58the two messages
15:59that I received behind
16:00It's still crazy
16:01and you have people
16:02anyway
16:02who are quite awake
16:03saying
16:04but the code
16:05is completely broken
16:05on GPT-5
16:07but it's true
16:07and people
16:08which indicate
16:09that it would be necessary
16:10tests
16:10the most in-depth
16:11well yes
16:11at some point
16:11if you want
16:12we are here
16:13to learn
16:13skills
16:14and things
16:15and those who want
16:15to have fun
16:16either they go on Netflix
16:17but indeed
16:18if we have skills
16:19advances in machine learning
16:21that we put forward
16:22and that we make
16:23highlight
16:24like the elite
16:25machine learning
16:26it's true that it is
16:27a little bit of a shame
16:28to have content
16:28of this type of quality
16:29the problem
16:30you understood it
16:30how is it
16:31they will manage
16:32their communication strategy
16:33to turn off
16:34any debate
16:35who could contravene
16:36to what they said
16:37as official version
16:39So I invite you
16:40to subscribe to the channel
16:41to support her
16:42because you understood
16:42that we are in a movement
16:43independent
16:44we all have the right to make mistakes
16:45on our technical point of view
16:46we don't judge people
16:47we judge the working methods
16:48but on the other hand
16:49that we know method
16:50and it's good to know
16:51who we are dealing with
16:52and what are the methods
16:54which are used
16:55both on algorithms
16:56both on people
16:57to know their reliability
16:58and if we can trust them
17:00and there I am extremely disappointed
17:02I'll let you meditate
17:03on all that
17:03to share on social networks
17:05because I think
17:05that there are things
17:06that must change
17:07like what today
17:08I don't want to yet
17:09to keep quiet
17:10I wish you a very good day
17:11and see you soon
Recommandations
2:24
|
À suivre
2:16
2:52
26:31
22:48
23:23
14:51
0:52
29:31
4:13
3:42
Écris le tout premier commentaire