Comment  j'ai trouvé le prompt parfait pour GPT-5 ? - Vidéo Dailymotion

#ParlonsIA CHAT GPT - Perplexity -Claude

https://youtu.be/kTkVVm-xhfA  1 like de la video = 1 merci ❤️ MES FORMATIONS → https://parlonsia.teachizy.fr/   Découvrez les secrets des chatgpt prompts pour exploiter pleinement l’intelligence artificielle : grâce aux meilleures growth hacking techniques, l’ai development et les ai agents ouvrent la voie à l’autonomous ai, aux ai tutorial, à l’ai video et au prompt parfait, y compris le prompt parfait chatgpt. Vous vous demandez comment utiliser chatgpt ? Nous montrons comment utiliser chatgpt efficacement, avec des chatgpt prompts optimisés, afin de utiliser chatgpt dans vos projets. Entre ai news, latest ai news, et l’évolution de l’ai future, l’ai et l’intélligence artificielle s’orientent déjà vers gpt 5, chatgpt 5 et les usages avancés de artificial intelligence.   🔗 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   #gpt5 #chatgpt5 #promptengineering   ⌚ Timeline Vidéo   0:00 |  Le  "Prompt Parfait" pour ChatGPT-5.  1:02 | L'anatomie d'un prompt "parfait"  2:48 | J'ai piraté ChatGPT-5  5:15 | Pourquoi l'IA est sous-utilisée : L'exemple du diagnostic médical.  7:56 | Le cas Luc Julia  11:10 | DÉBUNKAGE  17:05 | "Reasoning_Effort"  20:51 | "Verbosity"  22:28 |  99% des tutos sont du marketing

Transcript

00:00Hi there, in this video we're going to talk about the perfect prompt for ChatGPT5, here's how to do it.

00:06Oh my, if I had known this before guys!

00:09But what I just learned is still crazy.

00:12No, but really.

00:12What we are going to see now are the specialists in French, even international, prompt engineering.

00:18So you're going to have to brace yourself with a lot of humor in this show, because we're going to talk about perfect prompts.

00:25Do you know who we're going to talk about?

00:27No, I know who you're not thinking of, the one who made 25 tutorials on the perfect prompt, another one.

00:33The one who is here, down below, who is hidden.

00:36And we're going to look at how he created the perfect prompt, and here's how to do it.

00:41What we are going to see together is how the perfect prompt is structured.

00:45What we were shown on all social media, including Twitter, is the anatomy of ChatGPT5 on the perfect prompt.

00:52The role, the task, the context, the reasoning, the output format, the stop condition.

00:55Are there really things that didn't exist before? That's the question I asked myself.

00:59Well, there may be stop conditions, but in fact, they existed before.

01:03And the point I saw was acting like an expert, a tour guide.

01:07So I said something to myself.

01:09Today we have AI models that hold 90% of humanity's knowledge.

01:1590%.

01:16And he is asked to make a tour guide in a perfect prompt.

01:19So I thought, will we one day have a prompt that will show us that we can work with AI?

01:24Will we ever get it?

01:26So I forgot to introduce myself.

01:27So I'm certainly one of the very first to crack the code of ChatGPT5 and Owen.

01:34ChatGPT5, I did it live.

01:35I was with you in a tutorial that you have in the description and I came in without doing it on purpose

01:40while I was testing OpenAI prompts to see how GBT5 worked.

01:46And unfortunately, without meaning to, I got into the system.

01:48And so, here is the system prompt.

01:50So, you have the video if you want to see it, because what will surprise you in this video,

01:56It's just that what you're told is too often marketing.

01:59And today, I am outraged by the marketing we have on social networks.

02:02The only ones who can speak today are great specialists.

02:06So today, with the knowledge I gained from getting into the systems,

02:11What I'm showing you is that if we actually wanted to use AI

02:14and that we have university knowledge available,

02:19why we continue to interact by asking him

02:22“Do you have a tour guide or write me a LinkedIn post?”

02:25What I'm showing you here are three different situations.

02:27and we will come back to the perfect accounts.

02:29First, we have an image and I ask him to identify the fabric

02:33and to specify what an image is.

02:35And I don't have an answer that is convincing by the model.

02:38I'm providing a medical context and I still don't have an answer.

02:41which allows me to use it in a medical diagnostic context.

02:44And when I put a prompt system that you learn in my training,

02:48you have it in "The Best of AI"

02:50because my goal is to teach you how to use AI in professional fields.

02:53We are on medical diagnostic analysis.

02:57We're not about writing a LinkedIn post.

02:59We also do LinkedIn posts.

03:01But I think that today, with the computing power of these machines,

03:05write a blog post with GPT 3.5 or GPT 5,

03:09It's a bit of a waste of the model's power.

03:11I will make you, if you want, a complete video on the use of the model.

03:14But what I'm going to show you is that if you drive the model,

03:17we are able, from a simple analysis,

03:20of an endoscopy on which I did not name the section,

03:24but with a very structured prompt system,

03:26to allow him to define that it is a globular pedicle,

03:30to define the size, to define what is called a gradation

03:32with a structured clinical analysis,

03:34morphology, color, vascularization, surface and other elements,

03:38to bring what is called reasoning,

03:40because they are models of reasoning today.

03:41And I rely on this from the studies of Mensa International,

03:45which are the IQ Challenge studies,

03:47on which you have all the models that are tested.

03:50So today, we have models that allow us to have IQs

03:54which vary between 120 and almost 140 IQ,

03:58which is considerable.

03:59And we are really on reasoning models on some of them.

04:02The question is how to exploit them.

04:05in the most technical areas

04:08and which will allow us to assist our decision-making.

04:11In this context, I show you very clearly

04:14how we will allow models to acquire

04:17an advanced reasoning system

04:19to do classification.

04:21And behind, the objective for me,

04:23It's about saving you time as a professional.

04:25Really time in the cutting edge fields,

04:27medical, legal and others.

04:29And there, for example, you have at the end

04:31of the model's reasoning assistance,

04:34mail to colleagues,

04:36mail to patients

04:37and the report that you can transmit.

04:39This is the power of these models today.

04:41So let's get back to what I discovered.

04:43seeing this formation of the perfect prompt.

04:46And how was the perfect prompt created?

04:49And how did that bring us?

04:50So, I'm already saying it, I'm going to use a little bit of humor

04:52because otherwise I wouldn't know how to deal with the situation.

04:55But we are still dealing with a real expert.

04:57in the field of artificial intelligence.

04:59These training returns are excellent.

05:02I saw, people are very happy

05:03and I don't question it at all.

05:05and I saw that he was someone extremely qualified.

05:08Recognized expertise

05:09in France and internationally.

05:11They share knowledge at Vivatec.

05:13It's still a very big reference.

05:15IA Sumit and BFM TV,

05:17which is known for having very high quality information

05:19and sourced, referenced, controlled.

05:22It is one of the best news channels,

05:23it would seem.

05:24And I think Luc Julia must have also been on BFM TV,

05:27but not only that.

05:28It went to the Senate.

05:29Luc Julia, does that mean anything to you or not?

05:31If Luc Julia, the gentleman who passed by

05:33and that it was shown that this gentleman

05:35may not really be the designer of Syria

05:38and it's been maybe 15 years

05:39that he is said to be the designer of Syria

05:41and all the TV sets called him

05:43right left

05:44and that he was everywhere

05:45and everyone said

05:45It's incredible what he says, it's incredible.

05:48And no one dared to question it

05:49just when a guy

05:51who had 300,000 subscribers said it

05:52and when small chains like mine

05:55speak,

05:57people blow them up.

05:59So, you know, not too long ago

06:01I made a video about a guy

06:02who does machine learning

06:03Defend Intelligence

06:04and I showed that the prompts

06:06that he was using were not working.

06:08The only thing we had to do

06:09it's blowing up my content

06:10and behind incredibly

06:12my account has been deleted.

06:13It's surprising, isn't it?

06:14What is interesting to see

06:15it is that when you demonstrate

06:16only specialists

06:18internationally renowned

06:20may not be

06:21in good practices,

06:23the best way

06:24it is to put out the fire

06:25by deleting what is said.

06:27So I invite you to become aware

06:29of what I'm saying.

06:30One, Monday morning

06:31you go to LinkedIn

06:32you post

06:33go watch this video

06:34you're going to get slapped

06:35because what I'm going to show you there

06:37you're going to get slapped

06:38but a slap

06:39like you can't even imagine.

06:40So here we have instructions

06:42with a code structure.

06:44That's the new thing

06:45which is brought to us

06:46in the perfect prompt

06:47of this video

06:48that you will find again.

06:49From this expert

06:50who has 10 years of experience

06:51in tech

06:52I can't have a resume like that

06:53I wasn't born.

06:54and that he worked within GAFA

06:56of unicorns

06:57technological

06:57and opportunities

06:58incredible.

06:59And it is certified

07:00by Microsoft

07:01and OpenAI

07:01and the only one in France

07:02to hold these certifications.

07:04So what I ask is

07:05where does it come from

07:06this formatting?

07:07I suggest you

07:08that we will ask

07:09to ChatGPT

07:10if this formatting exists

07:11if they are variables

07:13Airness

07:14Clarification

07:15and Autonomy

07:15which are official

07:17recognized

07:17and if this formatting

07:19in a chat interface

07:20is official.

07:21Luke

07:22if you see us

07:23kisses

07:24and thoughts

07:25we are thinking of you.

07:26We will ask

07:26to extract the text

07:27to ChatGPT

07:28and translate it in full

07:29in French

07:29to understand

07:30what is it again

07:31this type of instruction

07:32and is it really

07:33we must integrate

07:34this type of instruction

07:34inside GPT 5.

07:36So we see

07:37that it is an instruction

07:39in format

07:39it's a code

07:40which could look like

07:42to XML

07:42but it's not well formatted

07:44it's a mistake

07:44formatting

07:45code specialists

07:46didn't hesitate to tell me

07:47for me it's not formatted

07:48like this in XML

07:49to err is human

07:50but I appeal

07:52to those who code regularly

07:53and now

07:54we will ask

07:55to the model

07:55to go and search

07:57in all documentation

07:57GPT 5 official

07:59to check us out

08:00if the functions

08:01are official

08:02whether yes or no

08:03if they are functional

08:04is it a function

08:05official or not

08:06no function

08:07documentation

08:08proves that there is

08:09Ernest

08:09Level

08:10Ask Clarification

08:11Autonomy

08:12no function exists

08:13none

08:13and this function

08:15Ernest

08:15is used

08:16in audio input

08:17and I specify

08:18in API

08:19not in a cat

08:20the model

08:21can't stand

08:22the attributes

08:22Ask Clarification

08:23and Autonomy

08:24in the tags

08:25is it functional

08:26the answer is no

08:27it doesn't make tools

08:28API part

08:29and it is not recognized

08:30by apprenticeship

08:31by strengthening the model

08:32so no

08:33the model

08:33doesn't know

08:34Really

08:35use this instruction

08:36other part

08:37the model

08:38I won't go into details

08:39is capable

08:40by comparison

08:41vector base

08:42to understand

08:43that clarification

08:44No

08:45it means

08:45that he does not come to ask questions

08:46but the concept

08:47autonomy

08:48In fact

08:48does not exist

08:49in GPT-5

08:50it's a system

08:51said agentic

08:51it's a system

08:52decision-making

08:53he will decide

08:54to launch tools

08:56research calls

08:57API calls

08:58if you tell him

08:59how to use

09:00tools

09:00and that's what you learn

09:02to configure

09:02in my training

09:03you have the training

09:04to learn

09:05office automation

09:06it's the best

09:07of AI

09:0760 hours of work

09:09legal

09:10medical

09:11office automation

09:12images

09:12videos

09:13when you do

09:14instructions

09:15in system

09:15real instructions

09:16in system

09:17you should not do anything

09:18take up duties

09:19which are not

09:19official tags

09:20which are not functional

09:21and who are

09:22prototypes

09:23not supported

09:24This must be avoided

09:25so I don't really care

09:27but overall

09:29what you have there

09:30these are things

09:30which are invented

09:31from scratch

09:32that's roughly what it means

09:33In any case

09:34I invite you

09:35to show me

09:36officially the documentation

09:37who talks about these variables

09:38and this formatting

09:39because what you are going to

09:40discover in the sequel

09:41of the discussion

09:41goes even further

09:42surprise you

09:43such

09:44if you should have

09:45to write

09:46this text

09:47in a system

09:48which is said

09:49from xml

09:49that's how it is

09:50that it is written

09:51Exactly

09:52in this structure

09:53here

09:53the guys who code

09:54don't hesitate to tell me

09:55but not in this way

09:56it's not like that

09:57that it is written

09:58if you should have

09:59mark these instructions

10:00in a chat system

10:02you absolutely do not have

10:03no need

10:04to bother you

10:05to format them

10:06you can simply

10:07put them on

10:08in this format

10:09response rules

10:10enthusiasm

10:11clarification

10:12autonomy

10:12Oh yes

10:13I forgot to tell you

10:14but what does that mean

10:15this text written in English

10:16in a learned manner

10:18you ask to translate

10:19in French

10:20the model tells you

10:21it means enthusiasm

10:22so do you know

10:23a variable that is called

10:24enthusiasm

10:25is what you ask

10:26a variable that is called

10:26autonomy

10:27it doesn't exist

10:28but hey

10:28why make it complicated

10:29because I don't know

10:31ask him

10:31ask him the question

10:32don't hesitate

10:33but don't hesitate either

10:34to retweet all this

10:35because today

10:36as I told you

10:36the specialists

10:38of AI

10:39produce content

10:40who for me

10:41does not rely on anything

10:42on prompt methods

10:44of work

10:45and that for me

10:46it's part of it

10:47In any case

10:48I ask to see

10:49where does this information come from

10:50official formatting

10:51if the model should format

10:53This is what he considers

10:54you put a text format

10:56in a format

10:56simply instruction

10:57you must answer

10:58with a high level of autonomy

11:00it doesn't mean anything

11:01autonomy for oneself

11:02the model

11:02This is not the type of instruction

11:04that he needs

11:04it's an instruction

11:05much more technical

11:06and structured

11:06because it's a system

11:07which is said to be agentic

11:08a decision-making system

11:10who can take

11:10behavior loops

11:12we'll talk about it

11:13a little further

11:13in the video

11:14but what he will understand

11:15is that he will not ask questions

11:16and he will be on a tone

11:18unenthusiastic

11:19good if you want

11:20a little something

11:20not enthusiastic

11:22it's up to you

11:23and otherwise in the calls

11:24in API structures

11:26JSON or others

11:26that's how it is

11:27that you put

11:28the structure of your query

11:29this information

11:30she is simply

11:31present here

11:32you have the formats

11:33OpenAI officials

11:34and it's marked here

11:35and you see

11:35how it's formatted

11:36so in java

11:38in python

11:38and in curl

11:39so in no case

11:40it's formatted

11:40as it is formatted

11:41in the example which is here

11:42which will be very interesting

11:44it's only now

11:45that we saw this first part

11:46we will see that

11:47he introduces us

11:48a new variable

11:49you see how it is marked

11:50how it is structured

11:51reasoning is strong

11:52then this variable

11:53it exists

11:53does it need to be formatted

11:54like that

11:55in a cat instruction

11:56the answer

11:56it's no

11:57the target system

11:58equals low

11:59doesn't even exist

12:00otherwise you bring me

12:02the proof

12:03formatting

12:03look how it's formatted

12:04reasoning is strong

12:05minimal

12:06there is no such layout

12:08complex

12:08and we are on

12:09a code part

12:10reasoning is strong

12:11minimal

12:12and piton

12:12here is the formatting

12:13So I'd like to

12:15that we have business

12:15to specialists

12:16who are called

12:17even on BFM TV

12:18and VivaTech

12:19but I don't see

12:20no this documentation

12:21wherever it is official

12:23how it translates

12:24in terms of use

12:25of the model

12:26because whatever anyone says

12:27what is in the end

12:29important to you

12:30in terms of users

12:31this is how to use

12:33correctly these models

12:34how to learn

12:34to work with these models

12:35how to get out

12:37disinformation

12:39of the vibe

12:40and marketing

12:40that you have

12:42on the networks

12:42first of all

12:43if I ask a question

12:44very simple

12:45to OpenAI

12:45to GPT 5 cat

12:47premiere date

12:48world war

12:49I have an answer

12:50of the model

12:50now you no longer have

12:52than to use

12:52simply

12:53the variable

12:54reasoning effort

12:55thinking

12:55in this way

12:56no need to put

12:58of the code

12:58in a cat version

12:59there is no need for it

13:00and in this case

13:01you will pass

13:02of a system

13:03who had not

13:03of reasoning

13:04to a system

13:05which will tip over

13:05in medium system

13:06it's here

13:07that you will find yourself

13:07in the mode

13:08operating

13:08of the model

13:09so you go

13:09increase

13:10precision

13:11accuracy

13:12of the answer

13:13we will send

13:14the variable

13:15in medium type

13:16and you will see

13:17that the system

13:18will tip over

13:19by opening

13:20a reasoning system

13:21because it is a variable

13:22which is known

13:22by the model

13:23so by clicking on it

13:24the reasoning

13:26will arrive

13:27it will decompose

13:28the situation

13:28and he will bring

13:29at the end

13:30an answer

13:31and so that

13:32these are the tokens

13:33from the black box

13:34reasoning

13:34of the model

13:35before answering you

13:36the written model

13:37his reasoning

13:37and that's how it is

13:38that we are talking about

13:39thinking model

13:40and so you have

13:41the answer

13:42here it is

13:42short answer

13:43November 11, 2018

13:44so as you see

13:46we don't need it at all

13:47to format

13:48in this type

13:49formatting

13:50which is neither known

13:51nor recognized

13:52neither official

13:52the person

13:53who made this tutorial

13:54said

13:54if I do

13:55reasoning low

13:56I have an answer

13:57which is not too much

13:59structured

13:59and if I put

14:00in high effort

14:02I have an answer

14:02very structured

14:03it's not entirely true

14:05and it's important

14:06to understand it

14:06I will explain to you

14:07one thing

14:07which is important

14:08in API

14:08that's to say

14:09on systems

14:10playground

14:10but which is not

14:12the case

14:12in a discussion

14:13when you continue

14:14a discussion

14:15with the model

14:16if I resume

14:17the same discussion

14:18and I resume

14:19the same variable

14:20medium

14:21no more no less

14:21I send back

14:22the request

14:23you know what

14:23I am doing

14:24the model will use

14:25the previous reasoning

14:27and will summon him

14:28to his new answer

14:30and you see

14:31that the model

14:32produced a response

14:33a little bit longer

14:34a little bit I said

14:35it's exactly

14:35what will happen

14:36in the specific case

14:37and that's explained

14:38by OpenAI

14:38in the official documentation

14:39it's called

14:40reuse

14:41from the reasoning context

14:42and it allows

14:43to activate the

14:44previous response id

14:45when you use

14:46the same context

14:47the benchmark rate score

14:49that's to say

14:49statistical improvements

14:51significant

14:52assessments

14:53improves

14:54you send back

14:55the previous one

14:56element of reasoning

14:57which allows

14:58to make climb

14:59the score

15:00of quality

15:00of the answer

15:01so if you return

15:03the same request

15:04and that the model

15:05just before thinking

15:06and you put it

15:08in low

15:08in fact he will use

15:10his reasoning system

15:11previous

15:11and it improves

15:13the answer

15:13that's why here

15:14you have an answer

15:15which is better built

15:17it's not because

15:18it's high

15:18that you have

15:19more length

15:20that's a completely different one

15:21variable

15:22that we will see later

15:23so you understood

15:24that in this structure

15:25promptly

15:26to put

15:27the structured variable

15:28like that

15:29reasoning effort

15:30with this formatting

15:31and the target

15:32it has no interest

15:33that the airness

15:35does not exist

15:36in the variables

15:36that autonomy

15:37does not exist

15:38in the variables

15:39and that all this

15:39is not present

15:40in the documentation

15:41official

15:42that the simple fact

15:43to click

15:44and write it

15:45in a chat interface

15:46in this way

15:47that's enough

15:48and if you want

15:49format

15:50curl

15:51javascript

15:52or python

15:53this is the format

15:53that it should be noted

15:54who is here

15:54so the format

15:55who is here

15:56does not exist

15:57it doesn't come out of nowhere

15:58I invite you

15:58to show me

15:59where does it come from

15:59repost this on

16:01LinkedIn

16:01Twitter

16:02wherever you want

16:03you have

16:04content

16:05by the greatest

16:05specialists

16:06who today

16:07I don't see

16:08where do they come from

16:09their statements

16:09and unfortunately

16:10it's practically

16:1199% of tutorials

16:13so the efforts

16:14of reasoning

16:15you have 4

16:16you have what is called

16:16the minimal

16:17the low

16:18the medium

16:19and the high

16:19so 4 levels

16:20of reasoning

16:21that you find

16:21in the documentation

16:23OpenAI official

16:24on benchmarks

16:25and now

16:26we're going home

16:27on the variable

16:28verbosity

16:29so what is it

16:30verbosity

16:30so I allow you

16:31to meet me something

16:32you saw that suddenly

16:33different formatting

16:35look

16:36you have a space here

16:37you no longer have the structure

16:39from the beginning

16:39Short

16:40the code has changed

16:41is this XML

16:42well no

16:43it is badly written there too

16:45he is different

16:46from what we saw at the beginning

16:46so the question

16:47that's why at the beginning

16:49you write the code like this

16:50and why it transforms like that

16:52if you have the explanation

16:53I want it

16:53but officially

16:55that again

16:56there is no need at all

16:58to write it like that

16:58we will resume our structure

17:00on reasoning effort

17:02this time

17:03we will remove the reasoning

17:04we will put verbosity

17:06the verbosity

17:07you saw that

17:09the variables

17:09you have them

17:10officially

17:11in the documentation

17:12that's just the variable

17:13you don't have to look any further

17:14so no need to write text

17:16complicated with layouts

17:17scholars

17:18that look like nothing

17:19I set the variable low

17:21and the model

17:22I will resend the request

17:23will answer me

17:23a sentence

17:24In fact

17:25what does that mean

17:26this system there

17:27I will explain it to you

17:27in more detail just after

17:28it's that the model

17:29had an apprenticeship

17:30by reinforcement

17:31who tells him that

17:32when you say low

17:33you just answer one sentence

17:34when you say medium

17:36it's a little bit like

17:37ristretto coffee

17:39the coffee

17:40long coffee

17:40medium

17:41you have something

17:41a little more developed

17:42and the high

17:43you understood

17:44the model

17:45will develop using

17:47what is called verbosity

17:48that is, a detailed answer

17:49and so he will open the subject

17:52from multiple angles

17:53provide examples

17:54and have

17:55a more gulpable vision

17:57this information

17:58you have it

17:59if you ask the GPT chain

18:00to understand the value

18:01verbosity

18:02the verbosity

18:03are categorized

18:04in three values

18:05minimalists

18:06balanced

18:07and detailed with nuances

18:08and to go further

18:09to understand

18:10we also have score structures

18:12and that actually explains

18:14that in fact

18:14we use a standardized scale

18:16which allows to describe

18:17behavioral intensity

18:19of the model

18:19very short answer

18:20telegraphic style

18:21balanced response

18:22little context and concise

18:23and the detailed model

18:25nuances and examples

18:26to put it simply

18:26for those who want the essentials

18:28low it's a tweet system

18:29medium is

18:31short encyclopedia

18:32and high masterful response

18:34Here are the explanations

18:35of these variables

18:36in conclusion

18:36no need at all

18:38to integrate this type

18:38formatting

18:39all that

18:40you would know it

18:40really use them

18:42I tell you

18:43because at one point

18:43I am limited

18:45surprised to see

18:47this type of content

18:47by professionals

18:48I don't know

18:49if we have to laugh about it

18:50or you have to cry about it

18:51but at some point

18:51where does it come from

18:52this type of formatting

18:54where do they come from

18:55this information

18:56where you have

18:57the systems

18:58self-reflection

18:59it exists

19:00a capacity

19:01that we call

19:01self-consistency

19:03models

19:03the models are capable

19:04to have an analysis

19:05of their reasoning

19:06and do draft prompting

19:08and I'm talking about it

19:09in training

19:10prompt engineering

19:11Excellency

19:11we approach

19:12all these subjects

19:13for professionals

19:14of AI

19:14those who need

19:16to scale

19:17their activity

19:18and to delegate

19:19complex tasks

19:20or to create

19:21systems

19:21chatbot

19:22or systems

19:23agents

19:24because I have training

19:24agent

19:25in this case

19:26it's important

19:27to define

19:28the behavior of the model

19:29but it is not

19:31because

19:32you give him

19:33of autonomy

19:34which is a variable

19:35which is not known

19:36by the model

19:36or tell him to think

19:38before acting

19:38the model is already thinking

19:39it's a model

19:40of reasoning

19:41he will define

19:42his level of reasoning

19:43in I

19:44so automatically

19:45so what we do

19:46this time

19:47we use

19:47the variable

19:48reasoning effort

19:49we pass it in I

19:50I find

19:51that he responds better

19:51when we put use

19:52we can remove it

19:53and see how it goes

19:54but I find

19:55that in general

19:55he took better

19:56instruction

19:56so logically

19:57it should trigger

19:58anyway

19:59ah he didn't trigger

20:00we are good car

20:01so you will see

20:02This makes me think one thing

20:03I said

20:04I find that he responds better

20:05when I tell him to use

20:06reasoning for I

20:08and in this case

20:09you saw

20:09it triggers reasoning

20:10so for me

20:11this little point there

20:12makes the difference

20:13in understanding

20:15of instruction

20:15it allows to show

20:17A

20:17that we absolutely do not need

20:19to go through the phase

20:20we tell him

20:21think before you act

20:22and briefly explains

20:23your logic

20:24Why

20:24there is the logic

20:26she is already here

20:26the reasoning of the model

20:27there is no need

20:28to make him do more

20:29you are talking about tokens

20:30of response

20:31and from there

20:32the thinking is over

20:34and the model responds

20:35behind

20:35and he gives you an answer

20:36more complete

20:37because he was told

20:38high verbosity

20:39There

20:40so on this model there

20:41you have an answer

20:42more complete

20:43so overall

20:44these prompt structures

20:45In my opinion

20:46they are not based

20:47on these structures there

20:48on nothing official

20:49even worse

20:51I think that

20:51these instructions

20:52I found them nowhere

20:53I find that

20:54very worrying

20:55by professionals

20:57so I invite them

20:58to find

20:59and give their source

21:00to understand

21:01where do they come from

21:02these formats

21:03sometimes in structure

21:04falsely xml

21:05which change along the way

21:06and then

21:07there is a question too

21:08that you need to ask yourself

21:09on how you prompt

21:10you have seen everywhere

21:12you have a specialist assistant

21:13in strategic analysis

21:14the mission is to analyze

21:15economic impacts

21:16potentials

21:16of a sudden rise

21:17of the price of oil

21:18so you wait

21:19to have an incredible relationship

21:20for the impact of oil

21:22on the economy of a country

21:24and therefore all the mechanisms

21:25so the model

21:26we are a model of reasoning

21:27we are on the best models

21:28Today

21:29who are capable

21:30to make connections

21:31and correlations

21:32the very serious test

21:33MENSA shows us

21:34this system

21:35who are capable

21:36to find answers

21:36to problems

21:37who are unknown

21:38they manage to have

21:39an IQ

21:39which is between 125 and 140

21:41how can we

21:42Today

21:43have answers

21:45of 800 words

21:46on complexes

21:47also important

21:49than an oil shock

21:50which could modify

21:51the costs

21:51on an economy

21:52of a country

21:53you have a document

21:54which is 800 words long

21:55with answers

21:57trivial

21:58trivial

21:58the oil shock

21:59will have

22:00a deterioration

22:01from the current account

22:02yes the current account

22:03it will empty

22:03because the essence

22:04costs more

22:05it will compress

22:06real income

22:07households

22:08well yes

22:08it costs more

22:09so if we have

22:10a model

22:11of such a level

22:12of power

22:12to achieve

22:14to a triviality

22:15similar

22:15me today

22:16I have trouble

22:17to have

22:17applicability

22:19use

22:20of this type of prompt

22:21saying

22:22we are

22:22with models

22:24incredible

22:25AGI

22:25who are going to work

22:26we don't have

22:27a report

22:28which holds the road

22:28there we have nothing

22:29for me we have nothing

22:30we don't have an analysis

22:31complete

22:31of an oil shock

22:32on an economy

22:34of a country

22:34so the strategy

22:35and the way

22:36to count

22:36as I tell you

22:37are very far away

22:38of what you are sold

22:39as a concept

22:40that for me

22:41these are not

22:41methods

22:41that we can use

22:42in the professional name

22:43you will not have

22:44a report

22:45that you can sell

22:46to a company

22:47saying

22:47here is a report

22:49on strategic analysis

22:50of the oil shock

22:50on the country

22:51I am ready

22:51to debate

22:52with the specialists

22:53specialists in general

22:54as until now

22:56what i see

22:58working methods

23:00that I can't find

23:02functional

23:03which are not even

23:04present in the documentation

23:05official

23:05so the question

23:06today to ask yourself

23:07is this what you want

23:08continue to use

23:09ChatGPT

23:10being aware

23:12that you have

23:12among the models

23:13the most educated

23:15on human knowledge

23:16they own 90%

23:18of knowledge

23:18and we limit ourselves

23:20to make posts

23:20LinkedIn

23:21and the blog

23:22with

23:23if you want to use

23:24these prompting methods

23:25and this way

23:26to interact

23:27it's up to you

23:28In any case

23:29I teach

23:30different methods

23:31I think that today

23:32people

23:33who need

23:34of decision-making

23:35assistance

23:36medical

23:36legal

23:37decision making

23:38report analysis

23:38need

23:39other types of strategies

23:40and if you want to know more

23:42you have all the information

23:43the page

23:44my training

23:45training courses

23:46should I say

23:47and you have the possibility

23:48to contact me

23:48by email

23:49by PM

23:50and if you want to train

23:51your teams

23:52don't hesitate

23:52to contact you directly

23:54I will answer you

23:55as soon as possible

23:56if it's not done

23:57subscribe

23:57share this video

23:59on the networks

24:00because today

24:01is it really

24:02we were shown

24:03creation

24:04of a perfect prompt

24:05and methods

24:06of interaction

24:07from my point of view

24:08we are very far from it

24:09and unfortunately

24:10I saw a lot of things

24:12which in my opinion

24:13weren't even

24:14officially recognized

24:15by the instructions

24:17GPT cat

24:17and OpenAI

24:18but of course

24:19I'm waiting

24:20the evidence

24:21and the elements

24:21in a video

24:22or in a two-person video

24:24where we will debate

24:25on the subjects

24:26I'll see you soon

24:27and see you next time

24:27Thanks a lot

24:28for your attention

24:29I'll see you soon

Comment j'ai trouvé le prompt parfait pour GPT-5 ?

Catégorie

Transcription

Recommandations