Big, big stress, because when I see the quality of the presentation, I think I'm going to stay on one slide, because I don't have the little pictos and everything. So, I didn't really have the time, etc. And I'm going, in fact, not by a desire for conflict or anything, but really trying to show another vision.
That is to say that there, so first I'm going to introduce myself, so Bruno Giacconi, I was in IT for 12 years at the FIA, those who don't know Formula 1 maybe they know, so I was responsible for the IA deployment, etc. and then for reasons because it is a federation that is moving towards England and then that I could not move so we stopped there really not long and so it
really allowed me to exploit a lot of things to really enjoy myself and that's what I'm going to try to communicate to you today I'm going to explain a little bit my starting position in relation to all these current evolutions.
I think, because I've been going a lot on LinkedIn since I'm no longer at the EFI, I'm overwhelmed, I see a lot of posts with a lot of IA consultants, we end up with crazy graphics, with crazy theorems, we get lost.
And in fact, what do I see behind all this? I see the human species which in fact turns its ego as much as it can he has little time left to express like that its power of organizational reflection etc all that I am sorry really to tell you some certainly present it
already others not but it will disappear it will be taken over by the IAEA so it will still have to go to another level of approach and there all these consultants who really do the max to explain to you precisely the triggers the imbrications the ones to the others etc
when in fact at the moment we already have even if we know the security problems behind still a little bit at the moment that too it will disappear when we already have systems that they are already able to do the trigger to do the loop to do the goal of the site of the there.
And what I would like you to understand, and there will soon be more, you know, powerpoints with crazy diagrams and all that, it's something that will disappear. We are already, for me, in the old world, with these things a little complex, etc.
There, we are going to start a new world, where it is no longer the engineering that comes first, it is the vision. because we still have a little bit of time left. The vision, the posture in relation to Alia, that's it.
So that's super important and just to say something that marked me in my childhood, my dad was an intellectual etc. Once I told him, but dad we agree, eternal life it does not exist and I was very small and he answered me well no he told me well you see well it is inevitable we will not be able to go against before dying at 50 years old now at 80 you can imagine that it cannot stop and those who will tell you no it is impossible they are wrong so now the big effort is every time you see something happen to be able to see further.
For example, when CHGPT arrived, six months later, you see prompting engineers everywhere. Yeah, I know how to prompt. No, but me, that's how you have to do it. It was obvious. I see that, I say to myself, if really IA is what it is, in two years, the prompting is over. And we were not wrong.
Above all, never do your prompting yourself, because in fact you will always be a little below what you need. There is nothing better than an IA itself to prompt, but it is a small example.
Always take a little bit of advance and make an effort to see what will happen later. Here I have raised my head a little, I know that I am really outside the I am perfectly aware of that, but that's it, there is passion, there are convictions and really we are going at full speed.
So today, in fact, what interests me is this aspect, and that is really important, and that the IA can bring you the information, but she won't be able to do this job for you, it's the techno wave, it's important, even if it is taken in relay, even if it has done a lot, you have to know everything that comes out, why, how, it will actually drive you in, and if I did that, and if I put such a thing in place, and you will see in fact if you, and we criticize a lot, you know the
the google thread like what we would be spied on and everything I love it in fact that's what it's almost what saved me a little because in fact it allows me as he saw that I was interested in lia every day I have a lot of stuff on lia so I arrive at the end of the day
attention the catch is to fall into the effect of hype it's new new stuff that's it's great wow and after you are in the low curve of the hype everyone knows this curve etc. So always take a little distance.
I see something new coming, I take a little distance, but I start to think about it and I also start to think about what it will give later. There will always be a sequel. All the new things will give a sequel. Always think of this sequel because it will make you innovative. And that's all that remains for us. I tell you, face to face, that's all that remains for us.
The technicality, unfortunately, it will explode and in fact after this week it is how to evaluate a solution it is good to go for example on chat gpt do two or three tests do two or three montages the problem is that in
In fact, you will always be a little, not a little, largely below what the solution can produce. So you will be in the incapacity to say, no, I know well, it's good, no, it's impossible. You know two, three little things, but in truth, if it's based on tests, it's going to be very complicated to really feel the solution that you have in your hands.
And in fact, I put in place a little technique that allows me to get to the end, to still understand the limits of an IA.
It's very simple, you find an application field that you know well, that you master quite well, not necessarily in IT. And you say to yourself, with this domain that I know well, I will actually produce an application. And out of the blue, since you know the application domain well, you are going to make an application that will still push the plug quite far in terms of functionality, etc.
Because each domain has a degree of complexity. And you're going to take care of it. And these areas of complexity, the area you're going to attack, you're going to submit them to the AI.
And there, you're going to quickly realize what the AI is really capable of doing.
So to make it short, not long ago, there was the Astra model. You know, there was the Dev Day of OpenAI. The Astra model that came out. I see that and I say, I said to myself, it's crazy, because they really care about 3D modeling. I said to myself, wow, it's crazy, it's the first IA that puts 3D modeling so far ahead.
For me, modeling is huge, 3D printing, the thing, it's huge. That is to say, it's a step in reality. We start talking about volume, etc. But between modeling and understanding the world, there is a gap. And so I had to find a solution to see how, for example, the Astra model plays with this modeling, what it is capable of and how far we can go.
And in front of me, I will not go into personal details, I had my partner who had been telling me for a while to renovate his apartment. 3D modeling, apartment renovation.
I had a company when I was very young in IT and then I was bought, I lost everything, it was dramatic. I found myself on the street and everything. And suddenly I had to sell houses for six months. But except that I learned a lot of stuff.
What is a roof? What are the closets? What are the wall carriers? window, door, volume, etc. I said to myself, I have everything and that's really what I encourage you to do. Go and draw in your passion, knowledge, etc. to tell you, here I am going to make an application on this or that theme and it will actually allow me to really work around the capacity of this IA.
So I started there. So I chose the environment offered by the OpenAI platform, Astra, since I know that this solution, a priori, models. And I started slowly at the beginning. I started slowly.
The idea was, I give a plan, even if it is not very good since it must be addressed to anyone. I put a house plan, even drawn by hand, I put the footage and everything. And the IAEA will have to be able to finally rationalize all this and really make a real architect's plan. And then possibly be able to move the fences, etc., redo everything.
Then there is the other step that I did not start at all, which is much more complex, which is to propose that yesterday offers you volumes of life as much to tell you that at the moment it is impossible you will understand why but it can already do nice things like develop and there you see then less total admiration creation of agents trigger imbrication of things one to the other is magnificent problem is that what I am looking for is to see how far this frontier model goes.
The example I gave you is really the baker who kneads the dough and as long as you are not in the hard it is impossible to feel the heat of the dough and how far we can go and what bread we can make of it. Too big, too long, there are a lot of worries.
So I mounted this little thing and I started to make the plan. My friend told me it's extraordinary but why don't you go any further etc. there I say let's go I like it
let's go and so ah yes so we agree my approach and it's an advantage in my eyes and come on we will say it more or less messy since that too the fact of having a slightly messy approach in simple dialogue with a thread of discussion afterwards you have to change it when you exceed the number of possible contexts but you will have done everything in advance so that they do not lose anything in your construction in fact they take up all the information for a good continuity that's important I would need a lot more in my opinion than 25 minutes but hey we're going yeah 20 minutes so I'm going to try to make it short and so here I am I put it in place and I started to perceive all the limits, all the problems that come with it.
The problem is that we are faced with models that are not, unlike what Yann Lequin is trying to do, and I really hope he will succeed because it really changes everything, we are on models that have no knowledge of the world. The only thing they know how to do is predict the next word.
But the knowledge of the world, what is a piece of life? for him, it doesn't mean anything. What is conformity? It doesn't mean anything.
We're going to have to put everything in fact, to standardize that in md files, markdowns, to say here is a passage descent corridor, it's going to be minimum time, width. This table there, you can't actually put it back in the wall.
He knows how to model super well, but he has a modeling capacity you have no idea that is to say that we were going on one I was going on an ikea page I was taking ikea elements to go fast with its all metrics the furniture and everything is ok no problem I do it to you he does it to you in five minutes he did the furniture with several pauses he had understood by himself pose on the ground there are adjustable feet pose against the wall etc and And once you put the furniture in the app, sometimes you end up with a TV set in suspension and in the air. How is that possible? My application, my algo, had to prevent that. But in fact, you realize that his knowledge of the real world is so low that he does a lot, a lot of nonsense.
Even if the goal is not just him, we agree. agree it is the application is its construction of the application which is capable through algorithms that we will have put in place to say such a piece is against a wall and as it is fixed to the wall the only thing we can do is to lift it to mount it to move it on the sides in no case I can move it back to me and it must be algorithmic algorithmic
The goal is not that the IA, Astra, does that itself, it is to find an application, to build together an algorithmic application that deploys all that and I told you I do not have the skills that an engineer could have in IA.
What I am telling you is that you, each one of you, you are obviously able to do this type of application. You You can't even imagine how much you can do it. You just have to get started.
The agents, for example, the sub -agents. Well, yes, Bruno's solution is extraordinary. It's built, it's fabulous, it's dental, it's perfect. It's ultra -safe, it's all you want.
But know that there is another life, there is another possibility, which is through a thread of discussion via the OpenAI structure, creation automatically at your request but he will do it after a while when you start dispatching too many subjects at the same time he will create you your agents your sub -agents who will be extremely well orchestrated between them so as not to break a version that they have just provided etc everything is hyper -cadenced native good to force to advance on this I also realized
realized that we can't, it has disadvantages, for example at the moment we can't work simultaneously more than four agents, more than four sub -agents, so we created 17 agents but there are only four who can work simultaneously, you see, and all that there, I would not have known it without creating this application, it is really a real way to create, to know
to be there so basically it's bad I mean don't expect something exceptional we are well agreed so there we will imagine that there was the conversation between me and there that I deposited the documents that we agreed etc.
and then in fact or that I draw my plan by hand it is also an envisageable option if I can not stay on the the screens I do not have enough time so we will immediately click and display problem
here good in short and there we fall on the app so we have we have an approach a little type photoshop see with the little things that will allow you to respect such or such a place with the dismantling systems etc.
and in fact here at the beginning there were not these integrated cupboards so So we positioned them, but it's not like I'm positioning, I'm doing it by hand. No, the integrated cupboards are part of the application, because you have to know that before discovering an apartment and when you renovate it, we have several steps. We have the second work, with the bathtub, with everything that is integrated. It's quite, I'm sorry, it's a bit of a particular area, but when you know it, it's really interesting.
Here, I can add a window at any time. moment add a window I can obviously there in theory the shift but I think it starts to be blocked because you see there is also a notion of volume I can it's very complicated by looking at the screen and his mouse here I can break certain volumes here it is
c 'est le bug de typique de microsoft quand il veut faire sa démo et là on dirait que tout est bloqué on va changer de plan comme ça ça ira bien et en fait il ya un système de vases communicants avec le modèle archi qui va vous permettre en fait de beaucoup mieux percevoir ce que vous have it is horrible there is its bug in all directions I am really sorry here it is and for example
so I can obviously select in all its armchairs etc. are part of a gallery that was modeled etc. you see I was talking to you about a piece of furniture under the TV this is the one that made me it was quite boring so we can move them like that but also manually there are plenty of possible options here so if I go on a
chair I have my product with all the finishes of the sofa obviously composition of the sofa the alternatives I can change the sofa we also modeled other sofas and And I can, of course, there are problems because I also tested that, you know, we can now host the site directly in OpenAI, so I also experienced that. And so, in fact, I can see, in fact, big slowdowns as I accumulate the modelings.
That's barely five minutes of modeling and there, we may not see it, but everything can open up inside, inside everything has been respected etc I had to remove some options because in fact it weighed too much on compared to the infra open ai and go so that you can see a little
bit a little nice thing we can obviously we are going to communicate by visiting the apartment so there obviously I turn I do what there is what it takes and like that I can see concretely what the little decoration looks like which is good what we did etc. is it interesting etc.
so what we are going to do we are going to take a slightly different angle like that in the kitchen I do not even explain to you so here it is really a terrible puzzle because it is really complicated everything is real that is to say each box corresponds to boxes we will say standard which come from a supplier and if you shorten for example your kitchen, the system must, if you are on a somewhat intermediate length, if your questions do not correspond, it must understand that it must add, as a cook would do, things for the too much or the not enough. It's really fine, it's really ...
I attacked myself and I realized a little late that the IA is is to do less, okay they model, but to understand what a kitchen is, to understand what a dishwasher does, it's a huge complexity. So we have made good progress, but there are still a lot of things to do.
And there, because for me, the total IA does not suit me in the applications, because in fact there is such instability, I am not talking about development. Development is what knows how to do better. So I take that to test my IA.
On the other hand, integrating the IA in the the solution itself, I go all the way. I only do what I succeed almost at a glance. If that's not the case, I drop it. Or then it takes engineering, it means a month of work to frame the thing, for me it's not envisageable.
There, small nice option, that you understand a little the purpose of this thing there, that you see, that seems to you, I think, a little strange.
I take a picture that I framed as I wanted, it had to fit here, and there it is one of the rare ones, apart from the reception, it is one of the rare API, because in addition the Air de Rien, it costs quite a lot and I would not want to spend too much money either.
So here I Pinterest and in fact it will allow the customer, my partner, to really be able to project themselves. Because it's beautiful 3D, no but you will see, we enter a completely crazy world, it's really, it's completely crazy.
That in the end I will make sure to have the possibility to the person who pinterest to define when even there I am always you will see it is always a bit of a cocooning approach it is always a bit of the same style but you will see that in fact it is not you will see but at the end here and we end up with finished things baum it is here I put myself anywhere in the apartment and I will sublimate and it seems like nothing a
A person who renews, beyond the plans, so that they finally understand, yes, it's the right solution. You still have to be able to project yourself with something really nice and that corresponds to the real code. We are in the real.
And there, it seems like nothing, all that, 3D navigation, the archi mode, it's a shame because I have a power problem anyway. but we can do a lot of things change furniture finally we can really even the luminaires finally good it's huge the plan mode I didn't show you but I'm still going to show you because it has to be useful that's good to develop if there is no utility really it's
not cool so there for example you have that you have done your thing and everything normally I am able to move ah bah tiens voilà là j 'ai moins de problème d 'une fenêtre ça peut se décaler bien évidemment etc il refusera une pièce sans porte voyez ou alors une pièce en dehors de la salle de bain sans fenêtre il faut quand même respecter les normes il ya plein de subtilité et le sens
des portes une porte ça s 'ouvre d 'un côté de l 'autre ça peut être une porte coulissante plein a lot of things and it integrates all that the system I can not go into the details but everything is a little here you see I have a lot of things good in short and in fact for example what is
important is to be able to once you have defined your plan etc. it is to say to come to a master of work or even artisans with the master plan so what is interesting is that after
After doing that, at any time, you can export the plans, layout plans, technical files, channeling plans, we will say, go download PDF, here it is, it's instantaneous, it's pretty fast,
and you can therefore at any time give this plan, totally measured in length, in width, across, to your master of work. It's not just playing around, etc.
no no no when you start delivering solutions it is really important to stick with a necessity a reality a concrete reality and I even went further it's nice to put to know that a house well it has water spills out of nowhere it is managed so it is crucial during the first exchanges with lia to determine the technical gain and if he does not understand
Obviously, the IA will explain to him, but a technical gauge, that's where the water is going to flow. But if, for example, there, the bathtub, you place it rather here, there will be a big problem. Because, in fact, you can't pass a technical gauge in the ground, otherwise you smash everything, and it's priceless, and then it's flat.
There is no, there must always be at least 2 % of axis. so it is obviously I have put in place an algorithm that allows you to calculate all this according to the positioning it will calculate the shortest path by stretching the walls to always tell you yes it is in order where you placed it I can and I will put it but
in the archi version you have seen 3d the small chest that goes well to hide obviously we are going to the end of things it is really necessary that it is used to see it is extraordinary and that there was a thread of discussion see two see three because after a while bug we exceed in fact the capacity but I will show you another little thing very important for
me it's all the admin part and in fact everything is documented in fact I left and you know You know, you can, in fact, without having crazy knowledge, without having to say, yeah, a skills, a this, a that, cybersecurity and all. No, you take the thing up front.
You tell your IA, I would like to, if I am audited, he will understand right away. If I am audited, I would like to be able to have the ISOs that are going well. And stay even a little vague, use two terms, two, three. Well, he understands, he gets into the audit situation, we're not kidding anymore, so we're really going to have to, in case of audit, all the technical files and everything, they must be visible, everything must be traceable, versioned, etc. And that's what we did. No need for crazy skills to put that in place.
And of course, all the skills are are available, so of course, everything that is Markdown, there are some everywhere. And above all, why did I do this admin with consulting the skiffs, downloading them, etc. I'm not showing you everything, everything is noted, nothing is put aside, everything is versioned, etc.
When you go there, what is interesting is to be able to say at the end, I have a new IA, I have a new technology that has just arrived, I have to be able to test it. And you have to be able to migrate what you have done in order to, in the case of emergency, be able to go on a new solution without losing too much along the way.
you must immediately have a cloud code you plug it into your on your file tree you have nothing to say ok ok so the skills are there very well to the cq ok you go up there ok very well then I will maintain this standing the graphic chart everything is everything is understood in it explained the slightest button the effects etc I can obviously change everything a text size etc it's not to please me it's
in fact in case of migration always think migration to not stay on astra why would I stay on astra if then claude code brings me the same capacity or even better in
modeling here is roughly my speech I know that I am pro it's not at all what we are used to to see, but I could see the limits. It costs a lot. You see? No, that's not cool. No, really, it's something that I'm thinking about.
There, I'm going to go to Anthropique, because you saw, recently, a 10 % reduction in prices, because in fact, there is a way now to... It's not a diffusion, it's another system to replicate their IA. Which means that everything is cheaper, learning is much cheaper, so for the final client it's cheaper.
And there, out of nowhere, to do this work in normal time, that is to say, all that took me 30 hours. Well, at the time, it's not 30 hours, even a team and everything, it's time, it's huge. 30 hours but it is with a subscription at 500 dollars it is astra the I do not use that too it was important for me one you know we offer you a model and then different levels very high
the reflection but where to know where it stops it starts to create no well for an application if you have to take the model of more than 8 because if you get out of this paradigm in fact you you find yourself with a banking thing all the time that you are constantly forced to recalibrate recalibrate recalibrate no the most powerful model it will really build in total autonomy and sub -agents and with an amazing cleanliness it will even overtake you extraordinary so very
expensive I had a surplus of 2000 dollars what there is a moment it blocks you how two two days of waiting, it will be impossible, I'm too much in the fire, I'm going to lose my, no, boom, credit, credit, credit, so good. If ever there is OPDY who allows you to reset the limits. Ah, but I already reset them all, all, all, and they made me crazy gifts, I got 50 ,000 credits at the start, I don't even know why, Why? I got a mail, yes, no, no, no, not understood. No, it's a crazy thing.
So I started with a lot of goals. As I told you, I started messy.
No, but me too, at work, hyper -structured and all, we are obliged. There, no, the goal is to be messy.
What is he going to do to get organized? How? Because I'm going to ask him, on the other hand, to be super organized.
Attention, each agent has his role, etc. But it's not up to me to tell him. It's up to him to understand the context and what is his power of reflection. So, that's it. It's extraordinary.
Thank you all.