L'Etat Britannique teste l'Intelligence Artificielle : DESASTRE !

Hello friends. New information about AI along the same lines as the previous one. I assure you I'm not just looking for information that confirms a post or something. I'm very open minded about these stories. But there is still a groundswell of interest regarding artificial intelligence that is extremely worrying. So the latest news comes from the United Kingdom. The British government. The British government, you have to know one thing, it has many flaws. The British administration in particular has many flaws, that's for sure. I criticize it quite often. On the other hand, it has world renowned expertise in technology in everything related to IT management systems. By the way, there's a standard called Light Today it's at version 4. Many people are in Lite or the ITL foundation certification. I recommend it by the way. It's a highly respected certification in the business world and you can prepare for it in literally three days. Sure, if you're really motivated in three days or if you take it slowly in 15. But overall it's a certification that looks great on a CV that opens your eyes to a certain number of organizational aspects of companies and their processes and it also allows you to acquire the vocabulary you need to understand a certain number of things. Who created was created by the British government. Okay, so it's used by millions of companies around the world and it comes from the British government. So they really have very recognized expertise in this area. So here's what they did. They conducted a study based on copyot. So they took, because they have a subscription with Microsoft, they took 1,000 licenses for artificial intelligence that goes with The Microsoft offer, that is Copilot. They took 1,000 licenses and they took about 700 volunteer employees and they took 300 others who were designated volunteers. We're talking about a thousand users in total. And then they gave them a steering committee. This was to use it in their work. And the data would be captured to see if we have a positive impact, if we have a productivity gain with the use of AI in the administration. And the result, my friends, is quite frightening. Before telling you about Leah's results and many people are asking the question will Leah take my job? All this I would like to indicate the link that is attached here below the video and in the description of our partner Kyuss. Kyuss can help people find a job or jump from one job to another. If you're fed up with your job, check out Kayuss. They are based in France. Under certain conditions their support is covered by France Travail. Contact them on my behalf. And generally the people I sent to them were Very happy. Now, what was the result of this story? Well, the result, I almost want to say, unsurprisingly, given what we know today, is that they did not see any gain in productivity. There was no increase among the people who used AI compared to people who did not use it. So I'll give you a few details. Among the people who used B, we found that the emails, well, the emails were generally of better quality, were written a little faster, there were fewer mistakes, and all that. Well, written a little faster because these are emails that fundamentally are not generated by artificial intelligence. You're not going to say, reply to this customer? No, you're going to have to give them the information. Often you have to go and get this data from a portal. You have to call to get it and all that. So it's not an email, it's not just text, it's not literature. Email contains tangible, objective elements. So whether you're connected to it or not, you have to go and get it. So that's the first point. So that's why those who responded to emails via Copilot responded a little faster than the others, lets say, but not excessively faster. Now, here's the other problem. The problem is that those who used Copilot to process information that was in Excel, they provided Excel files and all that to generate results from there. And, well, it took longer. And why? Because the results that were given by Copilot were very often either wrong or imprecise. So when you take the time to give it to Copilot, to prompt it, to tell it, to get feedback, to analyze it, to find errors and correct it, and so on, well, it takes more time than someone who hadn't used Leah, who had gone straight to a manual analysis of Excel. Same thing with everything that was PowerPoint. The PowerPoint presentations, the ones generated by Leo, were generated faster, but they had so many errors, so many mistakes that it took a long time to correct and all that. So you see that all in all, they found that there was no gain in productivity within the administration. So if you remember the last video. Well, for those who didn't watch it, I talked about an MIT study that was published in Fortune on entrepreneur on lots of very serious resources. Well, this study showed that 95% of projects implemented in companies did not provide any return on investment and were a loss. We spent a lot of money and very often it was driven by, well, how should I say, greed. We thought we were going to fire a certain number of people and replace them with AI. In the end, these People were fired and then we ended up with someone who told us nonsense. When we look at your return on investment, well, it was zero. It was even negative in reality, because we didn't even get our money back. So you see, it's something quite interesting because we have, and very often we have this in tech, we have this gap between hyper promises, we have the hype promise and all that stratospheric promise. In the end, it's not the same thing in execution. You know, I'll just give you an example. My personal experience with an AI that I canceled my subscription to Today, called Cloud. I create articles, so I needed an AI that would help me search for information, so it would Google for me, so to speak, use a lot of databases and all that to help me. So I needed an assistant somewhere and I thought Claude could help me. Well, I took out a paid subscription and here's what I found. Every time I asked him to do a search for me, he would respond with a value judgment, telling me, because I do studies and research that are contradictory, right? If an administration says something, if Macron says something, and so on, I want to verify and validate what he said. Is that true or not? Except that Claude assumes that everything said by a company by an administration, by, by. He tells you this is the truth, it is the source of truth. And then he becomes very peremptory, very aggressive, and he will not, he will not look for it as long as it was said by an authority. For him, it's full. Stop. It stops there. Well, I stopped my subscription and I hope that they will go to the authorities who defend it to try to get money, because useful from the user's point of view, it is completely useless. Search engines have this advantage, which is that I can search for anything they provide me with information I can search for. The Earth is flat. The search engine will not make value judgments. It's not that I believe the earth is flat, but you get my point. If I search for something as crazy as the Earth being flat, well, he'll provide me with websites of flat earthers with their theories and everything. Where today people allow themselves to make value judgments. Feel he'll teach you a lesson, won't he go and teach you a lesson? So and so. Ah, there's a context, for example, of cars that were made by Volkswagen and other German companies that had a system for rigging emissions. For Claude, that wouldn't have existed today. Well, of course he has it in his database, but at the time you wouldn't have been able to do research in that direction, because, no, it's forbidden. So. So there you have it. I'm telling you. And for the developments, for writing the code, I've noticed with Claude or with Chat, GPT or someone else, I've contacted him and I've noticed the same pattern. The pattern is as follows. When I develop by hand, well, I develop, I write the code, then I test it. I realize that there is a syntax error. I forgot a semicolon somewhere or something. I checked, I correct it and I continue. So I may be making slow progress, but I am making steady progress. With Leah, it looks like this. I ask her to write me some code, I explain to her what I want. She thinks for 30 seconds, she spits out a few nice blocks of code. So that's great. So I make fast progress. When I run it well, there are loads of errors. So what do I do? Since it's not my code, it's very complicated for me to correct it. For me, reading it line by line, it, understanding the logic and all that, and correcting it takes me almost as long as writing it from A to Z. So I have to give it back to her, explain the errors and ask her to correct it so it will spit out another one. I retest it, there are still errors and so on. And for example, I have attacked a code that could have taken me personally eight hours to type. And then I spend days with Leah doing the same thing. And the problem is that the more time you spend with Leah, the more you say, shouldn't I have started from scratch? The problem is that it's too late because now you've wasted hours with the AI, you have to continue by doing so. If you start from scratch, well, you've wasted your time. So from time to time, and here, and here, I agree with what the British administration has noted, which is that, yes, from time to time I can get lucky. And a piece of code that would have taken me an hour, I can get it in maybe five minutes. It can happen. Yes, it's true. I can save 55 minutes and say, wow, that's great. Unfortunately, this experience of saving a lot of time all at once is not generalizable. It means, yes, it's true, I can do it, but it's just a stroke of luck, but the next day I can find myself wasting hours of frustration on something else. And that's it. So it's not completely perfect, maybe for individuals, for certain use cases, for certain people. And of course, it's like all tech hypes. Yes, the hypes they come out with have a certain validity within a certain scope. The problem is that since they are, they want to sell as much as possible. So what do they do? They say that it's general, it should go everywhere. It should go in cars, in fridges, in microwave ovens and so on. And then everyone. The reality is that it's not that simple. That's it, my friends. So I'll leave you. I'll remind you of the link for Cus. They can really change career paths in a positive way, of course. And then we'll see you in a future video. See you soon.