YouTube Instagram TikTok How it works FAQ

The fastest way to turn talking videos into clean, searchable text. Free, private and no account required.

About

Built by Rishav Raj with one goal: free transcripts for everyone.

rishav@yttotranscript.com

Transcribe

YouTube to text
YouTube with timestamps
Instagram Reels
TikTok videos

Learn

How it works
FAQ
YouTube transcripts guide

More

Chrome extension
Contact
Privacy policy
Terms of service

© 2026 YTtoTranscript. Made for creators & researchers.

Not affiliated with YouTube, Instagram or TikTok. Respect creators’ copyright.

Deepseek a résolu le plus gros problème de l’IA — Transcript · YTtoTranscript

Home/Transcripts/YouTube

YouTube Transcript

Deepseek a résolu le plus gros problème de l’IA

Yassine Sdiri · 3,353 words · 17 min read · EN

Watch on YouTube

Share X LinkedIn WhatsApp

Below is the complete, readable transcript of Deepseek a résolu le plus gros problème de l’IA by Yassine Sdiri on YouTube. Read the full text, copy any part you need, or generate a transcript for any video with our free tool.

For years, the I's that everyone uses have all had the same problem. They cost billions to develop and are locked down by a handful of American companies. But recently, a Chinese AI solved this problem and nobody saw it coming. You may have already heard about it because more than a year ago, during its first

outing, it caused the biggest single-day loss in the history of the stock market. Nearly $600 billion has gone up in smoke, and it has just repeated the feat with its latest artificial intelligence model . Tatia, her name is Dipsic and she has become almost as powerful as the best GPT and cloud chat models, except that she

is open source and free. And all that power that was once the property of a few American companies, anyone can now use. I'm going to explain in 15 minutes how they did it , what it changes for you and how to take advantage of it. Now I know what you're thinking. You might

think that something so powerful, free, and Chinese... well, that smells like a trap . And it's true, there's one that's not necessarily where you expect it to be, and I'm going to show it to you at the end of this video. Before explaining how this AI is revolutionizing the market, let's see what it actually looks like.

This is the Deepsy interface. In fact, you create an account and you land directly on a very simple interface that you probably already know because it looks a lot like Chat GPT or Cloud. And in terms of use, it's exactly the same. You just have to ask him what you want. You have

three main distinct modes: fast, expert, and vision. And as its name suggests, Rapid is there to help you get an answer as quickly as possible. Next, expert mode is for the most complex tasks and finally vision mode is for providing it with images or documents and allowing it to analyze them in detail. And as an example, we're

Transcribe another video

Paste any YouTube, Instagram or TikTok link to get a free transcript.

Free · No sign-up · Unlimited

Add to Chrome — transcribe in one click

even going to ask him to create a small 3D platform game for us in order to push him to his limits and to his thinking. We'll come back to it a little later when he's finished. But the important point to note, which you may have noticed on this interface, is that you can see that he writes very, very

fast, much faster than Claude or Chat GPT, and this will be very important later in the video. Okay, so the game works great, nothing revolutionary but it was just to give you a little visual demo. To achieve this result in a single message, a single prompt that I sent him, is still quite

impressive. Of course, if you want to go even further, you simply need to ask him, refining and making modifications with him. But the really interesting detail about the interface is elsewhere. Whether on chat, GPT or cloud, you are surely familiar with this button that systematically pushes you towards the paid version. Sometimes it's towards the

€20 version and sometimes even for the €200 per month version. And of course, dipsic, you can look for that button for as long as you want. You won't find it . And it's very simple, it's because it doesn't exist. You are not even asked to pay to use this tool; dipsic is free,

completely free. So naturally, I wanted to test their brand new AI model for you. I've been using it every day for over a week now, instead of all my usual tools. And in this video, we're going to look at four points together. First , we will understand where this AI comes from and how a small Chinese lab

succeeded where no one had done before. In part two, we'll see what it's really capable of doing, but also its limitations. In three parts, how you can use it, including for tasks you wouldn't even suspect. And finally, in four, we'll see if all this is risk-free because Chinese nia is free, of course, that raises

questions. So, to understand why Dipsic is making so much noise, we first need to know where it comes from. Because Dipsic is far from being a tech giant. It's simply a small Chinese laboratory that originally wasn't even a tech company but an investment fund. So it's a small team with far fewer people and especially far less

money than the American giants. They have a huge handicap because, due to US restrictions on China, they don't even have access to the most powerful chips, the ones that run all American computers. So on paper, he was clearly starting from a losing position. Except that often it is precisely this kind of handicap that makes

a team formidable. Because when you have everything, you often fall into a comfortable routine. When you have nothing, you transcend yourself. And that's exactly what happened because a little over a year ago, they released a model that rivals the best American AI for a fraction of the price. The figures are being debated a little bit

today, but we are talking about at least 20 times less expensive than what the training of Chat GPT cost. And that's what set the American stock market on fire because overnight, everyone was asking themselves the same question. If they can do that with almost nothing, then why do others have to spend billions? And to give you a

fairly simple example that explains the reaction of the stock markets, imagine that you are building your house, you put a lot of money into it to have the most beautiful house in the neighborhood and the next day you realize that the neighbor right next door has built exactly the same house, except that he paid 10 times

less. So, not only would I think you would be shocked, but you would also be very angry with the construction company that made you pay too much for the same result. Well, that's exactly how Wall Street felt about Silicon Valley because Americans, to go faster, like to do one thing: always add more

money, more machines, more power to crush the competition. Dipsic simply couldn't do it . So, instead of all that, they were just smarter. In fact, to help you understand a little better, a classic AI, when you ask it a question, will use all its power, everything it is capable of doing, even for a very simple question.

Imagine for 2 seconds, you run a company with thousands of employees and you have a small, specific accounting question. Are you really going to call everyone into a meeting? Well, of course not. You will go directly to the accountant and the others will continue to work quietly on their side. Well, Deepsic, that's exactly what they've

put in place. They have a model that is gigantic, but for each question, it will only awaken the few specialists it really needs. All the others remain asleep. As a result, it has the power of a very large model and the speed of a small one. So naturally, he responds much faster, and that's

important. Now, that idea, they clearly didn't invent it. It existed long before. Except that before them, no one had pushed it this far. And most importantly, they did something that almost no one does in this shark-infested environment. They share almost everything. Well, not absolutely everything, it's important to be precise. The exact recipe for training the model

and the data they use, well, they're keeping that to themselves. But in all other respects, they are remarkably generous in this field. First of all, they make the model itself freely available, meaning that anyone can download it. And above all, he publishes a huge number of papers, dozens of research papers where they explain their

findings, their tips, including how to run the engine much cheaper. And that's exactly the kind of thing that American laboratories keep carefully hidden, literally like trade secrets. So now, to properly measure the impact of this approach on the market, let's imagine something. Let's take a farmer who finds a method to produce twice as much

wheat with half the water. What does he normally do? Well, he's keeping his method well hidden to himself. He will crush his competition and sell his wheat at whatever price he wants. Well, dips psic, they do the exact opposite. They publicly explain how they achieved this for free so that even their opponents can learn from it

. And that triggers two phenomena. The first is competition. Dipsic sells them a nia at a fraction of the price of the Americans. So all the others are forced to either fall in line or explain to everyone why they are 10 times more expensive. And the second one is even more interesting. Since

almost everything is published, other AI labs can read, understand and apply their findings to lower their own costs. In other words, dips psic is not just content with undercutting the prices of the biggest AI suppliers, it is also giving other laboratories the means to undercut them. And you'll see why it snowballs. Because since

they did that, other giants, particularly in China, have started to open up their models in turn. And today, Chinese I-series cards cost 4 to 10 times less than their American equivalents thanks to these famous dipsic discoveries. So now we might ask ourselves why they are doing all this? And frankly, that's pretty clever of them

. When you are the challenger in this market, making the technology free for everyone is the best way to destroy the lead of those who are ahead. And besides, even you behind this screen in your daily life, you see the result of this approach. This pressure that they have managed to create, well, it even affects the AI that you

use every day. And if we return to our example of the farmer, the price of the infinite little will decrease for everyone, even for those who have never heard of our farmer and his new method. Lia therefore cost a fortune to create and to run. It was locked up with a few and Dipsic has just

solved this problem not only for them but for an entire industry and then of course when something costs almost nothing, you can afford to give it away for free and that's where you really make history. Okay, a free AI, fine, but does it really hold its own against GPT chat and Claud? Because

free nonsense that spouts rubbish is pointless and nobody cares . So, as we saw in the demo, it's capable of doing some pretty complex things. Then, if we look at the comparisons on some of the toughest thinking tests to update the different model levels, well, we realize that it does not just follow the American I.

Occasionally, in certain tests, it even comes out on top. In practical terms, if you ask it a complicated problem, something about logic or math, instead of giving you a random answer , it will take the time to reason step by step before your eyes, exactly like the best AIs of the moment. Furthermore, like the

best models, it has an enormous memory. Its latest model, the 4 Pro version, is capable of storing up to 1 million tokens at once, which is roughly 750,000 words. Basically, it's like giving him the entire Lord of the Rings trilogy all at once and he still has room left . And within that, you

will be able to ask her for any specific information, hidden somewhere, and she will be able to find it. For example, a very specific detail, page 800, line 14. After that, like the best, can translate, write, code and analyze. And for the vast majority of what you would do with AI on a daily basis, honestly, you wouldn't

even see the difference with paid AI. Now, is she better than Chat GPT or Claude ? Well, not for me. And I'll be honest with you, on a lot of things, it's neck and neck with its AI. But on the toughest tasks, the most challenging reasoning questions or the large files to analyze, the

most expensive American models still maintain a slight lead, even more so since the release of the latest Fable 5 model from Claude or that of Chat GPT. But what I really appreciate about them is that dips say it themselves and fully own up to it. In their own report, they state clearly that they are

approximately 6 months behind the best models on the market. and a company that itself acknowledges its lag in this field. Frankly, it's rare enough to be worth highlighting. So now, let's be clear, Deepsic is clearly not crushing everyone. It's just almost as good, free, and open. And for most people, that's more than enough.

Now, practically speaking, how can you use it? Well, the simplest way, as you saw at the beginning of the video, is the website or the app on your phone. You create an account and you're good to go, for free. But there's one thing that JPT chat or close doesn't let you do, and that changes quite a lot

of things. Like Deepsic's AI, it is open source; in fact, it doesn't just live in one application; you can find it almost everywhere in other applications. There are other tools connected to other services. So you're never locked into a single supplier, and that allows you to do incredible things. For the more advanced among you, you

may have already heard of cloud code or codex. These are AI agents that don't just answer your questions. They are capable of acting, they can write code, create files, launch commands, all autonomously on your computer. In reality, code or Codex are just tools. The real engine behind it is, for example, the GPT 5.5 model or even the

Opus 4.8 model from entrepicuic. Well, with these tools, most of the time, you are able to manually change the motor and therefore you can run them with dipsic instead. Cloud code, for example, if you wish, can connect directly to DIipsic and there are free tools like Open Code which are made specifically to accept any

model. However, in that case, it is necessary to switch to usage-based billing. This is called using the API. And since dips costs a fraction of the price of Clode or GPT chat for their model, well, for the same use, you will often pay much, much less. Now, pay attention to one thing. If we start from this

principle, the Clude code agent for example, it will run well on your machine, on your computer. However , as a result, the engine, the AI model itself, runs on the cloud and most of the time, it's on Deepsy's servers. Remember this well, we'll talk about it again in 2 minutes because this is

exactly where many fall into the trap. Throughout this video, we often repeat one word: open source. And I think it's time to explain to you what that really means because there are still a lot of people who are mistaken about this. In general, people believe that open source is free. This means that you can

download LIA and run it quietly at home on your small computer. Except that's where it gets quite misleading. Because to get an idea of the power of a DIIA model, we will systematically look at the same thing, its number of parameters. Basically, the more parameters there are, the more powerful the model is. And the latest

version of Deepsic, therefore the most powerful one which rivals the best models, has exactly 1600 billion parameters, which is a lot of zeros. So yes, remember, as we said, with each question, she will only awaken a very small part of it. Except that to make it run, you still need to have those 1600

billion parameters on hand. And it's so huge that you need real professional equipment, not just your little portable MacBook. And here we're talking about a machine costing at least tens of thousands of euros. So open source doesn't necessarily mean it's always on your computer. This means that the model is simply public and anyone can

download it and offer it. And believe me, that has a huge consequence. A Chinese woman is free. The legitimate question that everyone can ask is: is my data going to China? To be clear, on the official website, the answer is yes. And Dipsy isn't even hiding it. As you just saw, it's written in black

and white in their privacy policy. It collects, processes and stores your personal data in the Republic of China, your messages, the files you send, your history, your login address. Moreover, in the United States and Australia, Zipsic has been completely banned on all government devices. And in Italy, the application was even removed from

the Apple Store or the Play Store. And at this point, you might be thinking, "Superias, all this just to tell me it's too risky to use," and that's precisely where you fall into the trap. Because the trap is believing that you only have one entry point, the official Deepsic website. Because remember, as we have seen several

times, it is open source, so anyone can host it, including trusted actors. And today, you can use exactly the same model at Amazon, Microsoft, or European hosting providers. And now, well, your data no longer goes to China at all. They remain with the hosting provider you have chosen under American or European law, not

Chinese law. And that's the real strength of open source because a closed AI like Chat GPT or Claude, you will never be able to run it yourself or even on the hosting provider of your choice. An open AI like dipsic, yes, you will always have a choice. And even if Deepsic decides to shut everything down tomorrow, you will still be able to

use it. The model is Chinese, that's true, but where you decide to send your data, well, that's up to you, and ultimately, you have total control. I'm also going to give you my frank opinion on dipsic. Is it worth it ? If you want a very powerful AI, free for writing, translating, coding, and learning on a daily basis,

then frankly yes, it's well worth it . Try it, you have nothing to lose, it's free. If you're a developer running AI on a very large scale, meaning you use it a lot, then the price clearly changes everything and can literally reduce your token bills by a factor of 10. If you're handling

sensitive personal or professional information, then be careful, think twice before pasting everything into the official website and always keep in mind where your data goes. Because ultimately, what Dipsy has proven to us is that the most powerful AI is no longer reserved for a handful of giants. It is becoming a common good and is being

commodified. And the hardest thing today is no longer even having access to AI, it's mainly knowing what to do with it. That's exactly why we created a specialized school to train the next generation of generative AI experts. You will find all the key training to master the subject and develop your skills, weekly workshops with my team

and myself, and all the technical support necessary to never get stuck. Today, there are more than 700 active experts on the market and some of our students like Nadia have already landed their first contract worth several thousand euros. If you're interested, the link is in the description. I hope you enjoyed this video and that

you better understand Deepsx's impact on the market.