Speaker 1: Now, hold on to your papers, fellow scholars, because this AI may be history in the making. We just talked about how quickly ChatGPT is improving, and it is already old, old news. I mean, goodness, a new AI model just dropped that can do wondrous things, like explaining the Pythagorean theorem, look at that beautiful presentation. All this in one shot. Now, if you ask OpenAI's paid flagship AI model to give you a bouncing ball inside a rotating triangle, well, it's not quite it. But this new AI, look at that. Incredible. It is called DeepSeek R1, and this was not just one isolated case. Look, what can it really be? When tested on a variety of benchmarks, it is really close to OpenAI's O1, which you have to pay for. But this one? I can't believe it. For a whole day of prompting, it costs next to nothing. This is crazy. But it gets better, there is a free version of the model that is available to all of us right now. And even better. Yes. There is a paper that describes exactly how it works. Wow. And what you have heard here is just a start. There is so much to talk about. Dear Fellow Scholars, this is Two Minute Papers with Dr. Károly Zsolnai-Fehér. So, how can you use it? Well, you can run it on the web, on their official website. If you do, this version can browse too, and it gets better. Ha. Look. It even comes with a free keylogger. Now, jokes aside, the model itself is also freely available, you can run it yourself at home, and in this case, there are no strings attached. You don't have to pay with your data. I have attached some links on how to do it in the video description. And it is fast. If you have a formidable, but still a consumer desktop machine, it can write to you faster than you can read. So, if you have hard math questions, it doesn't even break a sweat. Now, you see that the thinking process goes on and on here, yes, this is one of those new reasoning AIs that is allowed to think before answering, and note that it can take a while. Let me say two things on why this is a turning point in AI development. One. OpenAI has almost always been number one, often by a mile, but now, they introduced a paradigm shift with their Thinking01 system on December the 5th. An incredible jump forward, however, of course. It costs a fortune to develop, and you also have to pay to use it. Not anymore. Just a bit more than a month later. We have a free and fully open solution with an incredible paper. That is absolutely amazing. What a time to be alive. Two, these AI models are getting bigger and bigger, and you needed a big honking machine to run them. However, now, even the smallest models that even run on your phone are getting so good that unless you are doing quantum physics, you don't need the bigger ones. At some point, even the weakest model is going to be good enough for almost everyone. And they run for free, super fast, in your pocket. Just think about it. You have a genius in your pocket. If you would have shown these results to someone just 15 years ago, they wouldn't have believed it. And today, it is here, and tomorrow, most people are going to take it for granted. All this, for free, for everyone, and you can even train it to do something else or build your own systems on top of it. Unbelievable. But it gets even better. The paper reveals that they also threw away a lot of complexity from previous systems, they call what it does self-evolution. Essentially, it improves itself over time through reinforcement learning. The inputs are questions given to the AI, and the outputs are scored high if they use good structural reasoning and find the correct solution. Really simple. Elegant. Can't believe we are getting stronger and stronger intelligence with these simple and elegant techniques, good use of reward signals, and, of course, tons of compute. And if you add that another one of those systems popped up since I started making this video, it is called Kimi, and, can you believe this, this one also rivals OpenAI's flagship system. And Google DeepMind also just showcased their new AI system, and a separate paper also appeared on how to automate clicking around in user interfaces to do useful things. And with all this, the age of AI is here. And we are getting all of these assistants for free, or nearly for free. And they are getting smarter and smarter not every year, but pretty much every week. What a time to be alive. So, what would you Fellow Scholars use this for? Let me know in the comments below. To run your own experiments on an NVIDIA GPU, check out Lambda. I use it myself regularly for these videos. Hmm. Look at that. You can generate high-quality images in less than a second per image. I did a ton more of them, and paid less than a dollar for all this. Crazy. Seriously, try it out now at lambdalabs.com slash papers, or click the link in the description. Thanks for watching.