Learn DeepSeek-R1 in 30 Minutes: Watch BEFORE It's TOO LATE!

there is one model which has taken the entire artificial intelligence Space by storm it was launched just 10 days ago and it has already become the number one app on App Store and everyone is talking about it Sam Alman from open AI is freaking out about it Nvidia stocks are plummeting and the entire space is talking about deep seek R1 from China hi I'm isan and this video is a in-depth guide about deep seek R1 I'll first talk about what is deep seek R1 why is it better than chat gbt then we'll understand how to use deep seek in its various forms and then we'll also talk about what can you potentially build with deep seek R1 and to be honest the possibilities are Limitless make sure you watch till the end hit the like button and subscribe and let's first talk about deep seek R1 the model first up what is deep seek R1 so deep seek R1 is a reasoning model which was recently launched from China and it has quickly become the most favorite model for everyone to use today if you're using child GPT everyday you have to check out deep SE carvan because of fun big difference now along with this model there was a research paper published about how deep C carbon was trained and the one word is reinforcement learning while open ey trained chat GPD through supervised finetuning in which all the data is labeled and the model is about to pick up patterns in all of that data and become a better large language model in general but in the case of deep C carvan it was trained using reinforcement learning but what is reinforcement learning imagine a newborn baby who's trying to take their first steps now instead of knowing how to do it it just tries a bunch of things out figures out what is the best way to St stand and stands for the first time in his or her life think about you learning to ride the bicycle for the first time you don't really know what muscle to move or how to move it you just try it out yourself and you figure it out and you're able to then in a week or so able to ride a bicycle that is the idea of reinforcement learning all the self-driving vehicles are able to use reinforcement learning to understand how to become a better driver all the humanoid robots that you might see who are able to walk are using reinforcement learning to learn how to walk effectively or do certain actions it's all based on a agent trying out different things and figuring out what can give him the most incentives and whatever action helps it increase the score it goes in that direction and does that even more but what this really means is whenever you ask a question to DEC caran it will think multiple times about the solution it will first come up with the answer it will then reevaluate its answer thinking of ways in which this answer can be wrong and then come up with a better answer over time so with the help of reinforcement learning deep seek is able to evaluate its answer over time and get better at it compared to other models that is what makes it unique and with the help of distillation which is basically a process in which a bigger model can be transferring its abilities to a smaller parameter model you have access to it anyone can use it even on a smaller laptop with lesser specifications and I'll show you how to do that in this video but let's talk about what makes it better than Char GPT like if you're you're already using Char GPT why do you need to use deep seek for a couple of reasons number one it's completely free and offers onp par performance with o1 which is a reasoning model by open aai which costs over $200 per month first reason second reason DEC car1 shows its thinking process to you which chat gp01 hides so you're not able to look at how is this model thinking this is important because if you want to learn about something it's really good if you can look at the process in which the large language model is thinking about it right that makes it unique in a previous video I talked about how you can learn to code by simply looking at the thinking process when you ask DC car1 to build something for you number three DC car1 is completely open source and you can run it locally on your own device as well and that makes it amazing because now you can take that model and you can run it offline and you can train it on specific data and then it would work on that particular context itself another benefit of using a reasoning model like deep C car1 is that you don't need to master prompt engineering while previously using chat GPT you might have to specify everything about what you want you will have to write lines upon lines and paragraphs for it to come up with a right answer in the case of DEC car1 you can simply write one line and because it evaluates and re-evaluates its answer the answer will be much better you don't have to worry about how in-depth your prompt is because it will automatically optimize its answer in the process of coming up with the answer itself and last but not the least the API cost of using DC car1 is 27 times cheaper than that of 01 by open Ai and that is enough incentive for all the companies out there to be switching from open AI models to deep seek model to save up on a lot of cost 27 times is insane to think about and that is the benefit of deep seek R1 and that is why it's making headlines around the world with that being said there are some shortcomings of using deep SE carbon which you should know about the first thing is whenever you query something whenever you give it some personal data at especially on the chat feature always know that that data is going to be stored in People's Republic of China which is generally not seen as a safe place to store all of your data that is why India banned Tik Tok That's Why Us is trying to ban Tik Tok and in general having your personal data being stored in servers situated in People's Republic of China is generally not seen as a safe thing number two it's not great with creative writing because it's a reasoning model it's best with logic with math with science questions with coding questions and that is what it optimizes over you might be better off using clawed Sonet for better creative writing purposes and one last problem with DC car1 is it censorship problem I'll show that to you in some time but essentially if you ask it anything about China or the government it will censor that information and will say I cannot answer this question especially on the chat version if you are able to do it locally then you can ask questions to it and it will answer it to you properly but these are three big problems with deeps car1 that you should always keep in mind enough of me talking let's actually go on the system let me show you how to run deep seek online offline and how can you build anything you want with deeps R1 all right I am on chat. deep seek. com this is basically the chatbot that you can use of deep seek just like chat GPD to ask anything that you want you can just go to chat.

deep seek. com make your free account sign in with Google and this is the interface that you will first things first you have deep think R1 now this is basically the R1 reasoning model that we' have enabled so you have to enable this and if you want to search the web in real time then you can take this box as well so I can ask it any math question that I want to for example what's the differential of 1 by X and it will give me proper thinking and the answer as well so the thing that makes it unique is that it will now think about the question and the answer and will re consider the answer if it is the right answer or not and will come up with the final answer for me this is brilliant because if you want to learn anything new in math in science or in coding you can look at the thinking process and understand how to Think Through the problem step by step so this is called Chain of Thought thinking in which basically this model is going to break down this problem in multiple steps and will show you its process for arriving at the solution and not just coming to it directly and that is why I love deep seek for learning learning anything new that I want so if you go over here you can see its process first I need to find the derivative 1 upon x with respect to X let me recall the rules for derivative the derivative is x^ n the derivative of x ^ n is NX n minus1 so it will do that calculation over here and then it will think of a different way to answer the question so the thing is another way to think about it is using limits derivative is the limit as H approaches and the function itself and then it will solve this using the limits approach and then it will again think about it wait let me make sure I didn't mix up any signs the derivative of 1 upon X is negative which makes sense because as X increases 1 upon X decreases so it's thinking a lot about this problem and it again thinks about it just to double check maybe plug in a value let's say xal to 2 then f of x = 1x2 derivative of x = 2 should be -1 upon 2^ 2 which is - 1x4 and it is going to do all of this right in front of you and then after trying out all of these approaches it comes up with the answer which is going to be -1 upon x² DX and then if you go down you can see the actual answer of this question that you've entered but more than this answer this thinking process is what you should always look at and that's why I love it let's say you want to ask it any question question around science what will happen if I start racing at the speed of light for example and explain it to me like I'm five and look at the thinking process it it takes some time because of so much load on the Deep seek server it doesn't always give you the answer but it's free so you cannot complain about that okay after a few attempts it is able to answer the question it thought for 22 seconds about it and this is the thinking process for a scientific question it talks about time dilation it talks about length contraction it talks about energy requirements it talks about causality all of these scientific you know terms and keyword it is talking about it is talking about photons traveling at the speed of light and then it is concluding putting it all together explaining according to Einstein's theories impossible for anything with mass to reach light speed because it would re re infinite energy and then it explains you the answer that's one example of a scientific question and now let me show you how can you code using deep C car1 so I just gave it a prompt to build a single person space arcade game in Python I didn't give it any other instructions and it just thought about it automatically so it created lists of things that it has to do so it has to initialize the P game window it has to create the player Sprite it has to create enemies that will spawn automatically it has to solve for Collision detection it has to figure out when does the the game really end and it just creates all of that code on its own so this is the code using py game in Python then I realize that I'm not going to be able to run this in this browser itself so I need to use JavaScript instead of python so I just said modify this code to use simpler shapes instead of images and I want to run this code on deeps chat itself to see how well it works so then it is able to remove images and just use solid objects and then I ask it to rewrite this code in JavaScript as you can see right here and then it is able to rewrite the entire code and modify it according to JavaScript and its syntax so it does all of that and then I ask it to run HTML one thing you will find out in this game is that you're able to move but you're not able to fire when I press space bar then I asked it I'm anable to shoot by pressing space bar check the code and fix it so it goes through the entire code base figures out what is the problem realizes that it has to do something with pressing the space bar and detecting that and then it fixes that changes it to a different code base and now we have this code with us let's see if it works now so I press red I am able to play the game and able to fire bullets and this is basically the game built just by using deep seek R1 this is just an example obviously you can build much better things with it as well you can ask it to make the game a bit slower and maybe adjust the difficulty level of the game and that's how it will work so you have the score you are able to go around and this is the game built entirely in JavaScript using deep seek R1 you can also ask deep seek to create simulations so in this case I'm asking it to create a rotating triangle with a free flowing red ball inside it in JavaScript so I'm basically wanting a rotating triangle and it should have a ball inside of it so it thinks about the problem comes up with steps to solve it and then writes the code for it over here as you can see it thinks a lot about it it almost thought for like what 124 seconds that is insane let's just go down and see what it wrote and if it really works this is the code for it written in JavaScript I'll try running it and as you can see it's not working the ball just comes out of it so then I ask it to change the code the ball went outside the triangle as soon as I ran the code fix it the ball must always remain inside the rotating triangle so it thought about it it figured out how to solve it rote the code again and then if I press on run the code as you can see at least it is better it is better than before I can keep asking it to refine the code if I want to make it better but this is an example of how can you create simulations inside of deep seek using just plain text in JavaScript or any other language that you want to you can also do some funny things with deep seek I just sent it a hey with multiple V and it thought a lot about it it thought for three seconds before saying anything and it says okay the user sent hey with a bunch of VI they seem excited or enthusiastic I should reply in a in a friendly and upbeat way to match their energy maybe use some emojis to keep it casual hey there with a emoji and then it answers you gives you this answer also let me show you the censorship problem so if you go over here and ask it about anything related to China it will not give you an answer for it so I asked it why does the Chinese Army keep attacking the Indian army and try to invade parts of Kashmir this land is owned by India is what I gave it and it first thought about it for a couple of seconds then it completely stopped it and said sorry I'm not sure how to approach this type of question yet let's chat about math coding and logic problems instead that's how it works then I asked it some more questions for example how many slaves were used in the construction of Great Wall of China and it gave me the answer for that for some reason so that's basically how it works there are some problems with it you cannot ask it anything about China because then it will start to censor a lot of the answers now let's talk about how to run deep C car1 locally in your own computer and this is really fun so all you have to do is to just go to ol Lama just go to ama. com and download AMA on your own system so this is basically going to allow you to run any open source model locally in your own browser in your own system so you'll download for Mac OS it'll take some time and once you're done with that you can basically just unzip it and install it as an app in your system and then I'll just click on open and then I can just open up AMA on my system itself move to Applications I can click on next and I can just install this command line prompt called AMA I can enter my password and then I can run my first model now I don't want to run Lama 3.

2 I want to run deeps car1 so let's not click on this let's just go to models on the AMA website and go to deeps R1 so this is the page where you can download deeps r1's distilled models now deep seek R1 the original model is a 671 billion parameter model which is 404 GBS that is huge that's a big model but you don't need to use that yourself you can very easily use the 7 billion parameter the 8 billion parameter model which is more easier for your system to handle essentially this is a process in which the capability of a big model is now condensed distilled into a smaller model which has lesser parameters but has almost the same capabilities so as you can see this is 1. 1 GBS the smallest model which is 1. 5 billion parameter this runs on a Raspberry Pi as well which is pretty crazy to think about but I think all of your systems can run the 8 billion parameter model to think about if it can run on your system or not just multiply the size of this model which is 4.

9 GBS by 1. 2 and if you have that much RAM then you can probably run this system run this model on your system so that's about it I will just go ahead and click on the 8 billion parameter model we'll go down and we'll see how to install it on your system and this is the prompt for it using AMA so I'll just open up terminal and I will say AMA Run Deep seek - R1 hyphen 8B and it's going to download this model as you can see this says 4. 9 GB so it will take some time and I'll come back to you once it is downloaded and in about 5 minutes the model is finally downloaded and I can now start quering deep seek R1 the 8 billion distilled parameter version of it which is running locally on my system and I can ask any question I want from it so I can say what's bigger 8.

11 or 8. 99 just for example and you can think you can look at its thinking process and get the answer as well right in your terminal this is not using the internet this is not using the chat feature which I've showed you previously this is deep seek model host locally on your own system there are a lot of things you can do with it you can basically ask it for help and it will tell you what all you can do with it for example I can just say show info and it will tell me that it's using the 8 billion parameter model it is again the Deep SE model itself and I can also ask it to show me some templates and I can see what can I ask this assistant or I can just say that I want to clear and now my my context of this session is cleared this is how you operate deep seek on your own local system using AMA in your terminal itself now if you don't like interacting with a terminal we can change that we can have a better UI and to do that you can basically just go to anything LM and you can download for desktop this is basically like a UI that will help you play with any open source model out there so I'll just click download unverified file by the way if you're liking this video please hit the like button and subscribe I'll be making a lot more videos around AI agents around deep seek around what you can do with these AI tools and I have been doing that over the past one or two years so make sure you subscribe and stay tuned for more videos around deep seek and AI in general okay this should be downloaded and now I can basically install anything LM in my system let's just wait for some time while it copies one thing to note is that Lama is running in the background and you can see it with this sign over here so it should run in the background because that is how we'll be able to use it inside of anything LM okay so this is just one example you can also use LM studio so if you don't want to use anything LM just go to LM studio and download it for Mac it works very similarly just like I showed you with anything LM but now we'll go with anything llm and let let's just run it for the first time on this computer okay it is good to go I'll just click on get started and search LM providers so you can as you can see you have so many options you can go with openi you can go with anthropic Gemini you have nvidia's models you have hugging face but we'll go with AMA and we'll go with the model that we currently have so if you don't know what model you have you can basically just open up your terminal we'll close this one open up terminal and just say o Lama list and you can see all the models that you currently have on your system locally and basically that is all you need going back we have the 8 billion parameter model I'll just go ahead let's just keep going ahead I'll skip the survey I'll call this let's say isan this is my workspace and now I am good to go running anything llm on my system now the way this works is that I can create a new workspace call it let's say deep seek demo and I can go to the settings part of it go to chat settings and I can change the model it will be using so I'll be using AMA and I'll be using the Deep seek r18 billion parameter model and you can tweak a little with the numbers over here so this is the chat history number of previous chats that will be included in the person's response then you also have the LM temperature which is basically how creative do you want the LM response to be I generally it's showing 7 you can keep it like that you can try going with 0 4 update the workspace and now you can basically go and ask any question and Run Deep seek locally on your system using anything llm let's ask it a simple question um how many letters how many RS in the word strawberries let's ask it that question and it will again think about it but this will all happen locally so as you can see this is the thinking so I need to figure out how many RS are there in the word strawberries let me start by writing the words the letters so strawberry written out like that then it figures out one by one if this is r or not if this is r or not so that's basically the answer is three and it is correct and this is how you use deep seek inside of anything llm running locally on your system the benefit with this over terminal is that obviously it looks nicer but also you can attach any file that you might want so you can attach any file and you can ask any question about that file to deep C car1 using anything llm you can also use any of these agents that you have over here so maybe you want to do some web scraping maybe you want to summarize some document you can do all of that using anything LM and accessing deep seek R1 through that that's it this is how you use deep seek locally on your computer now one insane thing you can do is to start using deep seek with perplexity and that is basically god mode this takes the power of proarch of perplexity and combines it with deep seeks reasoning capabilities let me give you an example of how it works go to perplexity doc I will just make a simple account and now as just click on try deep seek R1 so we'll click on reasoning with R1 and now I can ask anything that is happening live and ask it to reason it to give me some insights for example my focus can be web so I'll just say search the web in real time latest news in the space of AI and I will be using r1's reasoning capabilities to ask it and analyze the best ideas to start an AI business today in 2025 press enter and you will see it first comes up with the most important news in the space of AI and then it gives that that to R1 and comes up with the most important business ideas that you can start today and this is the best thing you can do with perplexity and deep seek we can start a new chat over here new thread saying I want to come up with let's say ideas search the web in real time most important news in the space of freelancing and give me 10 viral ideas for Instagram reels that will work today in January 2025 and I'll just press enter it will find out the most important news in freelancing and will then give that to reasoning R1 model which will then come up with amazing ideas for you to create reels and that is the best part about it now with the free perplexity account you get three R1 searches every day and if you get the pro subscription of perplexity which is about $20 a month you get 500 searches every single day with deep seek R1 now the best thing about using deep seek in perplexity is that first of all your data is not getting stored in People's Republic of China it is actually stored in us storage places data centers and second important thing to understand is that there is no censorship so as I showed you before if you ask anything about China or any other topic it will censor that information here it will not because perplexity is hosting this whole model on their own datab base on their own server so they don't have to worry about any of that so it first gives me all the important news with links over here seven trending small business ideas for 2025 and I can have a look at all of these all of these data points then I can look at ideas that I can make content about this is all the data that we have right now and then viral real ideas AI took my job here's how I adapted and you can look at the references for that as well trending audio use my sha dramatic POV sound this is insane so now I have access to the caption trending audio and what to actually say in the real as well this blows my mind freelancer mental health check-in text overlay 72% of us get Ghost Ed by clients how do you cope and this is incredible which means that you can now create datab bagged content which you know will work really well I've been using it for a while and I love how you can get the benefits of realtime news and data along with reasoning this can get you the best answers in terms of research it can help you out with generating content ideas it can in general just help you stay ahead of the curve and stay informed not only you will you get the news you will also get to know what it means for you using the reasoning model of deep seek on the perplexity platform this blows my mind this is the basics of how you can use deep seek on chat how can you use it locally but there's a lot more that you can do with deep seek as well using the Deep seek API there are a couple of ways to use the Deep seek API either you directly go to platform. deep seek.

com which is not working right now so you can go to open router and you can go to deep seek R1 and click on API and just click on create API key and this will help you with using the API of deep seek through open router now what can you do with open router basically you now have access to deep seeks API and you can build anything you want with it maybe you want to build an application simply you go to cursor Ai and you can download cursor and you can just ask it to build a website for you and all the content or all the code that you want it to generate will be generated using open router which has API of deep seek so that's basically how it works so you can essentially build anything that you want with open router and cursor another thing you can actually be doing is to use deep seek for automations so what I mean by that is you can go to make. com this is a website that you can use to create automations for anything that you do normally you can get start for free but the way this works is that over here you can create a trigger basically let's say you get an email on Gmail and you want to get a reply for that email as well done for you automatically so you can basically create a trigger which is getting the first email from Gmail it takes that email it sends it to open router which is using deep seek R1 model it then creates the answer the reply for that email and then it sends out that answer on Gmail as well that you can basically send to the person whose emailed it to you that's just one example of what you can do with it the possibilities are endless as much as what you want to you can get into marketing you can get into sales you can get into operations finance and use deep seek r1's API through open router for all of these automations maybe you want to automate your customer experience by creating clever scripts using deep seek R1 you can do that over here you want to automate social media posting you can do that with make and deep seek you can I can make a separate video if you want me to do that about how to do this step by step let me know in the comment section below if you want me to do that but that's basically how you use make.