DeepSeek Complete Tutorial! (Including Run DeepSeek Locally + Web Features Explained)

Key Takeaways

Hello everyone. In the first two videos, we talked about how to use Chat GPT and Google Gemini and explored their features. Today, we're diving into Deep Seek. It's currently the most popular AI chatbot in China. I think most people first heard about DeepSeek when it blew up in January 2025 after they released their R1 model. R1 was designed to compete with Open AI's Chat GPT01 model. But here's the wild part. It only cost about onetenth as much to train. That news sent shock waves through the AI world and even made investors question how much money companies are spending on AI. The very next Monday, Nvidia's stock dropped almost 17%. So now a few months later, let's take another look at what Deepseek can actually do and how you can start using it. Just go to chat.deseek.com and log in. You'll see the main interface right away. Deepseek's interface is super clean and minimal. Right in the center, you've got your main chat window. As for features, there's Deepthink R1, search, and upload documents and images. On the left hand side, you can expand the sidebar. That's where all your past conversations are saved. You can click on any of them to jump back into an older chat. And if you want to start fresh, just hit new chat to begin a new conversation. Now, aside from using DeepSseek in the browser, you also have the option to run it locally on your own device. Once it's set up, you can use Deepseek completely offline. That means none of your data gets uploaded to DeepSeek servers, which is awesome if you're really focused on privacy. The reason this is even possible is because DeepSeek is one of the few large language models that's actually open-source. Most of the big players out there like ChatgPT, Gemini, Claude, and Grock are all closed source. With open source, the code is fully transparent, which means the community can help find and fix bugs, improve security, and even customize it however they want. If you're interested in local deployment, you can do it through a tool called Ola. But for now, I'll just focus on how to use the web version of DeepSeek. And later in the video, I'll show you step by step how to set it up locally. First off, you can just start chatting right in the main window. Now, if you don't turn on the Deep Think R1 feature, it'll default to using the DeepSeek V3 model. With everything turned off, DeepS's responses are more direct and to the point, perfect for quick lookups, casual conversations, or when you want a clear explanation of something. For example, I ask it to simply explain what quantum mechanics is. It gave me a super clear, easy to understand explanation of the term. At the bottom of the response, you'll see a few quick action buttons like copy regenerate, thumbs up or thumbs down. All right. Next, let's turn on the deep think R1 feature. So, when should you use this? Deepseek's R1 model is better suited for things like complex reasoning, multi-step explanations, or when you need a structured summary. It's also great for longer conversations where you want more consistency from one message to the next. Here's an example. I ask it to help me assess a business decision. Should I move my supplier from China to Southeast Asia to reduce risk? Since this is an open-ended question that requires thinking from different angles, it's a great use case for the R1 model. With R1 turned on, you'll notice something called chain of thought. This lets you actually see the model's reasoning process step by step almost like how a human would think through a problem. Now, Chat GPT and Gemini also have this capability. But what sets DeepSeek apart is that they've specially optimized it to make that whole thinking process more visible to the user. This kind of reasoning not only gives you a more accurate answer, it might even spark some new ideas for you as you read through it. In my case, DeepSseek broke the problem down, looked at it from different perspectives, and then gave me a conclusion at the end. Here's another example. I asked Deepseek to help me arrange a six-month weight training plan. I gave it some background info and told it how much time I could commit. That kind of planning is suitable to turn on R1 feature. Same deal. You get that chain of thought flow. It does take a little longer to generate, but the answer is way more accurate and tailored. In the end, I get a solid workout plan, plus some helpful tips to keep in mind. Now, let's move on to the search feature. I'll turn off Deep Think for now and just focus on how search works. It's pretty obvious that Deepseek's training data has its limits. So, if you're asking about the latest updates, specialized topics, or need it to pull info from multiple sources, you'll want to turn on the search feature so it can connect to the web. For example, I ask it what new tariff policies were announced in the past week. And you'll see it start searching online right away. Once it's done, it gives a nice, clean summary and even includes source links so you can doublech checkck the info. Now if you need both upto-date information and logical reasoning then turn on both search and deep think R1. For instance I asked it about global EV market share in 2024 and also wanted an analysis of why certain brands gained or lost ground. That kind of question needs some smart reasoning. So turning on the R1 model is definitely the way to go. Same deal here. Deepseek will start browsing the web and at the same time you'll still see that chain of thought process happening in the background. Now we've got a full well organized answer. And just like before, every piece of information comes with its own source link. Next up, let's talk about another feature, the upload docs and images button down here in the bottom right corner. One thing to note is that Deepseek specifically mentions it'll extract text only. So, if you upload a photo, it's not going to recognize objects or scenes very well. It's really just looking for any words in the image. As for documents, you can upload something like a long PDF and then ask DeepSeek to search for specific info inside it or even summarize the whole thing for you. Right here, I uploaded the official R1 model description file from Deepseek's own website and ask it to give me a bullet point summary of the key info. And yep, just like that, you get a quick rundown of a long document. Super helpful for saving time, cutting through the fluff, and quickly figuring out whether the file has what you're actually looking for. Okay, so that's the features available in the web version of DeepSeek. Now, let's move on to how you can run DeepSseek locally on your own device. I set things up using Ola, and here's how it works. First, head over to their website, olma.com. They offer a bunch of open-source large language models that you can run locally, including Llama 3.3s, Deepseek R1, and more. Click download, then pick your operating system. In my case, I went with Mac OS. Once you open the installer, just follow the onscreen instructions to complete the setup. Next, open up your terminal and type-V. This command checks whether Alma is installed correctly and shows you the version number. All right, looks like it's working. Now go back to the Alma website. Head to the models section and find Deepseek R1. There are several model sizes available and the biggest one is over 400 GB, which is massive. So just pick one that fits your needs and your computer's capabilities. I go with the 1.5B version, which is lighter and easier to run. Copy the code snippet they provide, paste it into your terminal, and run it. Once the download's done, you've got Deepseek running completely offline on your own machine. Here, you can just start chatting right in the terminal. I go ahead and ask it the same question I used earlier in the web version of DeepS. And yep, it still gives you that chain of thought reasoning followed by a bullet point answer and summary. Super cool. However, using DeepSseek through the terminal isn't exactly the most userfriendly experience. The interface is pretty bare bones. But don't worry, there are other tools you can use to give it a nicer front end. That's a whole topic on its own, though. So, if you're interested, I'll make a separate video just for that. All right, that wraps up our overview of DeepSeek, both the web version and the local deployment setup. Now, for some final thoughts. While DeepSeek does have the advantage of being super costefficient when it comes to training their models, that's not really something the average free user is going to feel or benefit from directly. In terms of features, DeepSeek offers deep thinks, search, and uploaded file, which are cool, but honestly, ChatGpt and Gemini also let you do all of that for free, too. What Deepseek doesn't have are some of those more advanced tools like image generation, interactive canvas, or deep research. And those missing pieces kind of make it less appealing for everyday users. But if you're prouser, like someone working in enterprise research, or you need largecale deployment, DeepSeek starts to stand out, especially because it's open-source and has way lower costs to run and customize. Anyway, that's it for today's video. If there's anything you'd like me to dive deeper into, just leave a comment down below. I'd love to hear what you're curious about. Next video, we'll be taking a look at Claude, another major AI chatbot on the scene. So, if you don't want to miss it, make sure to like, subscribe, and hit that notification bell so YouTube knows to show it to you. See you next time. Bye.