Speaker 1: Oh wow, this is gonna be easier than I thought. Oh my, this is just working right away. What the fu- No, stop, nevermind. Guys, in two hours, I just created a perplexity clone that is specific for my use case, and it's actually as good. It is actually as good as perplexity for what I'm trying to do. And we're using DeepSeek, the chat GPT killer, to build this. And whether you are a beginner or you've been coding for a few years, this video, you are going to see from scratch, you're gonna see me create content-specific perplexity. Mark Zuckerberg incident with Bezos wife. And I can search the web, and it's gonna load sources, read all of the web sources. It's going to load them up right here. I can click on all of these. It's going to begin thinking. I can close the thinking if I want, but I actually like reading the thinking of the new DeepSeek model. And then it's going to generate a full response. So here it's going to generate a full response, date of the incident, January 20th, yesterday, basically. Mark Zuckerberg, key figures involved, incident Donald Trump's during 25. It is literally exactly, and it has sources, right? I can click on all these sources, right? We can go check these out. And I can click on these sources. Look, it'll take us directly to the source. And we can literally just ask follow-up questions. We can say, what happened with Elon Musk and the salute, salute thing. And we are literally generating, this is literally perplexity. And we're calling it Yap Search because we're actually building it directly into the app that I'm building, which is the cursor for writing content. The first thing that we're gonna do is we're gonna open Cursor. And I'm gonna go ahead and open a folder and we're gonna create a new folder. I'm gonna go ahead and make this in document. And I'm just gonna say new folder, and I'm gonna say DeepSeek and search. And we're gonna create this app here. We're gonna hit open. And now what we're gonna do is we're gonna press command I to open up Cursor Compose. And we can actually just say, I want you to fork and run this repo locally. We're gonna run the template that we use for all of these videos. And I have this saved right here in one command. We're just gonna run this real quick. Generate command, run command. It's forking, so it's adding the right files to the app, downloading them. We're gonna hit pop out terminal. The repository is being set up. Here we go. Now it gives us this link to local host. Let's go ahead and click on this. And there we go. We have our template ready to go. Let's go back to Cursor. And we're gonna say, I want to create a perplexity-like app that allows me to talk to the new DeepSeek Seek Reasoner model. A question and get an answer. I want to talk to DeepSeek using this model. What we're gonna do is we're going to come to DeepSeek. We're gonna go to deepseek.com and we're gonna click on this API platform. And what we're gonna do is we're gonna go to API keys and we're gonna grab a key. We're gonna grab this right here. So this is the Reasoner model. We're gonna click on this. We're gonna go back to Cursor. Here are docs. And we're just gonna paste this in just like that. Here is my API key. I'm gonna paste that in at the end before I run it. And what we're gonna do actually is we're just gonna get this your first API call just in case it needs this. We're gonna go back to Cursor and we're gonna paste that in as well. Here's my API key. So I'm about to paste my API key and run this. Restarted the server. So whenever this link pops up, I'm just gonna click on it and it should take us to the other server. Okay, so here it has this ask DeepSeek Reasoner something. Hello. It is thinking. Can I assist you today? What model are you? I have DeepSeek. Awesome. So this chat app is working. All right, so I gave it my API key and just ran the prompt and now it is coding. All right, it is done. Let's hit Accept All. Hey guys, for whatever reason, the seven minute clip that I recorded right here, for whatever reason, my screen software made it record like this. Not good at all. So I'm just gonna show you, really quickly give you a synopsis. So the model that we just added previously was the DeepSeek chat model and their new model is called Reasoner. So I was basically showing you that I wanted to get this reasoning and the reasoning is the best part about this DeepSeek model because it shows you how it's thinking. So I went to chat GPT and said, can they show me the reasoning? I wanted to find out if I could actually stream the reasoning and you can. And so I figured out how to do that and I pasted all these docs into cursor and then I basically just said, run it. And then here I am testing it and there you go. It's showing the thinking rather than just doing the normal chat. And there we go. Then I asked, what are the best APIs to use for cloning perplexity? And here I basically, I asked perplexity, how can I clone you? And then I found this one that I had never heard of before, which was called Tavoli and I asked, what is this? And so then I kept going. I went to Tavoli, got an API key and then I put the doc or then I said, how do I use Tavoli in a Next.js project to create a perplexity clone? And it told me exactly how to do it. And I copied the text, pasted it into cursor and then I pasted in and ran it. And here I am just speaking out a long prompt saying exactly what I want it to do. It's gonna search the internet, find the sources and then feed it through the reasoning model so the reasoning model can actually reason through all the sources. And the Tavoli API was actually pretty good. And then this happened. Let's go ahead and see what this looks like. So we're gonna type in, write a report on the DeepSeek reasoning model. Searching the internet for relevant information. DeepSeek is thinking. Analyzing search results. Source one, starting with source one. Source two, source three, source four. Oh wow, this is gonna be easier than I thought. Oh my, this is just working right away. What the fuck? No, stop, nevermind. Okay, it worked very well up until after the final output. It just started spitting out a bunch of like the same responses. It didn't give me that one final report at the end. Something went wrong towards the end. I have no idea what happened. So this is a really good tip on kind of figuring out the problem. We're gonna hit inspect and we're gonna see what the problem is here. We're gonna make sure that we're on console. And so this kind of shows us shit that we can send to cursor and it will help us fix. So this thinking is working, the thinking is working. And this is gonna just start going off crazy. And let's see here. Okay, okay, so this is failing hard. And what's going on here? Okay, so my input was tell me a joke. And here was this error log. Please fix this. All right, let's see if this works. Let's see if this works. We're gonna hit accept. And let's go ahead and try this again. Tell me a joke. Search the internet for relevant information. Okay, wow, it's like deciding which one it wants to use, but alternatively, universally. Okay, wow, this is thinking for a while. God damn it. Is it outputting something? Oh, it is outputting something. Something is very wrong. Afterwards, you're repeatedly outputting whatever the user inputted over and over and over again, really fast, and I can't even see the response that you're giving. Please stop doing this. Tell me latest on Google via Cling, et cetera. It's analyzing. Okay, I need to create a detailed report. Cling, major updates. Okay, okay, okay. Oh, okay, all right, I think we're onto something. Why did it only bring in one source? Oh, I know. It didn't. It brought in more than that. Okay, okay. That's good, that's fine. That is okay. This is what the output looks like when it's done. However, I want it to look like this. To start, ignore the toggles and focus and attach. Just make it look good to start like this. Then, each step should be shown on the front end. Then, each step should be shown on the front end. So, users, question, then search results, then thinking, then final results. All right, what are people saying about Deep Seek, new reasoner model, and how this affects OpenAI? All right, so there's the question. Searching the web, search results, thinking process. All right, ooh, so it's loading above. There's been issues about how you have the final output above the thinking area or the search results. I do not want that to happen at all. I want for the, like, I want it to be exactly as we have it now, except I want it to stream the final results and show it as it's loading. Right now, it's just waiting for a bit, and then it's just popping out with the final result. I want it to stream out. Okay, so it's done. Let's give this a test. Tell me latest on Jeff Bezos company in one paragraph. Oh, no, no. Oh, okay, we're fine, we're fine. Come on, baby, stream out, stream out. Yes, let's go, let's go, baby. All right, okay, so now we can change the formatting because it's super ugly. Okay, I want this, the background, light gray, and I want to make the interface look like, look like, ooh, yeah, yeah, yeah, this one. Okay, I want it to look like this in the center of the page right here, look like this, and make everything not in those weird boxes. Make the thinking in a slightly different format and put it in a different font than it is now. You're only changing the formatting here. You're not changing any of the functionality or where things load or how it works. I just want to start making it look a little bit better. Okay, there we go. All right, so now we're gonna type in one sentence on Jeff Bezos. Okay, the query goes there, search result. Okay, it's getting a little bit better. The query and the final results should not be in little boxes. They should just be, or the boxes should be the same color as the background and have no border. Okay, all right, now let's see how this looks. Let's give this a reset, please. Please tell me the latest on the Google AI tools, all of them that they are working on and how it's changed in the past six months. Okay, we have this query right here. We're gonna get rid of this. Okay, ooh, these actually show up as sources. We can literally click on these. That's pretty cool, okay. All right, so let's take some screenshots here. So this is what it looks like before. So that's what it looks like before. What is the best thing about turtles? All right, and once you enter it, it goes to the top like that. I like that, actually. What the? Actually, let's test perplexity. What is the Roman salute? All right, so this is what it looks like before, and we will press Enter. Okay, I like this dropdown. Okay, cool. So this is what it looks like after. Here are images before and after I enter the image. Let's put them in perplexity. I want you to put the question, or like the initial question, in the center of the screen, like you see here. Ignore all, add anything beneath it. We don't need to add any of that stuff. We are simply just putting it at the center, and then once I press Enter, it should just create this slight little, nice little animation, and then the top response should go to the top, and then you should load everything exactly as you do from the top of the screen down. Okay, it's done. All right, this is looking cool. What is going on with Elon? Ooh, come on, please. Nice thinking process. Let's go. We're gonna add a little spacing between these. Keep going, baby. Yes, let's fucking go. Okay, now what we're gonna do, let's just get all the little things out of the way. Now I want you to add some spacing. Between the sources and the thinking. How does perplexity look? Not that we need to copy them. What does the pardon do? All right. Oh, so it's showing sources like that. I like that more. Can you also show the sources like this instead of a list of links like you have it? Text a little smaller. What are the best models? All righty. Ooh, those are looking good, actually. Wow, this is looking really good. This was a huge improvement. Wow, okay. Detail the report. All right, this could look better for sure. Okay, so I'm gonna screenshot this. Let's bring this back into here. This looks good. Let's see. Tell me about the latest in video games. Any drama? Okay, now we have this fixed bottom bar right here. Let's see if this loads. All right, sources coming. Anytime soon. There we go. Show sources. Okay, that's pretty cool. All right, so that was awesome. Right now, after I've entered in a prompt and everything works smoothly all the way up until the final generation, everything is loaded out there and I can type another query. Instead of whatever I type replacing everything, I want to repeat the same thing and include follow-up questions, basically. All right, so this is looking good. I like this a lot better. Ooh, the nice little glimmer effect. Beautiful. Elon Musk. This is all looking good. This is all looking good. All right, so we have this report here. Research reports for creators. We do not need the exclamation point. Generate a research report. Make this a more elegant font and add a subtitle beneath this header that explains the app simply, beautifully, and elegantly. And I'm gonna make this bigger, too. Your AI-powered content research assistant. Your AI-powered content research assistant. Ooh, yeah. And then add a pill above this that says powered by the app thread. All right, that looks pretty good. We'll call this the app search. And I think we're just about done. Do more research. Do research for content in seconds so you can spend more time going viral. By the way, so I'm actually building this chatbot here because this is actually going into the app that I'm building right now. It is the cursor for writing. As you can see right here, we have our composition. We have our thread so we can record voice notes. And so anytime we get an idea, we can record it here. It'll show up here in the thread. DeepSeek is one of the most powerful models ever and that's why I really, really wanna start building with it. And so everything that we say just gets transcribed, thrown in the thread. It has an iOS app so you can always record your ideas into a YAP thread. And then once you're ready to create content, you can just very quickly generate a draft. And what you can do is you can press O1 here and what we're gonna do is we're just gonna press this YouTube by film booth, which is basically gonna take our thread and turn it into this clean output right here. And since O1 takes a little bit longer to generate, this will take a bit of time. And while that's loading, I can show you the research assistant that comes on the side here. So anytime we wanna do research or go through all of our bookmarks on Twitter, which I'll get to in a second, you can just chat here. You can say, we can actually say what bookmarks and notes have I saved on the new Deep Seek model. And it's actually going to use the agent to go through all of my bookmarks and my bookmarks look like this. And so these are the bookmarks that I've saved to this thread. And these are all the things that I found on Deep Seek. And we'd even have suggested sources based on what we're talking about here. And we can very quickly add them to the thread. Here, it just went through all of my different sources and you can see these are all my bookmarks on Twitter right here. And you can import them with one click, but describe in a paragraph, paragraph the reasoning aspect to Deep Seek. And you can ask follow up questions. And that's actually why I'm building this is we wanna build a better web agent. And we're gonna add what we just built today in this. And so if we like this response, what we can do is we can just add it to the thread. And so it actually shows up here in sources. We do research and synthesis here, we do ideation here. And then once we've done enough ideation and got all our ideas into our notes and sources, we can convert both of those using O1 into a clean format. So this is called, so if you go to app.yapthread.com, it would mean the world to me if you tried it out. I've been working on this tool for a couple months now. It's fucking awesome. I'm not gonna lie, straight up. I'll see you guys in the next video. I love you, bye.