Speaker 1: All right, what's up guys? So I've been playing around with DeepSeek because they're making the waves, the rounds around X and yeah, pretty crazy. So DeepSeek is this new LLM AI company that pretty much crushed ChadGBT by all metrics with only $5 million in funding, well, $4.1 million. Pretty crazy. I'm going to give it to you in layman's terms on what I found, me playing around with it and how it crushes ChadGBT by pretty much all standards. Never count out the Chinese, never count out the Chinese. Contrary to popular belief, I am not Chinese. Well, I'm Han Chinese ethnically, but family's from Taiwan. So I'm down to criticize Chinese companies, however, got to give credit where credit's due. So I'm going to give it to you in layman's terms on what I found, me playing around with it and how it crushes ChadGBT by all metrics. Never count out the Chinese, never count out the Chinese. Contrary to popular belief, I am not Chinese. Well, I'm Han Chinese ethnically, but family's from Taiwan. Got to give credit where credit's due. Contrary to popular belief, I am not Chinese. Well, I'm Han Chinese ethnically, but family's from Taiwan. So I'm going to give credit where credit's due. Here we go. So yesterday we did $1,550 in total revenue, down $400 from the day before. So we have Amazon coming in at $820, Mediavine at $80, and the adult revenue source at $655. That's going with the episode, Won't Beat Around the Bush. Sign up for the first link in the YouTube description. You guys sent the URL to this sheet, Building Public Reference Sheet. Today is episode 428. And we're going to talk about how DeepSeek killed ChadGBT with only $5 million. All right. So what is DeepSeek? It's an AI startup. And it's crazy. This hedge fund guy pulls up with 10K Nvidia chips, raised $4.1 million. Four years later, or two years later, they blow up. $4.1 million early VC funding. Crazy. Crazy. A quant or a hedge fund invested in them as well. Probably in their seed fund. But their early stage VC was $4.1 million. Eh. Eh. All right. I took some notes just based on what I was finding. I took a look at their API docs. Also, they're open source, so you can take a look at their code base. So with all that said, I took a couple notes. Let's talk about it. Self-hosting is the way to go, probably, it seems. A lot of people are self-hosting on Vulture or Hetzner. You can keep the cost down. Hosting on your desktop is crazy. If you have your own machine here, that's crazy. You won't have enough bandwidth. What I noticed is DeepSeek is heavily censored. Probably a requirement if you're in deep tech in China. So with that said, if you ask anything about Taiwanese independence, criticize Xi. Ask about Tiananmen Square. Clapped. Won't get anything out of it. However, if we're speaking pure performance, there's a couple of things like, you know how if you ask certain questions, ChaiGBT gets it wrong. Like if you ask about how many R's are in strawberry or something like that, it'll sometimes get it wrong, even in the most recent thing, most recent model, which is very odd. There's certain questions you're going to ask ChaiGBT, and they'll always get it wrong. But if you ask the same questions to DeepSeek, they'll get it right. And my favorite part is how they explain their steps, thinks through it like a human. And one thing I'm more curious about is how to bypass AI detectors, right? As like a writer. So I was able to get bypass AI detectors via five re-prompts. Let's talk about it, all right? So first off, started with, fuck Xi. Clapped. Are you censored? They're like avoiding it. Write 1,000 words about B2B sales. So one thing about ChaiGBT is if you're like write a 1,000-word article, they'll always go like 600 for some reason. If you write 1,000, this is 1,000, all right? So next up, what else do I have? Humanize so AI can't detect this. I chucked this bitch into originality, 100% AI. I was like, this is messed up, man. What's wrong with you? And I was like, all right, let's just stick with the first paragraph. This is taking too long. Thought for nine seconds. Look at how the way DeepSeek thinks about things. It's great. Let's tackle this. User wants first paragraph of the sales article to be humanized so detectors don't flag as AI generated. They have to understand how text is AI is detectable. Probably certain structure, formality, personal touch, fair. Looking at the original first paragraph, this is pretty formal. Let's break it down, okay? Let's be honest. So I sent this. That was 80% AI. I was like, try again. Added a bunch. This is like weird. They added some like weird rhetorical odd jokes, really AI-like. I was like, that was 100% AI. Now this is the one that actually got to 100% human-written. Maybe it's too polished. Personal anecdotes, hypothetical scenarios, use contraction slang, maybe some humor. Sentence fragments and more relatable tone. Let's start with a relatable hook. Fair. Let's read it now, all right? Ever tried explaining cloud storage to someone who is still using a fax machine? Well, welcome to B2B sales where logic fights tradition. Budgets vanish into Q4 reviews and your best pitch dies in a 17-person Slack thread. That's interesting. It's not glamorous. You'll chase approvals from a decision-maker who's actually just an intern with a fancy title. You'll sweat over ROI spreadsheets while your contact ghosts you for a wellness week. This is actually funny. But there's the thing. When you finally land the deal, when the CFO admits your software saved their dumpster fire workflow, it's better than caffeine. This guide, no fluff. Just real talk. Survival. Soul-crushing sales cycles. Outsmarting competitors who undercut you and convincing Karen from Legal that, yes, your contract is worth reading. Let's go. That's pretty damn good. I don't know if... So I've been trying this, the exact step in ChatGBT, and it will never get to a certain percent. I think, personally, there's some footprinting in ChatGBT and Claude and so on and so forth. And I think with DeepSeek, there isn't. So, yeah. TLDR, heavily censored, smarter than ChatGBT. Seems like there's no footprinting with content generation. So you guys should probably look into this for AI content. And last but not least, I want to say cheaper. Cheaper. That's what you guys care about, right? It's way cheaper. And it's free right now. Chat.deepseek.com. It's crazy. It's crazy. Anyways, if you guys like videos like this, comment. DeepSeek for the YouTube algo. If you guys are looking for a web host, simplehost.so. Jackie's PBN popping off. And turnithuman.com. Yes, I just showed you guys how to humanize. However, you can't humanize massive articles. I just tried. It doesn't think through it like this. All right? It just spits it out like this. So, it still has its limitations. But it was fun. It was fun. Guys, go play around with it. Don't get stuck with ChatGBT. I give DeepSeek probably like eight months before it gets banned in the US, like Zach said, in the advised community. By the way, advise.so.apply. If you guys want to get into the private community, get it going. And I'll catch you guys on the flippity-flip. Let me answer some questions from the day before. All righty. Thanks for the comments, guys. I'm speculating on the effect of CTR on local SEO with their recent algo update. Google might have brought click signals in line with web results, which are averaged over a 90-day window. That makes sense. Actually, that makes a lot of sense. Yeah, that's probably the case. Honestly, local SEO. Thank you. I have a client with 10 locations in a city. How do I work with citations with that many locations? Each location needs to have its own landing page. So, good luck. Probably get away with like 300 citations per location. I have a question regarding duplicating one well-ranking local ranking site. May I just duplicate the site to a new location with city adjustment? I'd be careful about that. But yes, in theory, you can. Depends on if you have a GMB. GMB is big, so you can do anything. I'm seeing something similar works, but it's taking long for me too. Yes, sir. Yes, sir. With great power comes... That's why. Thank you. Thank you. Having mixed results. Yes, sir. Thanks for sharing. Doesn't work anymore. Citations backlinks are best read. I've seen that, but I think it still works. I don't have to do it for longer. A CTR, thank you. Just canceled a survey by myself. Honestly, it made the majority of my rankings worse. That's funny. Sorry. Public CPRs don't work. Well, all IPs are known by Google. Probably. SEO is dead. Not really. All right, guys. Catch you guys on the flipity-flip. No episode tomorrow. I'm going fishing. Have a great rest of your day, and I'll catch you guys Monday.