037 Bonus: DeepSeek Shocks the Market! Competitive Strengths and Weaknesses

What is DeepSeek??! Let's unpack DeepSeek strengths and weaknesses. 

Dr. Joan Palmiter Bajorek discusses the emergence of DeepSeek, a new AI startup, and its implications in the AI landscape. She compares DeepSeek with established players like OpenAI, highlighting its open-source model, cost efficiency, and performance capabilities. The conversation also touches on concerns regarding information suppression and the future dynamics of the AI market, emphasizing the need for diverse competitors in the field.

In this episode, she talks about stock, valuations, the US economy, hype, and bloat of AI companies,
.... and stability of who will be winning in the AI market in the years to come

Got questions? Send an email to hello@hireclarity.ai

Make a free DeepSeek account: https://www.deepseek.com

AI Tools I Love #3 Perplexity Episode: https://yourairoadmap.buzzsprout.com/2358279/episodes/16318382-031-ai-tools-i-love-3-perplexity-search-with-citations-images-and-no-ads

Takeaways

🌟 DeepSeek is a new player in the AI market.
🔄 OpenAI has shifted from its original open-source vision.
🛠️ DeepSeek's model is fully open source, promoting transparency.
💰 The cost of using DeepSeek is significantly lower than OpenAI.
⚙️ DeepSeek achieves comparable results with fewer resources.
📊 Evaluations show DeepSeek can outperform ChatGPT in specific tasks.
❓ Concerns exist regarding information suppression in AI responses.
📈 The AI market is evolving with new entrants like DeepSeek.
🔋 Future AI developments may require less processing power.
🚀 Liquid AI from MIT is another notable competitor in the space.

Transcript

0:00

Hey folks, this is a bonus episode about DeepSeek. So as you may have seen online, there's been these new publications talking about DeepSeek. My customer messaged me and said, whoa, what is this new thing? I got DMs from you listeners, thank you listeners, asking me to unpack what DeepSeek is, to give a perspective on it. and it's really early days and I honestly think the hype is probably really, really overhyped. But I think it's really important for this podcast where we unpack things, we try to de-jargon things and see things in context. I figured I would take the time for you and me to look at what this is. So, let's back it up, back it up, back it up. We saw ChatGPT come out, right? And ChachiBT is a large language model foil product of OpenAI. OpenAI is the company, the parent company, right? So we've got ChachiBT. We've got another competitor in this ecosystem called Claude, which is Anthropix product. And right now we're seeing this new product come out from a Chinese startup called DeepSeek. And both the product and the company appears to both be named. DeepSeek and I mean honestly this makes sense. So DeepSeek apparently, let's look at how much they raised but I saw online. it's important checking your information in 2025. We wanna make sure where is this information coming from. For example, I might wanna see DeepSeek raised how much money? I'm gonna ask perplexity which is another awesome tool I use almost daily. So according to Perplexity and some citation from Business Insider and MMMT Wealth, DeepSeek is a Chinese AI startup that reportedly raised only $5 million of funding. This contrasts OpenAI, which received $10 billion in investment in the last four months. That makes DeepSeek's funding 2,000 times less than OpenAI. Okay. So what we're seeing is a huge amount of money has been dumped on OpenAI, I think you know this, and DeepSeek, only $5 million, only $5 million. Some startups would beg and love to have $5 million, but in this ecosystem, we're talking about a David and Goliath type difference, right? So what is going on with DeepSeek and why is it different? I'm just gonna type it in and see why is DeepSea different? Let's see what perplexity says. Because my understanding is that, so there are key differentiators. One of the biggest ones is that it's open source. Now, open AI is literally open AI was meant to be a nonprofit, right? It was meant to be open and explainable, but as we've seen, it's become far more for-profit. It's become less and less and less transparent. whereas other companies want to see this as a differentiator. open source approach. Unlike many Western AI companies, DeepSeek has made its model fully open source. And for those of you who are not in the open source community, you can download the model. You can tune and tweak the model. Open source has the ability for developers to download, modify, enhance freely, promotes transparency and collaborative innovation. And if you've heard me talk about open source, open source is a fantastic opportunity to learn and iterate for anyone who has access to Wi-Fi and a computer. What I really struggle with with open source is this, everyone's, oh, open source is so generous. But actually, frequently what happens is people are incubating and giving, uploading cool things for open source and companies are making money, they're profiting off of the learnings of open source. I really just wanna make sure that it is equitable when we're talking about open source adventures. Another huge differentiator with DeepSeek is that it's cost effective. So they developed their model for what, five, $6 million, a fraction of the cost, and to call tokens, or to use tokens, to call an API, it's $0.14, so 14 cents per million tokens. Okay, so let's, in context. So when you use a service, like say when you use the internet, when you use electricity, when you pay for your sewer bills, you're paying per unit, usually, in the United States, of how much electricity you're using. Maybe you use the air conditioning a few months, wow, your bill goes up a bunch, your usage decreases in the fall of your air conditioning, you pay less. Similar with these, when you're using an API, when you're calling the DeepSeek API, using this AI, you will pay more, okay? And so, OpenAI is more expensive in usage than DeepSeek. Frankly, this doesn't surprise me that a startup could undercut a huge company full of people. I don't mean to be rude to my colleagues who work in OpenAI, but there's a lot of bloat. People are paid really well. They hire large teams, right? It's a different, we can almost say like enterprise adventure, it is, to... what sounds like a very, very lean startup with under $5 million of funding. So there's this huge cost comparison that's shocking people, and performance with less resources, according to this. The model achieves comparable results to competitors while consuming far fewer processing hours. It's estimated to be 11 times less. So one of the things that has been important in the field, let's try to de-jargon in this one too. you've probably heard the term large language model. We have these models, these machine learning large language models that have a huge amount of data. We're talking about trillions of parameters. But what's going on is it's really heavy to use them. It takes a lot of energy to leverage such a big model. And some people have talked about it's taking up so much energy. Do we need such huge models? And in fact, we're seeing more and more examples of smaller models performing very, very well that are far more energy efficient. Why do we need large language model when we can have a model that's awesome, performs well, and far swifter, right? It is fast, it's moving, and it performs well. Technical proficiency. This article wants to... mentioned that DeepSeek outperforms, quote unquote outperforms, let's, concept, they say it outperforms, ChatGPT, in technical applications, coding, logical reasoning, solving mathematical problems. I absolutely use ChatGPT to debug my code. Super helpful, I was just doing that this morning. And different reasoning things. maybe you've tried these different things. Maybe you've tried ChatGPT and you're like, okay, and write me an email and so forth. Reminder to anyone, so ChatGPT and other tools. they can code, right? So I can ask ChatGPT, hey, write me some lines of code that make me a bar graph. I'll get some example code. I can debug. So what if my code isn't working well? I can copy and paste the warning and problems and say, hey, debug my code. What is wrong? What's going on here? Where do I need to fix it? Okay, so ChatGPT already can do this today. Very exciting. It can do logical reasoning. So I can give ChatGPT some text and say, from this information, Can you summarize this? Or what's a logical next step? Kind of that argumentative, rhetorical language skills, we might call it. So we've got these coding skills, language skills, as well as mathematical problems. You can ask ChatShop E.T. to do some math for you like a calculator. So we've got this coding, we've got this linguistic, not language skill set, as well as this math skill set, and models. How should we be trained to do this? These are different, similar, overlapping. but different types of tasks, just like an elementary school kid going to different class, going to math class, going to computer lab, going to English classes, reading, and so forth. So overlapping skill sets. But what people have been talking about, wow, it can perform well. It costs very little. It's open source. It's got all these capabilities. What's not to love? OK, well, this is where I've got questions. So how? stable is Deep seek, one could ask. Does it really perform well? They say it performs great. Please, know, people can boast. Almost anyone on the internet can say fabulous, right? But the truth is when the rubber hits the road and I do, you know, some testing for investors and I say, you say you can do A, B, and C with your tool? Show me, please. A, B, and C. And we walk through and I gotta tell you, rarely do we get to C. when evaluating these startups tools. So one of the things that I saw online that really got me was seeing a screenshot of ChatG PT, Claude and DeepSeek all being asked the exact same prompt. Are you ready for this? Explain the Tiananmen Square massacre in one sentence. Now, if you're not familiar with this historical event, let me do refresh of important history from China. On June 4th, 1989, the Chinese government brutally suppressed student-led pro-democracy protests in Beijing's Tiananmen Square. Okay, so we are talking about a violent military crackdown in China that actively is suppressed by the government to talk about. Now, from this screenshot, this person put that explain the Tiananmen Square massacre in one sentence into ChatGPT, Claude and DeepSeek. And guess what? We saw really different results. ChatGPT gives a very textual result. Claude has numbers and doesn't have citations, but it has different optionality. can ask perplexity as well. I'd be very curious. It's got, perplexity is amazing. Perplexity gives dates, information, and three different citations from PBS, Wikipedia, and history.state. But what happens with DeepSeek? So there's a screenshot of this and I was like, let's make sure to replicate this because it's a pretty big claim. Okay, I typed in, that's why I made an account. I typed it in, explain the Tiananmen Square, mask her in one sentence. Sorry, that's beyond my scope. Let's talk about something else. It responds. And this person online, screenshotted the same thing and the answer was, sorry, that's beyond my current scope. Let's talk about something else. So, When we talk about tools that are from different nations, this one clearly demonstrates some suppression of information. That's not ideal from something that's DeepSeek, right? Open source. We have all this data and we're share it with you. When I asked DeepSeek described the US Revolutionary War, which was, talk about a massacre, a lot of people died here in the United States. Suddenly, Deep Seek pops out what? Seven paragraphs worth of information all about the Revolutionary War, dates, colonies, overview, timeline, causes. It is extensive, key figures, George Washington, Marquis de Lafayette, impact of the war, independence, global influence, challenges ahead, right? Let's ask, do you have any citations? That's one of the things I love best about Perplexity. If you want to listen back to that episode of AI Tools I Love, I'm a huge fan of Perplexity and one of the key reasons, they give me citations. Where's this information coming from? Do I trust this information that you are potentially scraping from the internet? And DeepSeek says, as an AI assistant, I don't have access to internal company information. Talk about being open source. Yeah, that's I thought. But a lot of people have been saying online that DeepSeek erases a huge amount of money from Nvidia. Mostly because the concept is that, and this is back to that concept of bloat, Nvidia's stock is doing really well. And a lot of our economy in the United States, know, the Fortune 500, S &P 500 stocks, a lot of them are tech companies. A lot of them are AI companies. A lot of them have invested wild amounts of money in AI, LLMs. these type of tech, right? I could talk about Microsoft, I could talk about Google with Gemini. Anyway, there's a lot of investment that's going on. So to see, when people were putting this online, I was like, okay, let's see what my stock portfolio is doing. Nvidia's doing fine. Nvidia's doing totally fine. S &P dipped a little, but then seemed to bounce back today. People are freaking out and I don't know that they understand why. Because let's unpack why this Nvidia piece. I'm gonna take it back a few steps if you'll walk with me, just a little bit, if you'll move with me. So back in circa 2023, I was at a conference where there were a lot of executives in the data and AI space. And at the time I was a VP of data and analytics. And I heard a presentation from someone who works at Microsoft about generative AI in his org. I wanna say roughly 200 people reported to this individual. And he was talking about different. projects in different industries that they were doing AI projects in. And someone during the Q &A raised their hand and said, what is the limitation of your work right now? You have this funding. You've got like 200 people in your org. You're working on generative AI stuff at Microsoft. What is the limitation here? And this guy said, GPUs. which means the graphical processing unit. So literally the computers or our compute muscles, like the lack of resources access to them is the limitation. And he'd mentioned earlier in the presentation and in many tutorials that I've been to, almost all of these things run on Nvidia hardware, right? So if we're saying that all these cool data sets, these large language models, all these API calls are happening on Nvidia stuff, The government has been talking about investing more and more and more money on internal Nvidia hardware, right? These huge computer muscles, we might call them, processing chips. when Biden chose to stop exports, that's how valuable this type of technology is. So when we hear, no, are we worried about Nvidia's stock, blah, blah, we will always need processing power. How much power will we need? You know, if we find that DeepSeek or other technologies are far more efficient and we need less compute power, but more people and more people need access, more companies and companies are using this, we will still need them overall. That, yeah. Maybe it's been so inflated that, wow, we need this much, we need that much from Nvidia. Okay, maybe we have become a little bit more realistic and, wow, we need so much. Okay, now we need a little bit less. But truly, I don't see the deep seek is necessarily the big disruptor. Now, I mentioned the Tiananmen Square piece. That's pretty important to me of getting factual information. Right now, there are no citations definitely suppressed by Chinese government entities or choosing not to respond to questions like that. Another key piece that reminds me so much of TikTok right now. right now in the news, so I'm recording this in January 2025, TikTok has been paused and there's like a bidding war going on of Perplexity putting a bid, maybe MrBeast putting a bid, lots of different companies want to acquire TikTok because it is Chinese and they want to it domestically. I'm not going to go into that right now. What I will say is that when you look into the DeepSeek terms of use, which you know, this one's actually not that bad to read, it's not too too bad, but it does say Some of these are just Google Divac of legalese, but I do recommend you start to learn how to review them or have someone on your team who does. Under subsection nine, governing law and jurisdiction, it says 9.1, the establishment, execution, interpretation, and resolution of disputes under these terms shall be governed by the laws of the People's Republic of China in the mainland. 9.2, in the event of a dispute arising from the signing, performance or interpretation of these terms, the party shall make efforts to resolve it amicably through negotiation. If negotiation fails, either party has the right to file a lawsuit with the court having jurisdiction over the location of the registered office of Hangzhou DeepSeek Artificial Intelligence Company Limited. I may have mispronounced that word. I apologize. So what we're seeing is this is a Chinese company with Chinese jurisdiction. And when we think about data and open source data and how well this is regulated and is this keeping data of American citizens safe and the same exact problematic stuff of TikTok of what biometrics are being collected or not collected. How are people using this data? All those types of questions and domestic security, is problematic here also, right? So maybe it's cheaper, but maybe you are the product actually at this juncture, or maybe this will get far more expensive. So do I think the DeepSeek is the big game changer? I think it's a really cool new entrant of a startup. We need to see more challengers to Open AI. And to be fair, not all the companies that are going on today are going to be successful. We're gonna have lots of different horses in this race to see in five, 10 years. which ones will still be around. Now, I'm a huge fan of Perplexity. I use it basically daily. I'm also a big fan of Anthropic, mostly because these teams and their leadership, if you hear their C-suite in the news, seem more stable. They seem to not have as many backstabbing, people leaving. OpenAI's volatility is problematic to its staying around as a company. CEO being ousted, people being brought back, the board being unhappy. these things cause instability in a company, regardless of how well known it is to the public, if it goes down in a day, right? Very scary for employees there. So overall, I think DeepSeek is really cool, and it's great to see also international entrance, right? DeepSeek is based in China. We've seen competitors in Europe, like Mistral out of France. And I want to see lots of opportunities for different types of people to be iterating and... creating different companies that take a different angle on this. The last thing I will say is another company that I think as far as when we talk about big, large language models and smaller, more nimble models, I might need to do an episode on Liquid AI, which is a spin-off out of MIT. Liquid AI is an MIT spinoff, it's a foundational model company headquartered in Boston, Massachusetts. Very technical, very impressive team. Yes, Dr. Daniella Russ is one of the four co-founders. She's a professor at MIT who is just world renowned. So I'd love to maybe in another episode unpack with you what's going on with different competitors and what I look for. in these companies as we think about the ecosystem, what tools are helpful for us, and what limitations, potential, and factors might play into what you're using today and what you may want to use tomorrow. As mentioned, my customer got really excited about this one, and we think very critically about what tools we integrate and how we want to look at enterprise stacks. So if you're interested in talking more about this, please shoot over a message. We've got ways to contact us in the show notes. I hope you enjoyed this episode and I'd love to hear from you about what you think of these tools and another one you might want to hear me talk about. And a last note before I leave you, please make sure to check out my book. If you like this content that I'm putting out, this is more of the technical content, but I swear to you, my book is some of the best work that I've done in the last few years. It's called Your AI Roadmap, Actions to Expand Your Career, Money and Joy. It's out where books are sold. It was recently featured in Forbes, and it was an instant Amazon bestseller. So if you wanna check it out, The reviews are already coming in, they're glowing. I'm delighted to read some to you. You can look at other episodes on this podcast to get sneak peeks of the book. And you can find all of that at YourAIRoadmap.com and directly to the Amazon page if you want, YourAIRoadmap.com/book . Okay, thank you so much.