Everyday AI Podcast – An AI and ChatGPT Podcast - EP 460: DeepSeek Deep Dive - AI’s Sputnik Moment or a National Security Threat?

Episode Date: February 12, 2025

$100 million fine if your company uses DeepSeek? What happened, here? A few weeks ago, DeepSeek was the internet's darling. After grabbing international headlines and shaking the U.S. stock market to its core, it's been a shake week or two for the Chinese AI company. So what's actually happening here? Should you actually use the model? Is it safe? Is it really a SOTA open source model? Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Ask Jordan questions on DeepSeekUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:1. Popularity of DeepSeek 2. DeepSeek’s capabilities and benchmarks3. DeepSeek causing market disruptions4. Global Reactions and Controversies5. Analysis and Criticism of DeepSeekTimestamps:00:00 "DeepSeek: AI Revolution or Threat?"03:20 Daily AI news09:06 DeepSeek's Advanced AI Models09:56 US Alternatives Boost DeepSeek Legitimacy15:34 Data Security Risks with China17:02 Potential Chinese AI Ban Looms21:13 Confidence in Model and Media Blame25:26 "Data Privacy Concerns with DeepSeek"30:04 DeepSeek Model Cost Controversy33:12 DeepSeek's Costs and Legal Issues36:55 DeepSeek Bans Amid Security Leaks38:24 Deep Seek Ban on Devices42:44 Data Privacy Concerns in AI Platforms46:55 DeepSeek: Not Truly Open Source48:36 AI Podcast Success Story52:48 DeepSeek AI: Threat or Hype?Keywords:Generative AI, DeepSeek, Chinese AI company, US Senate bill, $100,000,000 fines, prison sentences, national security threat, AI Sputnik moment, Microsoft, Perplexity, AWS, AI predictions, OpenAI, o3 model, International Olympiad in Informatics, Alibaba, iPhone AI features, Apple, ByteDance, Baidu, Tencent, Elon Musk, OpenAI takeover bid, transparency, data privacy, open source, state of the art model, LLMs, reasoning models, regulatory oversight, proprietary documents.Send Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info) Start Here ▶️Not sure where to start when it comes to AI? Start with our Start Here Series. You can listen to the first drop -- Episode 691 -- or get free access to our Inner Cricle community and all episodes: StartHereSeries.com Also, here's a link to the entire series on a Spotify playlist. 

This is the Everyday AI Show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business, and everyday life. Meet Firefly AI Assistant, now live and Adobe Firefly, the all-in-one creative AI studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. Generative AI is already a black box that's kind of mysterious.

So should you be using deep seek a pretty state of the art and impressive model, but one from a Chinese AI company that has maybe more questions than answers? Sure, it has impressive benchmarks, low cost. is kind of, but not really open source. Yet a bill was recently introduced in the U.S. Senate that would levy $100 million fines for companies and $1 million fines for individuals just for using it and could even carry potential prison sentences for willful violations. So sure, DeepSeek is popular.

It's trending. Everyone's using it. everyone's talking about it. And big companies like Microsoft perplexity and AWS now have supported versions of their popular V3 and R1 models. So is this a Sputnik moment for AI or is DeepSeek a potential national security threats? We're about to dive in pretty deep today on DeepSeek answer those questions and hopefully a lot more. What's going on, y'all? My name's Jordan Wilson and welcome to Everyday AI.

This is your daily live stream podcast and free daily newsletter, helping everyday people like you and me, not just learn what's going on in the world of AI, but how we can all actually leverage that and take it to our advantage, grow our companies and careers. And I want you to be the smartest person in your company when it comes to AI. and where you can do that, aside from listening to this podcast, every single weekday when we do it. If you want to come jump on the live stream and ask questions, we do that at 7.30 a.m. Central Standard Time. But the other place you do it is on our website at your everyday AI.com.

So if you want more on today's episode, we're going to be recapping today's episode in our newsletter, as we do every single day, as well as sharing everything else that you need to be the smartest person in AI at your company. So please make sure you go check that out. while you're there, I'm not going to stop promoting this, y'all. Go listen to episodes 443 through 447 on our website. It's all for free. It's all there. Listen to it at your leisure. Listen to it on 2X.

I won't be mad, right? But go check out those episodes, our 2025 AI predictions and Roadmap series. Because guess what? Before any of this deep seek hoopla happened, I told you exactly how this was going to go down. I told you exactly how a lot of Chinese LLMs in 2025 were going to be a little more Trojan horse than actual state of the art model. So, not saying I warned y'all and told you what was going to happen, but, you know, go listen to those shows. All right. So before we get into Deep Sea, can talk about if this is AI Sputnik moment or if it is a national security disaster.

Let's first go over the AI news as we do every single day. All right. So, O3, the O3 model from Open AI has achieved gold medal status at the 2024 International Olympiad in Informatics. So Open AI's new reasoning model, O3, has achieved gold medal status at the 2024 International Olympiad in Informatics, marking a significant achievement in algorithmerithic problem solving and coding. This accomplish is notable as 03 reached this level without relying on competition-specific handcrafted strategies highlighting its advanced reasoning capabilities. So, O3's performance was driven solely by scaled up reinforcement learning, setting it apart from its predecessor,

01, which use tailored strategies specifically for that competition. So the model's cold forces rating now ranks in the 99th percentile worldwide, aligning it with the top human competitors on the planet. That's a considerable improvement from 01's 93rd percentile. And that is the 01 version, the IEOI version that was fine-tuned specifically for the International Olympiad in Informatics. All right. Our next piece of AI news, Apple is reportedly partnering with Alibaba to enhance AI

features for iPhone users in China. So according to the information. Apple is collaborating with Alibaba to introduce new AI features for iPhone users in China. The partnership aims to address Apple's declining market share in China or domestic competitors like who I have surged ahead with AI integrated smartphones. So yeah, Apple couldn't roll out their Apple intelligence in China. So now they have been looking for an AI partner. So initially Apple had partnered with ByDoo for AI development, but the results did not meet Apple's standards.

Other potential partners, according to the information, considered by Apple, included Tencent, BiteDance, and DeepSeek, though Deep Seek was ultimately not chosen due to its insufficient resources and experience. The AI features developed by Apple and Alibaba are currently awaiting approval from China's cyberspace regulator. All right. And that integration of AI tools is pretty crucial for Apple, especially after a dip in iPhone sales during the holiday quarter,

which is traditionally a peak seals period. All right. Last but not least, OpenAI has reportedly not received Elon Musk's takeover bid for OpenAI. Yeah, this is apparently more than just a Twitter spat. So Open AI, according to Reuters, plans to reject, not eject out of a rocket, but reject Elon Musk's led group bid. to buy OpenAI for $97.4 billion. So the bid was reportedly sent to Open AIs outside counsel, but has not yet been formally received by the board,

leading to confusion and tension between the parties. So Elon Musk's lawyer claimed the offer was sent as a detailed four-page letter of intent to purchase Open AIs assets, but Open AI, CEO Sam Altman, dismissed the offer as a disruptive tactic. So Open AI, which was co-founded by Musk in 2015, along with Sam Altman and others, is in the process of transitioning from a nonprofit to a for-profit entity to raise $40 billion in new funding for AI model development. So Altman emphasized that Open AI is not for sale and criticize Must attempt as a move to interfere with the company's plans to transition from a nonprofit to a for-profit entity. Legal experts suggest must bid complicates the valuation of Open AIs nonprofit assets, highlighting the need for regulatory oversight to ensure fair market value.

Yeah, I don't know. Guess who's in regulatory oversight right now in the U.S. government? That would be Elon Musk. So that's going to be an interesting one to see how it plays out. All right. Enough chit-chat, y'all. Let's talk deep seek. Live stream audience, love to see y'all joining me.

Thank you for tuning in. Big Bogey Face from YouTube, Douglas, Yoschelle, Zofaya, Sandra, everyone else. Marie, Bryant, thank you for tuning in. I'm curious, are you all using Deepseek, right? Samuel here says, I downloaded the Deepseek app and played with it, but I use it a lot in perplexity. Yeah, I'm curious. Is Deepseek in your company's radar right now? Is your company using Deepseek?

using it as an individual. You know, I want to hear from you all because maybe I have a tainted point of view, but we'll get to that here. But let's just go over. Let's go over the good stuff, y'all. We're going to go over the good stuff, the bad stuff, the questionable stuff, because let's be honest, since this model launched, it has been grabbing headlines for good and bad reasons.

yet I still think even after it's been nearly three weeks since its release, I still think there are way more questions than answers, which is not normal for a state-of-the-art large language model. So here is the real good stuff about deep seek. So they have very capable models that bench well. So they have their faster deep seek V3, which is their kind of transforming. type model. And then they have their powerful R1 reasoning model, right? So again, we have to start separating these models as the big companies like OpenAI, Google, and now Deepseek are all kind of

on board with the same thing. You have your kind of quote unquote traditional transformer models. Then you have your reasoner models and, you know, great benchmarks from Deep Seek. When they came out, you know, you can make the argument that it was a top maybe two or three model in the world. probably both of them. You could have made that argument. Also, this is great for the open source community and AI development as a whole. I'm going to get to that whole open source thing. I said it's great for the open source community. All right. The other thing, because we're going to talk about a lot of the problematic or unknown things when it comes to DeepSeek, but the good thing is now there are U.S. hosted alternatives. Right. So yes, this was launched by the Chinese AI. company Deepseek, which is a relatively unknown company, right, especially, you know, here in the U.S. and other parts of the world. So there's been a lot of, a lot of questions on like, hey, should we be using it? Well, there are now U.S. hosted alternatives via companies like Perplexity, Microsoft, AWS. So in the last couple of weeks, it's gotten this boost of legitimacy

as some of the biggest tech companies here in the U.S. have offered official support for Deep Seek's models. And in these kind of updated versions, user data is processed and stored on servers in the U.S. or other secure regions, right? That's important. And they use a modified deep seek model, removing some of the more problematic features like censorship, censorship or data sharing mechanisms. All right. So Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience. Meet Firefly AI Assistant, now live in the Adobe Firefly app, the all-in-one creative AI studio.

Powered by Adobe's creative agent, Firefly AI Assistant lets you start with your vision, just describe what you want, and shape the outcome as it takes form with the Assistant. The Assistant orchestrates multi-step workflows, drawing on 60-plus pro-grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator, Premiere. Lightroom Express and more to help bring your ideas to life. You can also get started with creative skills, a growing library of pre-built workflows for common creative tasks, like batch editing photos, creating mood boards, portrait retouching, and creating social variations. Every step the assistant takes is visible so you can refine, redirect, or take over at any time. You stay in the driver's seat as the creative director.

Adobe Firefly AI assistant now in public beta. See it today at firefly.adobie.com. All right, Samuel saying our company leadership discussed DeepSeek yesterday. It's quote unquote easy relatively to host and find two. It is, but it's way more problematic than that. Because we have to talk about the real not so good stuff. All right. One of the reasons DeepSeek literally caused market mayhem here in the U.S.

well, because of a couple of things. One, they reported that the model cost $5.5 million to train. And they said the model is fully open source. Those two things are not true. All right. So two, so I would say it was actually the three things that happened all at once. Number one, it had great benchmarks. That can't be denied.

Right. So when you come out with a new model that is benching as, you know, a top three model in the world going from a relatively unknown to competing or overthrowing on certain benchmarks, state-of-the-art models from Google and Open AI. That's point number one, very powerful, capable models. The fact that it's like, oh, this is open source, not necessarily the truth. And the fact that it is $5.5. They said it was $5.5 million to train. Not necessarily true.

And we're going to talk about that. All right. So it was kind of the culmination of those three things happening all at once that literally shook the U.S. economy. We saw a roughly here in the U.S. It was temporary. And for the most part, all of those companies have mostly recovered their market cap that maybe they lost. But essentially when this happened, a lot of the big U.S. tech stocks, specifically, InVIDIA, went down because everyone started saying like, oh, apparently this deep seat company only spent $5 million. to train this state-of-the-art model.

Reportedly they didn't have access, or they reportedly said they didn't have access to the most powerful Nvidia GPU. So they're like the rest of the world, who didn't care to really investigate or read or take their time for, you know, third party and independent analysis to come out, they essentially said, wait, the world doesn't need, you know, trillions of dollars worth of Nvidia GPUs.

Right? It's not the case. All right. So let's talk about some of the other things in deep sync that weren't that good. All right. So it's not truly open source, period. I'm going to go through and break down the fact that this is not an open source model. Just like, sorry, y'all, like meta, not technically open source. I'm a big fan of meta. I talk with a lot of their staff. It's not technically open source. Neither is deep seek. And it's not even close, right? So we should probably stop calling it open source and call it an open source. and call it an open, open weights, open ask, whatever you want to call it.

It's not an open source model. All right. Also, there's no real data opt out, all right? Unless you are downloading and fine-tuning the model yourself, there's no real out. There's no real data opt-out. If you are using the service just by default, their API, using it on their website, downloading the mobile app, which, by the way, skyrocketed to the number one most downloaded app, a free app, right?

If you're using it that way, there's no such thing as a data app out. Y'all, I kid you not. I saved their old terms of service and data privacy, in terms of service and privacy policy. And then when they updated it, I compared it line by line like a human. That's why I look tired, right? I don't sleep a lot. It might actually be Tuesday in my body if I'm coming with some hot takes. All right.

So the data that you upload, unless you have downloaded this and find tuned it yourself. Or unless you are using the AWS version, the perplexity version, the Microsoft Azure version, if you are using Deepseek, right, strictly out of the box, everything you upload, literally, everything you upload gets sent straight to Deepseek. You cannot opt out. Unlike U.S. companies that offer some data protection, there's zero. As soon as you enter anything, Deepseek, as essentially owns your data. There's nothing you can do about it. If you accidentally upload proprietary documents that maybe you didn't mean to, or maybe you didn't mean to, it's not yours

anymore. It's China's. And guess what? The Chinese government can access that if they want to, right? I'm not saying everything you send automatically gets sent over to the Chinese government, but the Chinese government can access any of that data if it wants to. All right. It's also an extremely censored model, although, like we said, the versions from Microsoft, Perplexity, AWS are not as censored, but try, you know, this has been well documented. Try asking the deep seek out of the box about any human rights violations in China. It won't answer you. Right. It is already banned in multiple countries and key markets. Let's talk about this new law. There's a new law. firms, if this law were enacted, U.S. firms could face up to a hundred million dollar fines

for using the model. Individuals could face a million dollar fines just for using the model, all right? And even prison sentences for willful violations. Also, I, like, I have to be real. That actual bill that was proposed by a U.S. Senator stands like an unlawful. icicles chance on the equator of making it anywhere, right? It's not going to become a law, but there's going to be follow-up versions of actual laws here in the U.S. that are, I think, that will be way harsher than, you know, the TikTok ban that actually stood in effect for like 24 hours.

There will be some sort of Chinese AI ban. I'm not saying a blanket. ban, but there will be heavy regulations and restrictions because we're going to find out, we're going to find out whether it's sooner rather than later. Well, that's actually going on. And I'd like to give Deep Seek the benefit of the doubt, but I don't know, a lot of the reporting, you know, when there's smoke, there's probably fire. All right.

They've already run into a handful of hacks, data breaches, and controversies. Like we said, this already led to a nearly two trillion. million dollars in temporary stock losses here in the U.S. economy. Most companies have already recovered all of that. InVidia, as an example, has recovered about 80% of those losses so far. And hey, here's the other thing to maybe keep in mind that goes along with those temporary stock losses. The CEO of DeepSeek is a Chinese hedge fund co-founder.

Stating facts here, y'all. Stating facts. All right. Speaking of facts, I'm just going to read this. I'm just going to read this. This was for Yahoo Finance. This was their coverage in the week after DeepSeek was released. Ready?

So it said this week, concerns surrounding Deepseek led a major sell-off of stocks in top American tech companies, including computer chip maker, NVIDIA, which saw almost $600 billion knocked off its valuation. Deep Seek itself. is owned by Chinese hedge fund High Flyer, which uses algorithmic trading techniques to place bets on market movements. And then Liang Wen Feng, High Flyers chief executive, is also Deep Seek's chief executive. All right. So a Chinese hedge fund co-founder is the CEO of Deep Seek. and what is this hedge fund, this Chinese hedge fund high flyer known for?

It uses an algorithm to place bets on market movements. All right. I'm not going to speculate on something this big, but I will say the internet is speculating that this was market manipulation, right? that Deepseek essentially shorted a lot of the U.S. big tech stocks because this is what the hedge fund, high flyer does, right? They use trading techniques to place bets on market movements according to Yahoo Finance. I don't know.

Did you guys know that? Did you do all this research? Or did you and your company just run? to deep seek and start uploading all of your data. All right. You got to think about it. So we're going to go quick here. All right.

And I'm just bringing the facts. All right. So now I gave you the big picture point of view. I gave the big picture good and bad. And now we're going to go through what's actually happening step by step. If you're listening in, I promise, I'm going to get through these next couple of slides I have here on my screen fairly quickly. because I want you to have the whole story, right? It is so important when using AI at your company.

That large language model that you are using is more than a tool, right? You have to have the utmost confidence in the model's output. You have to have the utmost confidence in its data security. And let me just go ahead and say this. All right. before I get to my list of facts and I want to take you all chronologically through everything that's happened, let me just say this. A lot of the traditional media here in the U.S. is to blame for this. A lot of, you know, social media, you know, influencers, which a lot of people, you know,

have opined that they are being paid by certain companies to put this information out there. I do blame a lot of this on traditional media. I was an investigative reporter before, right? They blindly just went with this $5.5 million. You know, they hardly no one reported on the fact that DeepSeek was started by this hedge fund that really just bets on markets, right? People rushed, right? There's this tendency. in the AI space that we all just want to click the big easy button, right?

We are looking for the easiest, most impactful shortcut, not necessarily caring or taking the necessary steps to understand what it might mean, the bigger picture, right? Is your company going to get $500 million? No, like I said, that bill is not going to make it. But, I mean, I'm also looking at the people out there that are just blind. kindly peddling Deepseek, right? Because, oh, it's cheaper.

It's powerful. Well, okay, when it came out, yes, it was one of the most powerful and most affordable models in the world. But guess what happened after that? Open AI released updates to their 03 mini. They also released pricing updates. Google, same thing. So now, does DeepSeek have this, you know, huge lead when it comes to speed open source,

which it is not quality? Absolutely not. Absolutely not. In the age of AI in 2025, there's no such thing as an insurmountable lead. And I think when this model came out, everyone just rushed to judgment, rush to savings, rush to speed, rush to fine-tuning models, made terrible decisions. Right. So again, I actually put a post about this on LinkedIn. Go back and look.

Who do you follow in AI? Were they blindly copying and pasting like a PR, like a PR, company, Deepseek's claims, telling you, you've got to use this. It's state of the art. It's cheap. It's open source. It's not. It's not open source.

Transparency is everything. Are the U.S. tech companies the most transparent out there? No. But they're much more transparent than what we have from Deepseek. And again, we have laws here in the U.S. I know we have an international audience, all right? but let me just talk about things here in the U.S. We have laws.

We have data privacy laws, right, that protect us, you know, and you do have to put a certain level of trust in a Microsoft, in a Google, in an open AI, and an anthropic, et cetera. But the U.S. government cannot just be like, yo, Microsoft, yo, open AI, hey, Google, let me take all this data and we're just going to use it for whatever purpose we see fit. It doesn't happen here in the U.S. That is exactly what happens in China, right? And everyone is just blindly saying, oh, build on top of deep seek. It's so cheap.

It's so cheap, right? They even discounted their prices. When V3 came out, they went with V2 prices for a certain number of weeks just to get everyone using it, right? I would not even be using aside from, if it's from perplexity, Microsoft AWS, I would not be using deep seek at all. Even from other providers. That's the other thing.

maybe you use tools at your company, you know, AI tools to, I don't know, meeting summarizers, you know, writing tools, whatever it may be. Are they using deep seek? If so, if you're uploading your company files, guess what? If they are using deep seek's API, guess who also has access to all of that, all of that information, all of those company documents that you're uploading and you probably don't know. China does. Right. I'm not one, you know, I'm not trying to go on a crazy guy, China, this,

trying to that, but you are, if you are using any, and there's thousands, there's thousands of pieces of software out there that are using deep seek and you might not know because they might not tell you. You might have a model selector. So you might have a little bit of, of agency when it comes to knowing, but you might not know. A lot of companies, they don't tell you what model they're using. And if you are uploading documents, you are putting yourself at risk. So this also goes out. to the software developers out there. If you're going to use deep seek, that's fine. But you need to make it abundantly clear that you are using deep seek, right?

And if you allow users to upload files, you need to be very, very clear and very explicit about what that means. All right. All right. Now let's get into what's actually happening. So deep seek, it is the AI chatbot from hedge fund high flyer. It was launched January 20th with. with kind of open source, not really, cost efficient R1 model, which is its reasoning model.

There was 10 million plus downloads within days. It went viral globally. It was the most downloaded app in the iOS store. And they are facing massive data breaches, security flaws, and government advance. So it was launched at a high profile event in China, and it was unveiled with top government officials, talking about these performance claims and bold costs. So the training cost was touted at $5.5 million, which is not true. All right, this was disputed most recently by Google DeepMind CEO Demis Hasabas.

So, yeah, it's not true. And it's quote unquote open source model intended to challenge established U.S. systems. So why did it go viral? Well, it climbed to the number one free app. on the US App Store overnight. A lot of big names in tech said this was the Sputnik moment, right? They said this changes AI development forever, right? This means that small tech companies, you know, that maybe don't have, you know,

a billion dollars to train a model. They don't have hundreds or thousands of top tier engineers. You know, what they were saying is this means that any company with just a couple million and a couple talented people can change the world. That was false. Also, there's been a rapid global uptake driven by that low pricing. Just the low pricing alone. Again, you look at it today, R1 and V3 from Deepseek are not these, you know,

sticker shock cheap anymore because of these recent updates from Open AI and Google, although for like a week or two, they were exponentially cheaper. And that caused everyone to rush there. So yeah, if you weren't, you know, downloading Deep Seek locally, right? If you have a computer or data center or a cluster with, you know, 100 gigabytes of RAM. But it did position itself as a direct competitor to chat. And the headlines, really, the unchecked media coverage led to unchecked social media coverage, led to, I think, millions, especially here in America, millions of business leaders blindly following the deep seek whale through the dark sea.

January 27th, that's when this story really started to unfold because that's when the U.S. tech stocks essentially tanked, right? Nvidia shares fell 17% in one day, 600 billion market share erased just from Nvidia. The NASDAQ and S&P experienced sharp sell-offs. and there was some financial shock over this low-cost Chinese AI model. Let's get to that $5.5 million, right? This was in DeepSeek's research paper. So, again, there's good things.

You have to tip your cap to them, right? Deepseek did release a lot of information about their model, right? Was it all 100% accurate? I don't know. Some of the smartest people in AI are saying, no. It wasn't, or at least it wasn't telling the whole truth because so much of what has happened in the last couple of weeks was predicated on that one number, $5.5 million, right? They saying that's how much it costs to train R1, this reasoning model, which at the time,

the reasoning models were extremely expensive, right? Like I said, O3Mini and O3Mini high have changed that from Open AI. Google's Gemini 2.0 flash thinking, you know, so it was exponentially cheaper than other reasoning models at least for a week. And that's what really caused a big bulk of this. But Google deep mind CEO, Demis Hasabas, labeled these claims as exaggerated. All right. Demis is obviously one of the smartest people in the world when it comes to AI, right? He's the Google deep CEO. Google essentially invented large language models for the transformer technology. All right. So here's what he said. He was quoted as saying, ready, quote, seems to have only

reported the cost of the final training round, which is a fraction of the cost, end quote. I said that on the show. I said that on Twitter. I said that everyone, right? When they saw that 5.5 million dollars. Everyone rushed and said, oh my gosh, big tech companies don't need 100,000 GPUs from Nvidia, right? And that changed everything. Guess what? Yes, you do. You absolutely do. Don't be foolish. Don't be foolish, right? This is not some groundbreaking world changing technological advancement. It's not. It's a great model. State of the art, good benchmarks. Would I ever use it? Absolutely not, right? Unless I'm using it inside of perplexity right now. And more than anything, I'm just doing it

to test it out. I'm not going to use it. I'm not downloading the app. Absolutely not. And this raised a lot of questions on the hidden costs versus the full development expense. So you also have to look at the semi-analysis report. We shared this in the newsletter. We talked about it on our weekly AI news that matters. So the semi-analysis report, and if you don't know semi-analysis, they are an industry analyst firm. They are, I think, one of the leaders, one of the independent leaders in looking essentially at the semiconductor industry and the, essentially the technology, the hardware that powers large language models. So this is a world leading analyst firm that this is all they do, semiconductors, GPUs, AI models.

So they reported their report revealed that DeepSeaks actual all-in costs were closer to $1.6 billion in hardware. And that included 50,000 Nvidia Hopper GPUs. So there's also separate investigations going that said because there's an there's a partial export ban to certain GPUs, essentially certain AI chips going to certain parts of the world, some of those including regions in China. Right. So there's also investigations going on on like, hey, did Deepseek acquire these chips legally?

Do they have these chips? Right. Again, there is so much secrecy. And I'm just saying what has already been reported out there. Open AI reportedly has evident. Right? You want to talk about even more things. Open AI reportedly has evidence that DeepSeek distilled its models from theirs, right? Which is not technically legal according to OpenAI's terms of service.

All right. So not only is it like, nah, it didn't cost $5.5 million. And but also a lot, even Open AI reportedly, uh, has evidence that DeepSeek essentially copy their paper, right? They copy their test. they just distilled, they used Open AI's models to go ahead and do some of their training. Let's think about that. Again, these are reports.

The smoke will continue to clear and we will get in the coming months the full truth, right? So that's why a lot of these things are not my opinion. These are facts that have been reported. Open AI also said, and this was less than 24 hours ago, Open AI CEO, Sam Altman said when he was confronted about this, he said that OpenAI has, quote, no plans to sue deep seek right now, end quote. I don't know. To me, I'm like, yeah, that means you're definitely suing deep seek.

Great, great PR response there from Sam Altman, or at least opening the door when you say, hey, no plans to sue deep seek right now. but they did say that they are looking into this distillation, these distillation claims. All right. Now let's go chronological so you can understand what's happening. So January 20th, R1 was launched as quote unquote open source. All right. Then you also had that $5.5 million dollar training cost claim touted at this event.

And then there was bold claims of world changing cost efficiency. So it was kind of quiet. for like a week, right? And then January 27th, that's when we saw the market shock and the cyber attack. So essentially, the deep seek app on January 27th became the number one most downloaded free app in the U.S. overnight. At the same time, we saw all the Nvidia shares drop. A lot of the other big tech shares drop.

And then we also saw cyber attack forces. The cyber attacks forced temporary API. sign-up limits on DeepSeek. All right. And that caused a lot of regulatory blowback and privacy concerns. So January 29th, Italy opened an inquiry over DeepSeek's data collection and storage. GDPR concerns sparked immediate European scrutiny. And by January 31st, Italy banned Deepseek and Taiwan banned public agency use. All right. January 30th, the cyber firm whiz exposed an unsecured database in deepseek's code. More than one million chat logs, API keys, confidential information leaked.

Did you know that out there? Or were you just rushing to use something free, something cheap, something fast? Did you care about this when clicking that big easy button, that big red easy button, right? Did you know? There were all of these concerns. February 3rd, research revealed unencrypted transfers on 3DES. February 7th, it was reported 100% jailbreak vulnerability. That's not good.

That's not good. 100% of jailbreak attempts were successful by another cybersecurity firm. And then there was on February 8th reported a device fingerprinting risk. Yikes. All right. More global bans. Australia banned deep seek on all government devices, February 4th, February 5th, South Korea restricted usage. February 5th, also OpenAI alleged on authorized API use to distill the models. And then February 6th, so last week, U.S. agencies such as the Pentagon, NASA, and Navy have banned deep seek on official devices. Texas did the same thing.

They were the first state to do so. Now let's talk about some of the legislative pushback. So February 6th lawmakers proposed a no deep seek bill for government devices. That one is very likely. Well, I won't say very likely, but that one is likely to go through. So in the same way other countries have banned deep seek on all government devices. We will likely see that here in the U.S. Also, guess what?

Every business leader that out there, you rush to use deep seek, right? All you people spinning up your apps, right, and shoving them down our throats, right? All you random social media influencers with, you know, oh, I have 300,000, you know, followers on TikTok, you know, so everything I say about AI, you need to follow. All those people were blindly pushing deep seek.

Guess what? When these bans happen, you're not going to be able to sell your technology to any government agency to any school or university. So good luck. Good luck. If you blindly went down this deep seek, you know, hey, it's cheap, it's fast, it's open source. Good luck having a business in the future. You're going to have to revert back, period. This deep seek model will not be allowed in the near, in the medium term future. It will not be allowed on most public devices. And I wouldn't be surprised if certain states take it even a step further. All right. And then we have that new,

on February 10th, it was introduced the new U.S. bill that would find companies up to $100 million for just using any Chinese AI. It did not name deep seek by name. It more or less covered any Chinese large language bottle. Again, that bill is not going to, that proposed bill is not going to become a law. It has no chance, right? It is extremely restrictive. I used to be a political reporter. It is, it is one of the, I'll say this, one of the most restrictive proposed bills I've ever seen, right? Like, million dollar fines for individuals in potential prison time who willingly violate this and use the, no, that that won't happen. Yet still, I do think a much less stringent bill will eventually become law. And then we,

We already talked about this, but on February 10th, Google DeepMind CEO, Demis Hossibos, called deep-seek's cost claims exaggerated and did say that $5.5 million was not the total training costs like the entire world believed for two weeks. It was the final training run. And like I said, according to some of the analysis, there was probably more than a billion dollars of total cost that actually went into building the underlying technology. You can't just look at the last training run and then say, oh, this is what the model costs, not how it works. Oh, let's get to the good stuff, terms of service.

Yes, y'all. I kid you not. I copied their old terms of service and then they updated it on January 20th. Yes, I'm a dork. No, I don't sleep much. Yes, I use AI to help me with this analysis. But if you look at their old terms of service section 4.3 and, compare it to the new terms of service.

It's pretty telling, ready? So the old terms of service essentially said that data processed only, that data is processed only to provide, maintain, operate, develop, or improve the services. All right. New terms of service, section 4.3, says this. It's to block processing, or sorry, to block processing, users must quote, unquote, provide feedback. Yeah.

That's a symbolic measure. There's no effective way. If you are using, whether on the app, whether you're using their website, their API, there's no way. There's no way to block data like there is. There's enterprise controls on open AIs, chat GPT, on Claude, on Google Gemini, on Microsoft co-pilot. There are plenty of enterprise controls that, that,

you cannot send your data. So not only is there zero way in the new DeepSeek terms of service to keep your data private, anything that you upload, or any service that is using their API by default, aside from using those more secure instances via AWS, Microsoft Azure perplexity, etc. There's no way. There's no way. If you're using their services directly and uploading data, that data belongs to Deepseek

and the Chinese government at any time can just come and say, yeah, we'll take that. And we'll use it for whatever we see fit. So there's no technical opt-out of data training. So nearly all of your inputs and outputs are automatically repurposed. So any data stored in China, as you know, is subject to access by the Chinese Communist Party. Still want to use DeepSeek directly? So like I said, data sent even via the API is sent to servers in China. and the Chinese government can access it under local laws.

What are the risks? Well, surveillance, disinformation, espionage, choose, pick your poison, right? And this is viewed widely, especially in the last week after the initial shock in hoopla settled down. It is viewed as a huge national security threat by U.S. lawmakers, right? Sorry, they're not the smartest bunch for the most part. I've been covering politics for a while. It's,

I'd say it's funny, but it's actually sad, right? You can go back and listen to old testimony from U.S. lawmakers on the internet. They didn't even understand what the internet was. They said it was tubes of information, you know, like a bank, how you would stick something into tube, right? So for the most part, U.S. lawmakers have no clue how technology works and AI. Yes, there are some that do. But for the most part, if, they did, I would have already expected a real bill, right? Not a knee-slapped bill that we saw,

you know, like a $100 million fines. But I do see some sort of restrictions coming sooner rather than later in terms of using Chinese AI. It's too important to the U.S. economy, period. There's national security concerns. There's geopolitical influence that people aren't really looking at. The potential for disinformation and misinformation is rampant, the more. that models like this are used and unchecked at either the federal or state level. Let's get to the last point here, y'all. Everyone's like open source, 5.5 million, state-of-the-art model, best model in the world. No, no, no, no, and no.

DeepSeek is not open source. Let me repeat this. Deep seek is not open source. Also, yes, a lot of the models that you think are open source probably aren't open source either, right? like Lama. It's not technically open source, although they are an open source leader. And although I do think deep seek's models are great for the open source community,

they are not open source. Okay. So yes, Deepseek did kind of release its full source code and model weights provided under an MIT affiliated license, but here's why it's not open source. According to the open source. initiative in their open source AI definition. So that's OSI is the open source initiative.

That is the most reputable organization out there on open source technology. So right now with deep seek, there's restrictions on usage in specific domains and or regions which violates OSI's requirements for unrestricted use for being an actual open source model. That's number one. And then the bigger one, well, Deepseek does not provide access to its full training data sets or sufficient details to allow independent replication of its models, which again is an open source requirement. So go go check out like I did, right? Go check out and read the open source initiative OSIs. Go read their open source AI definition and you'll see, no, Deepseek is not an open source

model. It's not even close. just because, you know, they, you can download a distilled version of it. You can, you know, a 70B version or, you know, 1B version or just because you can download something and fork something does not make it an open source model. We have to stop saying that. Is it good for the open source community? Absolutely.

Is this a true open source model? No, not even close. Yet every single journalist and every single Billy Boy in his mom's basement. that is a, you know, NFT bro turned crypto bro turned AI expert who's telling you it's an open source model using it in all your projects. It's a hunt. No, it's not. I don't know.

I guess people don't like to read. I guess people don't like the facts. I guess people don't care about seeking the truth. They're just deep seeking a shortcut. All right, y'all. That was a lot. And it took me a while to do this show. I literally, and I'm not exaggerating between, you know, comments on LinkedIn, you know,

DMs, emails, you all have been asking about this for weeks, right? Hey, Jordan, when are you going to come out with a show on DeepSeek? I said, not yet, right? When I started this everyday AI thing, if I'm being honest, right? I was like, hey, this is going to be a newsletter. And I'm going to do a daily podcast, so we have fresh and exclusive insights for the newsletter. I did not anticipate that this podcast, this live stream that turns into. a podcast, I did not anticipate that it would become one of the most popular and most listened to

AI podcasts in the world, if I'm being honest, right? And over the last, you know, I've been doing this almost every single day for two more than two years now. Is that right? Gosh, I've realized that I actually have a responsibility, right? And I take that responsibility very seriously. that's why sometimes I don't do, you know, hot takes on something right when it comes out, especially when there's a lot of factors at play that the average business leader in the U.S. doesn't understand, right? Maybe I'm a little pessimistic, right? And maybe that's good.

Maybe this is my old, you know, political reporting background, investigative reporting background that comes into play here, right? But it's the same thing. Anytime a big AI company puts out a press release, I never just read it back to you and take it as truth. I test it rigorously, right? All the claims that we can, right, with our very, very small team. But I take the responsibility of this podcast very seriously, right? Because I've received a lot of messages, especially over the last year, from literally, I get messages from Fortune 500, Fortune 100.

companies saying that they made an AI decision based on something I said in the podcast, right? They're like, hey, our team listened to it. We investigated what you said and you are right. So this is what we did. So I'm not going to, you know, the day that DeepSeek comes out, I'm not going to come and come to you with a show that says Deepseek is the most amazing thing ever. It's open source. It is cheap. It is powerful.

It is safe. It is transparent because it's open source. No, I'm not going to do that. I feel a little bit of responsibility because, like I started the top of the show talking about, generative AI is already a black box, right? We need to demand answers. We need to demand transparency. Speaking of deep seek, we need to seek deeply the truth. We need to understand implications.

We also need to be cognizant to the fact. that AI is about more than AI. Right? For the first time ever, I talked about this. For the first time ever in 2024, the top six companies in the U.S. by market cap were all in the same industry, right? AI tech companies, more or less.

They were all investing in AI. That's never happened before. Okay. AI is the new world economy. And so what happens with that? This is more than, you know, typing in a prompt on a large language model. It's more than that. This is global power.

This is geopolitical influence. There's very sophisticated misinformation and disinformation campaigns being run in the AI space. This is political power. This is military power. This isn't just a chatbot anymore, right? And that's why sometimes I try to take my time. I try to educate myself. I try to be informed.

I try to ask the hard questions and think about those hard questions. So I hope today you can at least see. And to answer the question, is Deepseek AI Sputnik moment, or is it a national security threat? I'll say it's something in between. Could it be a national security threat? Seemingly so. According to some of the bills,

U.S. lawmakers in certain states have put out. It does seem like it could be a national security threat, although I do not have any firsthand information, but a lot of the reporting and some of these initial bills and laws that have been proposed would lead you to believe that, yes, it is a national security threat. Is it AI Sputnik moment?

Absolutely not. All of the tech visionaries and big VCs and everyone else that, you know, on January 20th or January 27th that made those bold claims, they just wanted to be first. It's more important to be factual than it is to be first. The stakes are high, higher than you think. That's why I hope you can appreciate the time that went into putting this show together. And I hope that as you are making decisions, right, whether you are a solopreneur trying to build something, whether you are a leader of a department at a Fortune 500 company, whether you are working in big tech. I hope you take the time to fully understand the implications and everything that went on here.

Because like I said, it's bigger than AI. All right. And this is not the first time. This will not be the last time. So I hope this was helpful. Thank you for tuning in. If you haven't already, like I said, we're going to be recapping this in our newsletter. So please go to your everyday AI.com.

Sign up for the free daily newsletter. While you're there, like I said, this was already. I already did this. I already covered this before it even happened in our 2025 AI predictions and roadmap series. So make sure while you're on our website, it's free. You can just go listen to episodes 443 through 447. Thank you for tuning in. Hope to see back tomorrow and every day for more everyday AI.

Thanks y'all. Meet Firefly AI assistant. Now live in Adobe Firefly, the Allman One Creative AI Studio. Just describe what you want to create in your own words in the assistant handles the rest, orchestrating multi-step workflows across Adobe Creative Cloud apps, including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerates execution. Stand control with the ability to step in and refine at any time.

See it today at firefly.adobie.com. And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going. For a little more AI magic, visit Your EverydayAI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.