EP 453: AI News That Matters - February 3rd, 2025

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.

Did Deep seeks groundbreaking AI model really cost only 5 ish million dollars to train? A new analysis says it costs way more. A new copyright ruling on AI dropped, but hardly no one noticed or pay attention. And OpenAI not only dropped an impressive new reasoning model in O3 mini, but they're eyeing a $40 billion capital infusion just a week after announcing the $500 billion AI infrastructure plan in Stargate. Yeah, if you thought AI news was going to slow down, you know it's the end of December was ramping up, so we're going to have a slow fourth quarter. Absolutely not. There is so much to cover and we're going to be talking about those three stories and a whole lot more today on Everyday AI. What's going on y'? All? My name is Jordan Wilson and I'm the host of Everyday AI and this thing's for you. It is your daily livestream, podcast and free daily newsletter helping us all not just learn AI but how we can keep up with all this stuff that's happening. Because it's AI is literally happening around the clock. On the weekends when you sleep, guess what's also happening? Me, our team, we are keeping you up to date because it doesn't just help you to know what's going on, you have to know what actually matters and how it's going to impact your company or your career. So that's what we do most Mondays. We're back after a week or two off with our AI news that matters. So. So we are back for the week of February 3rd and if you're new here, yeah, like I said, we do this almost every single Monday going over the top news. One other thing, if you are brand new here, maybe checking us out on the podcast. Thank you. Check your show notes for the link to our website. That's your new best friend in AI. It's your everyday AI dot com. Go sign up for our free daily newsletter there. In our daily newsletter we recap each show every single Monday through Friday as well as there's like now 400 plus back episodes. You can go listen to learn from the world's lead leading experts. All sorted by category. All right. Also gotta give one more shout out live stream audience. Did y' all stick with our our 2025 AI predictions and roadmap series? It is still highly relevant. You know a lot of things that we're gonna be talking about today predicted this two weeks ago when we released it. So make sure you go listen to those on our website. All right, without further ado, let's get into the AI news that matters for the week of February 3, 2025. So let's talk about the hot topic. All right, and FYI, y', all, I'm gonna have a dedicated show on Deep Seek later this week. I intentionally took a lot of time on this. People were reaching out. Jordan, give us your take on Deep Seek. I said, nope, it's too soon. All right, but here's one of the reasons why. Apparently, according to a new new report, Deep Seek spent maybe a little bit more than it was initially believed training their new models. So according to a new report, deepsea hardware expenditure is estimated to be significantly higher than $500 million. And that's according to a new report by semianalysis. So despite these reported high hardware costs, Deepseek AI reported that its model training costs were only about $5.5 million. Yeah, pretty stark contrast between $5.5 million and every single other state of the art large language model that's ever been trained, usually for about 10 to 20 times that cost. And obviously when Deep Seek, you know, with their both their V3 and their R1 reasoning model, when they said, hey, we only spent $5.5 million to train this model, the entire US stock market was went up in arms, lost more than $2 trillion in market cap in reaction to this news. Y', all Speaking of that 2025 AI roadmap, I warned you of this. I said this is gonna happen right before it happened. So I wouldn't pay too much attention to it. Although I think Wall street got tricked, I think journalists got tricked. I think your favorite social media AI person got tricked. Guess who didn't? Me. Yeah, go back. You know what? I'm already starting out on a hot take. Tuesday, even though it's the news that matters. Monday, go. Go see who you follow in AI and go see what they said two weeks ago about Deep Seek and then check back in with me in two weeks. You know, I'm a former journalist, so I was not surprised that a lot of large news organizations essentially just copy and pasted what Deep Seek said and did not look into it. Which is one of the reasons, among many others, that, you know, the US Stock market saw a huge swing. But the new report from Semi Analysis highlights the substantial compute power needed for generating synthetic data, a key factor in Deep Seq's reported hardware spending. So Deep Seq's R1 model has been praised for its capabilities, with experts noting its rapid advancements despite US restrictions on chip exports to China. Yeah, that's one of the biggest things. It's like you can, there's only like one chip from Nvidia that can make a state of the art model like this. And right. Reportedly Deep Seek in the their last training run, this one that cost $5.6 million, wasn't using them. Again, these are all according to reports in what Deep Seq AI itself said. So obviously at the center of all this was number one, Nvidia, right. Saw their stock price go down, I believe it was, by 17% over the course of like two days because now everyone said, oh it's so cheap and you don't need Nvidia's, you know, highest powered AI to their highest powered GPUs to train a state of the art model. More on that later this week when we go in depth into the whole Deep SEQ saga on our show. But it was probably two of the companies in the spotlight aside from Deep Seq were Nvidia and OpenAI. Right. OpenAI, by far the leader in the large language model game, reportedly spent about $100 million training their old GPT4 mod model. So they were kind of thrust in the spotlight. So OpenAI CEO Sam Altman acknowledged the quality of the Deep SEQ model and emphasized the competitive nature of the AI industry. And also concerns obviously have been raised about the US potentially lagging behind China in AI development. Another thing I told you was going to happen in 2025. We are going to see China leapfrog the US and yeah, I called a lot of these quote unquote open source AI models Trojan horses. All right, so we'll see how that actually plays out. All right, our next piece of AI news obviously related because this really shook up the entire US economy. Not just the AI, the generative AI space, the large language model players, big tech note, the US economy. But OpenAI, at first they were like, okay, good job Deep Seek. And then we saw the reporting where OpenAI is scrutinizing Deep Seq for allegedly using OpenAI's own models to distill or to help make their own model. So According to reports, OpenAI is investigating allegations that DeepSeek used OpenAI's AI models to create its competing chatbot. So that was first reported by Bloomberg. So like we Talked about, Deep Seq's chatbot caused a significant market disruption causing AI linked US tech stocks to lose about $2 trillion in market. A lot of that has since been recovered. But pretty, pretty significant anytime you see $2 trillion in market cap erased from the US economy, pretty big deal. So Deep Seat claimed that its AI model was developed at a fraction of the cost and hardware needed compared to rivals like OpenAI, Claude, Google, etc. So OpenAI's CEO Sam Altman initially expressed, expressed admiration for Deep Seq, but did later acknowledge evidence of deep distillation or a technique where smaller model, smaller models replicate larger models. Right? So you have this whole kind of like teacher student paradigm between the models where, you know, quote unquote, smaller companies are just using outputs from larger models to distill information or to take or borrow outputs from a larger model to train theirs, which is against almost every single proprietary large language models terms of service such as OpenAI. So this is, this is happening quickly all and in our daily newsletter we'll continue to keep you up to date as we have this week. But the US Navy, interestingly enough, has banned Deep Seeks apps due to potential security and ethical concerns. And the US National Security Council is evaluating the AI apps implications. So, yeah, yes, Deep Seeks, you can download it, right? It's not a true open source model. All right, I'll get into that later in the week. But it is a fairly open source model in the same way that Meta's llama. It's fairly open source. Not 100%, but yes. So people are kind of missing the point. Everyone's like, oh, you can download it, it's open source. Well, it's not really open source according to all of the rules for open, like for something to be considered truly open source. But most people are still just using it via app interface. So there's a lot of intricacies with data sharing and really just privacy concerns. Right. In the same way that, you know, the whole Tick Tock ban was being talked about for two plus years. It was banned for, I don't know, a day. All the teenagers went crazy. US President Trump reinstated Tick Tock, but they are on a shot clock now. So I think we're going to be seeing the same things. At least the app on the iPhone could be banned. Right? So it topped the US charts for free iOS apps. Right. So a lot of people, you know, downloaded Deep Seek on their phone, on their iPhones. I wouldn't be surprised if it does get banned in the US eventually. It's already been banned in other countries such as Italy. So the situation is presenting a significant challenge for Silicon Valley in the global AI Competition. Also David Sachs, Trump's AI advisor and sar, I think suggested that intellectual property theft may have occur, occurred by Deep Seq. So, you know, it's not just OpenAI and you know, reporters hinting at this, this is from now someone in Donald Trump's official, uh, kind of uh, government in, in an official government role accusing Deep Seek of distilling from OpenAI's model, which is not, uh, not legal, at least according to OpenAI's terms of service. And you might be wondering, okay, well, what's the big deal? Well, I mean, number one, most people aren't downloading this thing and running it locally, number one. Number two, data privacy and concerns. It's a lot different, you know, if you are giving your data, especially as a business, you know, to a Meta, to a Google, to a Microsoft, to a OpenAI. Right. Because at least here in the US and I know we have listeners from all over the world, but I am putting this, I live in the U.S. the majority of our listeners live in the U.S. so I am putting this through that point of view. There's a lot of huge concerns when you are essentially connecting all of your business data and uploading it and sending it to a company based in China. You really have to understand how China treats its data and its link to the government, which is not the same here in the US So more on that later in the week. All right, more Deep SEQ news. And this one gets a little weird because Microsoft has announced that it will host deep seq R1 on its Azure cloud service despite those exact accusations that we just talked about. So yes, it now looks like OpenAI is rightfully so, pointing the finger at Deep Seek. Yet Microsoft, which is one of the biggest backers of OpenAI digitally, just announced that they will be making the R1, the reasoning model from Deep Seq, available on its Azure platform. So, yeah, pretty noteworthy as it involves a major player in AI Microsoft, which like I said, is OpenAI's largest investor, now offering access to the model accused of violating OpenAI's terms of service. And not only that, but from the company that I think was maybe kind of tricking the US economy into going down. More on that later this week. Right, so Microsoft's decision to host R1 is seen as making it more legitimate in the business community, despite that ongoing dispute and the probe into data data extraction practices by Deep Seq. Also, Microsoft announced OpenAI's Zero1 Reasoning model would be made available to, to even free users in Microsoft Copilot Enhancing Copilot's ability to address complex questions with their Think Deeper feature, which was previously only available to Copilot Pro subscribers. So yes, now you do have the O1 reasoning available to even Copilot free subscribers. All right, our next piece of AI news. Yeah, a lot with OpenAI this week, y'. All. So OpenAI announced partnerships with the US government to enhance research across various fields using AI, marking a significant step in AI's role in advancing science and national security. So approximately 15,000 scientists across diverse disciplines within the US government will gain access to OpenAI's latest reasoning models, according to the announcement. So researchers from Los Alamos, Lawrence Livermore and Sandia National Labs so those are all U.S. government entities, will utilize O1 or another O Series model. All right, so on Venada, which is an Nvidia supercomputer at Los Alamos with Microsoft's support in deployment. So OpenAI also announced Chat GPT Gov, a secure AI solution designed specifically for US government agencies to use Chat GPT so agencies can self host. That's an interesting one. Agencies can self host chatgpt to manage security and compliance, reducing risks tied to handling classified information with external AI models. All right, moving on from OpenAI, at least for a couple of stories here. Google has announced the rollout of Gemini 2.0 Flash, which promises faster responses and improved performance for users on desktop and mobile platforms. So the update is currently more widely available on the desktop, with mobile users on the Google Pixel smartphone still experiencing the previous versions. So Gemini 2.0 Flash enhances tasks such as writing, brainstorming and learning by providing a smoother user experience. The model also offers improved response to image inputs, although free users can only upload images and not other file types. So previously only available to Gemini Advanced subscribers, the 2.0 flash model is now accessible to all users, providing a significant Update from the 1.0 flash model. Yeah, this is pretty big because the front end of Google Gemini up until December, they weren't really transparent in what version you were running, right? So I actually did a ton of digging, even reaching out to people at Google, and finally I essentially got a hint that, you know, up until December, for the most part the model that you were using, you know, if you go to gemini.google.com right? So it's different than if you're, you know, using Vertex or Google AI Studio on the back end. But if you were just logging in and using the Gemini chatbot, you were generally using a version of Gemini that was six to nine months outdated, which is not good. So that changed in December, but now we have the official version of 2.0 Flash, where previously we had an experimental version of the Gemini 2.0 model, but now we have the official Gemini 2.0 flash inside of Google Gemini and the Google Gemini app. All right, a big US copyright case that I didn't really see a lot of people talking about, so I think it's important to chat about it. So this is the second kind of quote unquote ruling or kind of paper to come out of the U.S. copyright Office. And there will be more. So the U.S. copyright Office has issued guidance stating that AI generated outputs based solely on text prompts are not protected by current copyright law. So this decision highlights the lack of human control in AI generated works, which affects their ability to be copyrighted. So this new report from the US Copyright Office emphasizes that AI systems cannot hold copyrights and users of AI systems are not considered authors of the AI generated outputs. So the unpredictability of AI outputs as demonstrated by a Gemini produced image in the paper underscores the lack of user authorship. So the Office contrasts AI generated works with traditional art forms like Jackson Pollock's paintings, where human control is evident. So using AI as a tool to assist human creativity does not necessarily jeopardize copyright protection for the final human produced work. So, yeah, this one's a little tricky, right? So what this report essentially said, if you're using a text only prompt, probably like let's say you're using, you know, Mid Journey or, you know, Dall E, although I don't think anyone would be using Dall E, although reportedly, according to OpenAI's AMA on Reddit, they are closer to releasing the successor to Dall E. But regardless, if you're just using a prompt, right, to inside Mid Journey or inside any of the stable diffusion models or inside Runway, right. Inside a video platform, if you're just using a text output, you will very. It's very unlikely that whatever you produce, you will be able to copyright. All right? And in instances where there's a lot of kind of human back and forth and even human inputs, right, so let's say you create a piece of art by hand and then you go in and enhance it with, you know, an AI in painting from one of these programs, even from something like Adobe Photoshop, right, that's where it becomes extremely vague. I am not a huge fan of this ruling, just because of the gray area. I would say this ruling, we'll link to it in the newsletter and we did last week, but I'd say it's like 90 gray area, right? Because essentially they said in those instances where it's kind of almost like an equal input of human and AI, they're all going to be kind of judged for their ability to be copyrighted on a case by case basis, which does not make sense. I'm not saying I have the answers, but I, I don't think this is the ruling really any was hoping for because all it really said is, okay, well if you use text only, you can't copyright that, which I think a lot of people already knew. And then conversely the, the only other kind of new piece of information that I think that we really got is, hey, if you, if it's kind of equal parts human and equal parts AI, it's being tackled on a case by case basis. But the office fork, the office allows for copyright protection in works that incorporate AI generated content with significant creative modification. So, you know, let's say if it's mainly human work with a little bit of AI, you might still be able to get that copyrighted. Right? Which is huge because here's the thing, in most creative workflows in 2025, it is going to become very hard to create something without the use of AI, right? And now, you know, we've, we've seen all these stories now of, you know, publications and writers guilds trying to create systems to prove that your work is not AI generated or AI augmented, which I don't think is going to accurately work because that's not how generative AI and large language models work either, or diffusion models. Right. It's very hard to prove or disprove aside from if you are using a system that does have an invisible watermarking system. But those are also easy to get around. So it's, it's a pretty, pretty interesting announcement here coming from the second report from the U. S Copyright Office. All right, moving on. $40 billion. That is billion with a B. So according to reports, Softbank Group is in discussions to lead another significant funding round of up to four $40 billion for open AI, potentially valuing the AI behemoth at $300 billion. This would set a record for a private company investment if this does come to fruition. So the move highlights the escalating competition in the AI sector, particularly with the Chinese players now with Deep Seek and Quinn really entering the Fray in late 2024 and really putting their stamp on the race in early 2020. Bank has previously valued OpenAI at $260 billion, a substantial increase from $150 billion. That was the the valuation just a few months ago. This new round of funding reportedly is expected to be facilitated through convertible notes contingent upon OpenAI restructuring its business to limit the control of its non profit arm, which is originally how it was founded. SoftBank reportedly has about $30 billion available for investment with a commitment of up to 25 billion potentially directed towards OpenAI and its joint venture Stargate, which we talked about in the newsletter quite quite a bit over the past week or two. Stargate, if you didn't know, aims to bolsters the US position in the global AI race against China and others, complementing SoftBank's previous 15 billion dollar investment into that venture. So yeah, now it looks like Since SoftBank is one of the main financial supporters of the Stargate project, that kind of joint venture between, what is it, SoftBank, Oracle, Microsoft's in there, more on the data side. So now SoftBank reportedly going all in doubly on OpenAI by not just the Stargate financial support in leading kind of the financial push there, but also in this new round of funding for OpenAI. So yeah, if you didn't know, the Stargate Initiative is a 500 billion dollar joint venture led by OpenAI, SoftBank and Oracle and it aims to expand US's AI infrastructure with significant investment in data centers, job creation and technological advancements. All right, speaking of Oracle, one of the largest companies in the world just announced big surprise advancements in their AI agents capabilities. Are you still running in circles trying to figure out how to actually grow your business with AI? Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on gen AI. Hey, this is Jordan Wilson, host of this very podcast. Companies like Adobe, Microsoft and Nvidia have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead and some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use Gen AI. So whether you're looking for ChatGPT training for thousands of or just need help building your front end AI strategy, you can partner with us too. Just like some of the biggest companies in the world do. Go to your everydayai.com partner to get in contact with our team or you can just click on the partner section of our website. We'll help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on gen AI. So Oracle has introduced New AI agents aimed at enhancing supply chain management during its Cloud World event in Austin. So the AI agents are designed to automate and streamline task force supply chain workers, improving efficiency and decision making. Oracle's new AI Agentic tools are part of the Oracle Fusion cloud supply chain and manufacturing platform that is a mouthful and a half, targeting areas like product inspections and delivery instructions. So the move aligns with the broader tech industries trends. Everyone's going, you know, Agentic in 2024 and 2025, joining the likes of companies like Microsoft, Google, Amazon, Nvidia, Salesforce, OpenAI, everyone else investing heavily in AI agents. So Oracle's announcement obviously, like we just talked about, follows the Stargate project collaboration with OpenAI and SoftBank. All right. And then last but definitely not least, we got a new model that it looks like is going to, at least from a benchmarks perspective, be the highest or the best new model in the world. So we'll see how long that lasts until you know, Quinn or Deep Seek or Google or Claude announces a new model. But OpenAI has launched the O3 mini reasoning model. Not just that, but also for free. Yes. So a lot of people are saying this is in response to Deep seek releasing their R1 for free, but OpenAI did just announce this for even free users, marking the first time a wide audience can access the most advanced models which were previously reserved served for the 20amonth chat GPT plus users or the 200amonth chat GPT Pro users. So there is. Yeah, gotta love these names here. But Sam Altman did say that they were trying to fix their naming problem. So we have two new models actually O3 Mini and O3 Mini High. I know. And this follows 01 which was before called 01 Preview. And then we had 01 Mini. Then we got rid of 01 Preview and just got 01. But then we also added 01 Pro. We skipped O2 because of some, some, some naming rights over there in the UK from I believe a telecom company. And then we skipped to O3. But we don't have O3 preview. We don't have O3 Pro. Obviously not yet. Instead we have O3 Mini and O3 Mini High. Oh gosh. Head scratcher. All right, anyways, so oh, these models use the chain of thought reasoning technique and Sam Altman did say that they are going to have a little bit more transparency in what is actually happening in that chain of thought or those reasoning steps like other models have a little bit more transparency. OpenAI has been a little bit more closed off and showing everyone kind of exactly what happens underneath the hood. Aside from a summary. So according to OpenAI, the O3 mini responds 24 faster than its predecessor O1 mini. Okay so comparing O1 mini to O3 mini. So O3 mini responds 64% faster and costs 63% less per input token though it does remain about 7 times more expensive than than non reasoning models like GPT4O mini. But to me that makes sense. Right? You're not going to use in an O series model for everything if you're a company. Right. You're only using it for more complex tasks. Right? The, the reasoning models, I think they are much better for anything in coding, software development, stem anything in advanced mathematics, business intelligence. Right. I think it's great for reasoners but one thing that I love and we did a big 01 pro episode here on the show so if you want you can scroll back and listen to that. It was about 10 to 15 episodes ago but I think it's also great. These O1 03 models are just great for dumping a lot of data and saying hey this is who I am, this is what I'm trying to do. This is it requires some business logic. So I think a lot of people are saying I'm not, not I'm not in software development, I'm not in business intelligent. Well do you have a pile of data? Do you have a, a job that requires you to think in a multi step way and make a business decision based on that data? Right. I think it's a, the, the. The O. The O1 model. Completely love it. I'm using it all the time. Even though I thought I would still be using the 4.0 model much more I'm using the O1 and now the O3 mini high. What a name there much more than I thought. But the launch signals OpenAI's strategy to maintain its edge in AI innovation despite increasing competition. Like we said from more quote unquote open source models like deep seqs v3 and R1 its reasoning model. But the busy the biggest feature no one's talking about saving what I think is one of the biggest things for last it connects to the Internet. All right. O3 mini and O3 mini high connect to the Internet where the previous 01 preview, 01, 01 mini and 01 pro did not connect to the Internet. And that is extremely important to be able to quickly query the Internet, pull in information because one of the biggest things about using a large language model and why I still tell people, tell companies do not use Claude on the front end. Right. If you're using it on the back end, connecting it, you know, with a rag pipeline into your company's most up to date data. Sure, right. But so many companies are using large language models on the front end. Right. They're going to chat GPT.com Claude AI gemini.google.com Right. There's been a lot of problems. I think Gemini really improved in December and we've seen these updates roll out. But Claude, it's still not connected to the Internet, although we've seen some reports that saying they're getting closer to having that. And then maybe that might be suitable for enterprise companies to use on the front end. Right. On, on a teams or an enterprise plan. But you know, I'm, I'm personally very excited that OpenAI has included the ability to connect to the web in the new O3 mini and O3 mini high model. So there we go, saving what I think is the best little nugget for last. All right, that is it. Let me do a very, very quick recap of the AI news that matters for February 3rd. So according to a report Deep seeks hardware cost more than $500 million. Throwing a big corkscrew into that hole. We trained it for $5.5 million. OpenAI is scrutinizing deep Seq for allegedly using its model for distillation purposes. Despite all that, our next AI news, Microsoft is hosting that somewhat controversial model Deepseek R1 on its Azure platform. More OpenAI news, they partnered with the US government both for those 15,000 scientists and with Chat GPT gov. Google has rolled out Gemini 2.0 Flash for free users. And in the front end of Gemini as well, the US Copyright Office kind of clarified AI generated content copyright status. SoftBank is reportedly eyeing a 40 billion dollar billion investment in open AI even on top of everything, Stargate. Speaking of Stargate Partners, our next piece of AI news, Oracle unveiled some AI agents for supply chain management at its Cloud world event. And last but not least, OpenAI did release its new reasoning model O3 mini and O3 mini high to even free users. All right, was this helpful? I hope it was all right. Live stream audience, thank you for tuning in. You know if this is helpful, go click that repost button if you're listening on the podcast. I super appreciate it. You know, you make sure to check out our show notes and more than anything, please check out our website at your everyday AI sign up. Because like I said, so much happens in the world of AI and it's hard to know. Hey, should my company be using this model? You know what does how does this new news impact our department and what we're doing? Well, don't spend hours every single day trying to keep up. That's what we do. You can keep up in our free daily newsletter, your everydayai.com all right. Thank you for tuning in. Like I said, we have a big deep seek show coming at you this week. We're going to be covering the O3 model in more depth and we're going to be tackling operator as well. OpenAI's AgentIC AI system. So huge lineup of shows this week. I can't wait. Thank you for tuning in. Hope to see you back tomorrow and every day for more Everyday AI. Thanks y'. All.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us 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 everyday AI.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.