In this presentation Enrico Schaefer explains why DeepSeek’s recent advances matter for the future of AI and for business professionals. He contrasts DeepSeek’s visible chain‑of‑thought reasoning with earlier models like OpenAI’s O1, showing how DeepSeek makes the model’s internal question‑asking process transparent. He then details DeepSeek’s use of unsupervised reinforcement learning, where the model trains itself by querying a larger “teacher” model, and how this approach rapidly scales capability with minimal data and compute. Schaefer also describes model distillation—extracting the essential knowledge from a larger model into a smaller one—to accelerate development. Despite limited hardware, DeepSeek engineered layer‑level and GPU‑communication efficiencies, achieving high performance on modest resources. Schaefer argues these technical leaps will dramatically accelerate AI progress, forcing business professionals to adopt AI tools quickly or risk obsolescence. He urges viewers to join his “AI for the Win” community and YouTube membership for ongoing guidance on leveraging AI in business.
Chapters
Here is what you need to know about DeepSeek, and it's not what you're reading in headlines. It's not all of the clickbait and hyperbole and hysteria. What you need to understand about DeepSeek is how they did what they did, because it is going to inform what happens next in the development of new AI models. It is going to impact the pace at which we accomplish new benchmarks with AI, which is going to be greatly accelerated. The, the amount of change and the speed of change continues to accelerate. Just like no one saw that OpenAI would be able to create what it created with ChatGPT a couple of years ago, that even the top experts didn't see it coming, I think that this is not quite the same but similar. If you are staying on top of the industry and you understand the industry, you know that a lot of the things we're going to talk about today have been in the works and part of the progression of the industry for quite some time. However, no one expected it to happen this fast, this dramatically. So what did DeepSeek accomplish, and why is it important for you as a business professional looking to scale up using AI? Well, We saw Microsoft OpenAI release an O1 model late last year in 2024. And that was a reasoning model where instead of just guessing the next word, what would happen is the model would ask itself a question many different times, explore the answer, figure out which is the best question, and then do chain-of-thought chain of thought reasoning in order to try and figure out what the best question is and what the best answer is to that question. So going back and forth like a ping pong ball, analyzing and reasoning until it got to the right answers. That was a fundamental change in a new way for these AI models to assist us as business professionals. Now, a lot of the key functionality that has been added to, you know, ChatGPT in order to make it business friendly, such as the creation of GPTs, uploads, instructions. None of that is available yet, no one, but it will be in the future. And then that powerful model reasoning is going to be really relevant to business professionals because that's what we do every day is reason. Guessing the next word has been astounding in its abilities to 10x us, but the ability to help us reason is going to be another big jump. So chain-of-thought reasoning was a recent leap forward in AI and one that was expanded on by DeepSeek. And the interesting thing with DeepSeek is instead of just doing chain-of-thought reasoning where it's asking, you know, it's thinking and ChatGPT just shows its thinking, DeepSeek actually showed what the AI model was thinking and what questions it was considering and whether it decided that it'd gotten off topic and was going to start again. And that fascinated people because it showed the reasoning, the thought process that was occurring in '01, but also in DeepSeek. And I think that really kind of took people by storm because it's like, whoa, look at the way it is thinking. Look at the way it's reasoning like a real human brain. Right? Another massive leap forward in AI capabilities. The other thing that is really instructive about what DeepSeek was able to accomplish using less time, less money, less compute was this new reinforcement learning approach. Now, last year, about the same time, there was a big discussion about we're out of data. So these models which fed on the internet are out of really good data. 99.9% of the world's data is hidden behind a firewall. Owned by companies, not publicly available. So where do you get really good data in order to make these AI models better? Let's think of it another way. ChatGPT, Claude, these, these models are really good, and they trained on the junk that is the internet, right? So if they're that good training on junk data, a lot of junk data, up a high percentage of junk data, then what would happen if You had really good data. Well, where do you get this data? And there was a discussion about whether or not we could have AI generate new data and have it train on that. And that seemed like a possible idea, but there was a lot of naysayers because they were like, okay, but if the AI spits out bad data and it's training on bad data, now it's exponentially going to become worse. Well, with What DeepSeek was able to do was use, you know, unsupervised reinforcement learning. And what that means is unsupervised, meaning the AI model itself was training itself, and it was using a parent model in order to do so. And here's generally how that would work. Think about a student and a teacher. Let's just say that they use, you know, LLaMA or ChatGPT to train. Using the APIs that are available against the terms of service of those APIs. But put that aside. What the DeepSeek model was able to do is have the model ask a bunch of questions to one of the existing AI models and get the answers in order to scale up really quickly, very, really inexpensively, to learn everything that was important that was embedded In the established model. So it trained itself on one of the other models as a student would learn from a teacher, and it accelerated its learning because it could ask a lot of questions in a millisecond and get a lot of answers back. So that unsupervised reinforcement learning actually is not quite creating its own data, but very similar to using AI-generated output, which is generally accurate in order to be able to say yes or no, this is the right answer, this is the wrong answer, which is how these AI models become smarter. So the third thing that happened here is, is very similar. It's this model distillation. It's taking one of these other models and distilling it down in order to take the prior work product and prior accomplishments of another model in order to start there and level up. And so If we if we distill ChatGPT down to its essence, down to the connections and the vectors of what's important and what's not important, and we have a way of pulling that out of the system, then we can just simply build on top of that. So what does all that mean for the industry? What does that mean for you as a business professional looking to 10x on AI? What it means is that things are going to start happening much more quickly. The level of competition, the approach to all of this is going. to start to accelerate. The other thing that I think we should mention that DeepSeek was able to accomplish is since they didn't have the best GPUs, they didn't have the greatest technology and compute available to them, what they had to do is they had to engineer solutions. Think about it on the hardware side. They had to optimize everything that was happening within the layers of the neural network. Within the GPUs, how the GPUs spoke to each other and related to each other. It had to engineer efficiencies that weren't baked into the GPUs that it had available to it. It had to engineer a solution that was much more robust and efficient and cost-effective than what was previously available. So Yeah, now we're gonna see everyone mimic those things. One of the principles you need to understand is in the AI space, there are very few secrets. Everyone kind of knows what everyone else is doing and how they're doing it. Now, there's some stuff that's baked in that they can protect, but the concepts that drive the innovation are publicly available generally and universally. So that means everyone is constantly using every— the latest and greatest innovation to then leapfrog What was the latest and greatest, the top dog yesterday gets leaped tomorrow endlessly. And we're seeing that big loop. So what does that mean for business professionals? Here's my take. You know, things are accelerating quickly. You've got the exponential curve where technology is driving innovation at exponential rates. You've got change happening in ways that are instead of decades, years and soon-to-be months and potentially weeks, it means that you as a business professional need to learn how these AI tools work and how to use these AI tools. If you do not level up on AI, it's not that you're going to be inefficient or that you're not going to be as competitive or that you'll have to go about it some other way. You're going to be obsoleted. You are going to be so obsoleted so quickly in ways that are historic. In ways that have never happened before, we didn't have to fear them. We didn't have to think about them before because everything took time before. Everything took years and decades. The technology would come into place, but it would take ten years before adoption. That is not the new reality. The AI age is different. So if you can't figure out how to use AI to deliver services, and fifty percent of our economy is service economy, professional. Professional services, intellectual activities. Then you are going to be in trouble because it's not going to be you doing the work. It's going to be a much superior tool called AI that's doing the work that you're doing today. And why would anyone hire you? You'd be like a you know a shovel digging you know a pool for a resort using a spade versus a steam shovel. No one's going to hire someone with a shovel for that job. And you will be standing there with a shovel. You will be left behind. The age of the AI elite is coming. In order to be part of the AI elite, you have to understand how AI works, because if you don't understand how it works, you're never going to be able to get the most out of it. You're going to run down rabbit holes and paths that are frustrating and don't get you the efficiency you're looking for. And then once you understand how the tools work, you need to learn how to interact with the AI tools. Figure out which are the best tools that are available. Figure out how you're gonna change your business model in order to leverage these tools. You know, all of these different things, some of which we don't even know yet. But if you're not standing there ready and able and willing for the next opportunity, known or unknown, you are going to get crushed. So I'm Enrico. I'm here to help. I'm here to work with you. My community shares back with me and teaches me things every day. I try and share with you. Join AI for the Win. It is a community of business professionals designed to allow all of us to power up to 10x together as this technology develops. I also want to mention that we now have a membership for— on YouTube where you can get access to exclusive videos and shorts and also live streams. And it's like $9 a month, so it's really cheap way for you to get in on the really meaty stuff that, that is going to level you up. Tips about like what exact AI tool for what exact purpose you should be using today. A lot of little snippets that are going to power you up quickly and endlessly. So go ahead and take a look at that. Go ahead and join the membership. You could always, you know, uh, unsubscribe later, but I think you're going to find this great value in it. So if you're on TikTok, go to YouTube. If you're on YouTube, go ahead and become a member of our AI for the Win community. Till next time.
