In the last week or so the main topic in markets, finance, technology, and politics has been the release of DeepSeek. Everybody has been talking about DeepSeek, and the discussion hasn't stopped yet. In fact, this may be the beginning of the conversations in the next phase of AI development. As you may have already heard multiple times, Chinese AI lab DeepSeek has released its R1 AI model. Not only they made the model available for the public, they released as an open source model. There are a few interesting points that make this event a significant one. Not only it caused panic in markets, it also made everyone involved in AI space to reevaluate their approach in the development of these technologies. Let's take a look at a few of these key points:
- DeepSeek R1 is Open Source.
- DeepSeek R1's benchmark tests have demonstrated comparable or better performance against leading models like ChatGPT.
- DeepSeek claims to have only spent $5.5 million in developing the model, while AI efforts in the US and elsewhere are spending tens of billions.
- DeepSeek didn't have access to the latest and greatest Nvidia GPUs.
These are some of the interesting points why DeepSeek has been a big breakthrough in AI tech, there are of course other aspects that are not less important as well. There are also many unknowns, and speculations. For example, it is not clear what data what used in training the model. What is clear though, DeepSeek has achieved something that other AI labs and companies have not despite having limited financial and technological resources. What I find the most fascinating and the most important in these developments is that the model was released as Open Source and now others are considering Open Source strategies as well. I will return to open source side of it after we explore the other points.
U.S. and the west have declared a trade war against China years ago. This put Chinese companies in a difficult situation. While China's economy has been growing and it has established trade relationships with many countries around the world, especially the US and the west, something happened a few years ago. China has been an attractive market for the US and western companies for its cheap labor and resources. Over the years Chinese entrepreneurs were great partners for the western companies. Chines products dominated the global markets, everybody was happy. However, this economic growth also has influence in industrial and know-how growth as well. Chine was longer just a worlds manufactory, entrepreneurial spirit and desire to innovate was striving as well. Tech companies with great potential were emerging rapidly and still are. Chinese businesses do not need to copy IP anymore, they are building comparable or even better products that can compete with the global brands. In my opinion, this what worried the politicians and business leaders in the US and the west. The competition. Hence, unfair competition practices were implemented. I don't buy the excuse of CCP rule is what the bans were about. Nobody has issues with CCP when there was a trade need with China for cheap labor and resources. Now that the competition in the space of technologies and innovation, all of the sudden CCP is an issue.
Global politics of these things is broken at this time. However, I am hopeful one day parties will come to reasonable resolutions and find a win-win way forward. DeepSeek's release of the model is significant here. It has demonstrated that any political actions, sanctions, and bans are useless. When it comes to innovation, some may find it as a challenge when there are artificial obstacles are created and work harder to accomplish better results. DeepSeek was able to build a model that caught everybody in the space by surprise. It has shown there is more than one way of achieving success in AI. It has shown there may not be a need for huge amounts of investments. It has demonstrated that even the limited power hardware can be utilized in creating quality AI models.
Before DeepSeek events unfolded, the main strategy was more GPUs, more powerful GPUs, and more power is what was needed to make the next leaps in AI. The main focus had been hardware, infrastructure, and energy. Now the topic and focus has shifted more to software, algorithms, and methodologies. There is great possibilities in rethinking the methods, developing better algorithms that will use the hardware, infrastructure, and energy more efficiently. It both the hardware and the software where innovations should continue, and one should serve as a distraction for another.
Pre-DeepSeek the idea was the established AI companies are well positioned to continue with the innovation. Startups had no chance in the space, because they would have to raise crazy amounts of money just to be relevant in the space. Since startups are usually a gamble for investors, the obstacle of needing the expensive computing power could automatically disqualify startups from investments and innovations with great potentials would lose at the start. Now things are different. Compute is still important, but not everything is about compute now. There are ways to innovate in the space even with limited compute. That is what DeepSeek was able to show and encourage everybody to reimagine the AI approach.
There are speculations, and reports that DeepSeek may not have been completely honest about the costs. One of the recent reports suggests DeepSeek may have spent somewhere between half a billion to a billion in developing their model. These numbers are significantly higher than $5.5 million, but also not as high as what other AI companies spending. Even if that is true that DeepSeek actually had much higher costs, it still shows the possibilities of achieving high results with less costs. There are also speculations that DeepSeek actually had access to the best Nvidia GPUs. Even it this is true as well, the amount of GPUs available for them still would be a lot less than what leading AI labs and businesses utilize.
What everybody agrees on is that DeepSeek indeed came up with an innovative methods and algorithms that is a lot different the approaches already used by ChatGPT and others. Many experts have expressed that there are lessons in this achievement, and DeepSeek being open source will help all involved in rethinking and improving their methods as well. While DeepSeek may have cause temporary panic in markets and AI space, the end result is a win for everybody. This is an important moment to reevaluate, rethink, and reimagine AI. AI experts will optimize their new models, AI investors will be smarter with their money, AI companies will have to reconsider open source and closed source models. Sam Altman just recently admitted that open source might be a better way moving forward. I am not sure if that means OpenAI will finally embrace open source approach for future developments and releases. But one thing is clear, it is viable option.
I don't think DeepSeek made their model open source for altruistic reasons. I believe they didn't have a better option. DeepSeek is already not trusted as they are considered to have close ties with CCP. Even with open source release, it will take a very long time to establish trust in this space. They did the right thing with making it open source, because everybody can verify and test their software. That is the beauty of the open source. Not only it can be tested globally by anybody and tested, but also brilliant minds will learn from the code and methods, and build on top of it better technologies. I believe this is the best way for AI to move forward. This will remove confusion, fear, and concerns from many. The AI development can be more efficient, effective, and responsible if done open source way.
Nobody will own AI. It will not be like Microsoft dominating PC markets, Google dominating search, and Amazon dominating e-commerce. This cannot be the case for AI. AI models will have embrace open source otherwise many investors may end up losing huge amounts of money on whatever they bet on. Open source will facilitate global collaboration. Now companies can build products that are closed source built on top of open source models. That is a fair and acceptable practice in the free market. But efforts to create monopolies and dominance with AI technology will not be a right path forward.
DeepSeek has played the game well, and even reached the top rank in the App store. I am sure this is not the end of DeepSeek and we will see more innovation from this lab. I hope they will stay true to open source, now they have established this a solid standard for AI tech.
In conclusion, we may see continuation of political stupidity when it comes to AI and competition in this space. In fact, such stupidity has already shown its initial signs. US senator Josh Hawley has introduced a bill that could make it illegal to use DeepSeek in the US and people may get 20 years in prison and fined for $1 million. Politicians have lost their minds. I doubt anything stupid like this would actually become a law. But even considering this and wasting money and time on such bills is insane. The hypocrisy of these politicians is at an all-time high.