In recent weeks, a Chinese artificial intelligence company has unveiled a new artificial intelligence model called Deppseek R1, sparking a huge controversy. Why? Because now the American models (e.g. o1) in the AI race had a competitor from a country other than the US, and get ready, all this content was OPEN SOURCE! Deepseek-R1 offers almost the same capabilities as the o1 model developed by Open AI, open source and free. It competes with OpenAI's $20 model, and you can install it locally and use it without an internet connection. But why do they do this? Let's take a look.
DeepSeek-R1
1. Deepseek-R1's skills and the importance of open source
Deepseek does really well in math, coding, physics and more. You can find more detailed information about the benchmark tests by accessing the paper from the link below, if you want to learn more about this field, I strongly recommend you to read this paper. Among open source AIs, the best before deepseek were the llama models developed by meta. Deepseek almost dominated this market and managed to become the best open source AI model in such a short time. In my view, it's great for developers and users alike that open source projects are growing and can compete with paid models. I'm sure we will continue to see more and more new models in this environment that deepseek has created, I can't wait for the future!
2. How Reasoning works
The R in its name means reasoning. So how does this work? When the user gives any prompt, it thinks and comments on the prompt just like a human, evaluates how it can fulfill the user's request and creates the result accordingly, and while doing this, it shows us what it thinks instantly on the screen. This is basically reasoning. You can find more information from the links below.
3. How was it developed so cheaply, at little cost, and made available for free?
They developed the model in a very different way, I can explain it in simple terms as follows. We have a heavy rock and we don't know how much it weighs and we need to lift it. First we send 10 people, strong, smart and agile people. They try and try and try and fail to lift the rock and then they get another 10 people and they try and lift the rock again and then another 10 people and so on and so on until they lift the rock. What other AI tools do is to send 100 people at first, but deepseek's method works more efficiently. That's the main reason why it's low-resource and cheap. And why is it free? Because the people who developed this model have been training it as a model that evaluates the stock market for many years and it has evolved more and more to become the R1 model of today. Therefore, the company does not expect any revenue from this model.
4. Deepseek's impact on the US stock market
With the release of DeepSeek, many American tech companies suffered significant losses. For example, nvida shares fell by about 14%. This is primarily because deepseek is a model with low resource consumption. But if you ask me, the resources on how deepseek trained this model are also open source, so American companies will also release such a model in the near future, and my prediction is that this decline will not be long-term.
5. Deepseek's Chinese censorship
If you go to Deepseek right now and ask about past negative events related to the Chinese state, the model says it cannot talk about that and suggests talking about something else. So clearly the Chinese state has an influence on the model, and I think that's one of the negative features of the model. I think it's important for an AI model to be free, no matter what government it comes from, and I hope they will step up in the future.
6. API Pricing
When we look at the api pricing of O1 and deepseek, we see that there is a huge difference. Currently, Open AI's price per 1 million tokens is around 60 dollars, which is a price that can be called normal compared to the resources and energy used by open ai models. But when we look at deepseek, it provides a pricing of 2 dollars per 1 million tokens, so deepseek offers its API 30 times cheaper than open ai. I think this is a difference that cannot be ignored.
7. Can a Chinese company be trusted
This question is one of the most puzzling questions for users. When we look at Deepseek's privacy policy, it is clearly written that all the chats and data we make on deepseek can be used in the training of the artificial intelligence model. This is something we may not really want. It's up to you to take the risk and use it, but installing and using it on your local computer doesn't even require an internet connection, everything works locally and I think it's a safer way.
8. The downsides of Deepseek
Although it is a cheap and powerful model, I couldn't get as clean a result as the open ai o1 model during the time I used it. Yes, it does reasoning well and understands what I want, but it needs to improve a little more in terms of code quality. However, it is too good to be compared with the early days of open ai, but it would not be correct to compare the level of artificial intelligence when open ai was first released about 2.5 years ago with this period.
9. Benefits for small startups
I think Deepseek will make a big contribution to small-scale enterprise use, because if they want to develop an AI project, they will be able to build their own infrastructure and manage it over their own servers, and maybe they won't even need to use an API connection. In this regard, I think it will also provide a positive contribution to artificial intelligence applications.
10. Future development of artificial intelligence
I think that all future AI models will produce better results with less resource consumption and the rapid development of AI will be faster. At the end of the day, it is a win-win situation for the user when a company like deepseek competes with giant companies like open ai. In this competitive environment, models will continue to evolve rapidly, as open ai recently introduced the o3-mini model and opened it to all users for free and added reasoning mode to the model. So artificial intelligence will continue to develop rapidly in this competitive environment and in the future, and let's see what innovations we will see in the future.
Lastly
Overall, it's exciting to see companies like DeepSeek competing with giants like OpenAI and introducing powerful AI tools. In this article, I discussed the deepseek-r1 model on various topics and gave my thoughts. See you in future articles.
Sources:
DeepSeek-R1 Paper: https://arxiv.org/abs/2501.12948