In this guide I will explain to you What is Deepseek and how to take your first steps with this AI. I already showed you what are the main differences between Deepseek vs. Chatgpt and now it is time to learn a little more closely this new proposal that is putting in trouble to leading companies such as OpenAi, Microsoft and Nvidia. Don't miss it!
What is Deepseek and why is it generating so much stir?
To start, you should know that Deepseek is a Chinese company that, since 2023, designs and trains language models. Currently, it has two models, which are the most powerful. One is Deepseek-V3a traditional language model, which generates text from text. That is, when a user writes a request, the model calculates what is the most likely and composes it.
On the other hand, Deepseek-R1 It is what is known as reasoning model. This combines the potential of the language model with more computing time, forcing the model to “take their time” when composing the answer. It is not only that it returns the most likely text according to the user's request, but that the Prompt Input and solve it step by step, task by task.
Deepseek's exploits
Currently, Depseek has achieved some important technological feats. To start, he has achieved match the performances of leading modelsthose of OpenAi. Deepseek-V3 manages to overcome in some reference tests to your rival, GPT-4O. For its part, Deepseek-R1 does the same with O1.
So, to begin with, we have two models launched by Depseek that stand up to Openai. This, in itself, is not really a great victory. The thing gets really interesting when we discover that Both models are open source. This is a blow on the table!
And there is even more. Being open source, Deepseek models can be executed at home. Depending on the number of parameters that selections, you could even have a model with the ability to reason running in your mid -range laptop. It sounds good, right?
Wait, that we are going to put the icing on the cake. Deepseek says that training these models has cost you much less than Openai. Besides, Your API rates for the reasoning model are 50 times cheaper than those of O1. He even states that he does not even need the most pointers in Nvidia to execute his infrastructure, since these models, to equal capacity, are much more efficient.
After all this, we could say that these are Depseek's keys:
- His models are up to Western models.
- According to its creators, the training cost of the models is much lower than what Openai has cost.
- The execution costs are also very low compared to the most powerful openai models, especially if we compare the rates of
- The models are available for free for everyone. They can be adapted to any need and execute at home.
Deepseek is also a chatbot that you can use
In addition to everything that I have explained so far, Depseek is also a chatbot application that you can download in Android either iOS. Additionally, it is available in the browser in the URL chat.deepseek.comso it is compatible with any device, since you run your cloud models. This is the most interesting option for most users, although, as I explained in the previous section, Models can be executed locally in quite modest computers (especially Deepseek-R1).
From here, I explain the steps to start using Depseek on your device. All indications have been followed from a desktop equipment, although the procedures are the same on the mobile. Once you have accessed your main page (you have the link above), You will have to open an account. This is a mandatory step. As I understand that you are not in China, your options to open a new profile are to use your email or use a Google account. In iOS it is also possible to open an account with Apple's login.
By the way, if you see the superior message, It means that Depseek is experiencing problems. In this case, what seems to occur on the platform is a malicious attack. You may not be able to register at this time.
In any case, when you have created your account, you will quickly see that the interface is very similar to that of Chatgpt. You just have to introduce a petition and press in intro so that Deepseek processes it.
The answer appears in the form of text, as it does in Chatgpt.
These answers are possible thanks to the Deepseek-V3 model, a classic LLM. Now, if what you need is for the model to apply a simulation of human reasoning and spend more computing time, you have to activate the button where it says Deceptk (R1). In the mobile application it appears as deep thinking (R1), in Spanish.
Acting this option is highly recommended when you need to solve a more complex request, such as a riddle, an enigma or a mathematical problem. By the way, you can activate deep thinking at any time during a chat or open a new one.
In this case, I have asked the following question:
- When is the following operation correct: 11+3 = 2?
After “thinking” for 18 seconds, it has come to the conclusion that the correct answer is that this operation is possible if what we add are hours of the clock or, otherwise, Whenever we use 12 units cycles.
Additionally, you have an option to upload documents. It also admits images such as entry, although what Depseek does is not analyze its content, but Extract text through the OCR technique. On the other hand, it also has a search button that, after trying it several times, does not seem to work very well.
And you, what do you think of Deepseek? Leave me your opinion below. We read!
