DeepSeek is everywhere: What is it and what possibilities does the viral competitor of ChatGPT have?

By
Stephan
-
0
1

This chatbot is currently storming millions of cell phones, browsers and PCs. Find out everything there is to know here.

Skyrocketed overnight: DeepSeek. The name belongs to the Chinese developer, which specializes in artificial intelligence. It is also the name of its AI-powered chatbot.

DeepSeek is a serious alternative to models like ChatGPT, Gemini and Co and is already causing quite a stir, for example at Nvidia

Particularly exciting: DeepSeek has already caught up with or even surpassed ChatGPT in terms of downloads in app stores. But what makes DeepSeek so popular with users and what exactly can the bot do?

Like other LLMs, DeepSeek also filters its answers to problematic topics and refuses to answer questions about drug prescriptions. But DeepSeek is also subject to China’s state censorship. The AI avoids statements about the Tiananmen massacre, the oppression of the Uyghurs in the Xinjiang region or the political independence of Taiwan.

Answers that include the name of Chinese head of state Xi Jinping are also withheld. This censorship is actively monitored by a Chinese government agency, but it can be circumvented with the right questions or by running the LLM locally.

DeepSeek: Features and reasons for the hype

The download charts in the App Store and the lively discussions on Reddit indicate that DeepSeek is very popular with users. There are several reasons for this:

1. Free and versatile features

DeepSeek offers its AI chatbot for free on the official website – no subscriptions, no hidden costs. Alternatively, there is also an app for mobile users (iOS and Android) and for PC.

The AI works in real time, answers questions, explains complex topics, analyzes documents and provides programming support. On top of that, it is capable of creating content – from articles to poems to structural templates.

Just like you know it from ChatGPT.

But the AI can do even more: If you use the R1 reasoning model, you can see the thought process that the chatbot goes through. This way you can better understand the result, get more information and improve your prompt skills by realizing which of your statements the bot reacts to and how.

In addition, DeepSeek’s image generator is already available on Github, which we hope will soon be included in the app.

2. Open Source instead of Black Box

One of the great advantages of DeepSeek is its openness. The code is completely open source, which means that developers and researchers can freely view, adapt and even use it for commercial purposes.

If you are interested in doing it yourself: Click here for the Github page

In contrast to this, there are other providers whose AI models run in opaque systems and have strict terms of use. DeepSeek shows that open source is not automatically inferior: in tests, the model has proven that it can keep up with the best private alternatives.

3. Two models that are causing a stir

DeepSeek’s two main AI models, namely V3 and the R1 reasoning model, are particularly impressive in math, programming, and word problems.

One thing in particular is causing a stir: its good performance compared to ChatGPT, which was achieved at a comparatively extremely low cost. This is interesting for both the economy and programmers, because the costs for using DeepSeek’s API are significantly lower than for ChatGPT.

And we users?We can use the AI from China for free and will benefit sooner or later from the fact that DeepSeek is increasing the (price) pressure on competitors like ChatGPT, Gemini and Claude. Provided that the hype turns into lasting interest.

Here are the details:

  1. DeepSeek V3:

    • Strong benchmarks: The Large Language Model (LLM) has outperformed Llama 3.1 and GPT-4o in benchmarks such as programming tasks.
    • Cost-effective training: V3 training reportedly cost only $5.5 million, compared to OpenAI’s $80 million.
    • Cheap Tokens: Also impressive: 1 million tokens cost just 55 cents, while OpenAI charges $15 for the same amount.

  1. DeepSeek R1:

    • Strong benchmarks: R1 excels at logical reasoning and outperforms other models in math tests like MATH-500.
    • Pure reinforcement learning: R1’s ability to tackle math, code, and brain teasers is made possible by the use of pure reinforcement learning This technique allows the AI model to learn independently and make decisions based on the environment and incentives. A technical milestone.
    • Cost efficiency: This makes it the first Chinese open-source model to operate on a par with the market leaders, but at 90 to 97 percent lower cost.

But be careful: Some analysts point out that the actual figure for the Chinese company is not known, so we cannot say for sure whether these claims are true.

And that’s not the only problem with DeepSeek.

A chatbot with a dark side?

Like many AI models, DeepSeek reflects the values of its country of origin. The chatbot adheres to the guidelines of the Chinese government, which means that it does not answer questions on political or sensitive topics such as Tiananmen or Taiwanese autonomy.

In general, questions about human rights violations in China are not answered.

However, this censorship only applies to the online version. Those who run DeepSeek locally have unrestricted control!

What do you think of DeepSeek? Have you tried it yourself yet? How do you like it compared to ChatGPT and Co.?

Do you think DeepSeek’s country of origin is problematic? Feel free to share your opinion in the comments! Please keep the discussion factual and be considerate of others.