DeepSeek

AI made in China

On 20 January 2025, the bombshell dropped: DeepSeek, a hitherto largely unknown start-up from China, released its latest AI model, DeepSeek-R1. Since then, nothing in the AI world has been the same - really? It is undisputed that DeepSeek has developed a powerful model with relatively little effort. Despite having fewer resources, it can compete with significantly larger and more expensive US models such as those from OpenAI. Account registrations and app downloads exploded. Ironically, the US embargo on the most powerful NVIDIA chips presented the Chinese start-up with the challenge of developing a powerful LLM with limited resources. They have obviously succeeded.

However, after the initial hype, the first doubts are now spreading. The development was probably not quite as favourable as initially claimed. Calls for data security are getting louder. Nevertheless, DeepSeek is making a remarkable contribution to current AI research.

DeepSeek's approach shows that competitive AI solutions are now possible without huge budgets. This is a good sign for the comparatively financially weak European AI scene.
Tim Strohschneider | Head of GenAI | adesso SE

Open source approach as an alternative

Unlike ChatGPT, DeepSeek makes its AI available as open source. The advantages for both companies and the global AI community cannot be denied:

Full control and customisation options

  • Customizing: Companies can customise the model to their specific requirements. This is particularly interesting for regulated areas.
  • Local implementation: The open source nature makes it possible to run the models locally on your own servers, which significantly improves data security and data protection.

Cost savings

  • No licence fees: In contrast to commercial solutions, there are no ongoing costs for using the model.
  • Efficient entry into AI projects: Start-ups and smaller companies can utilise powerful AI technologies without large investments.

Promoting innovation and collaboration

  • Open source promotes an active developer community that fixes bugs, extends functions and develops new use cases.
  • The model can serve as a basis for academic research or new commercial applications.

Faster further development

  • Open source creates numerous variations through fine-tuning, which in turn can be easily customised for your own purposes.
  • Synergies are created between different projects and industries as the model is accessible to all.

Independence from large providers

  • Companies that use open source models such as DeepSeek are not dependent on cloud providers or technology giants such as Microsoft, Google or OpenAI.
  • This reduces the risk of vendor lock-ins and gives companies more strategic flexibility.

DeepSeek's open source approach could provide a long-term counterweight to the closed models of established AI providers and democratise access to artificial intelligence worldwide.

Caution with sensitive data

The easiest way to operate the chatbot is via a web browser. However, caution is advised here. All requests and results end up on Chinese servers. It didn't take long for a DeepSeek database containing sensitive data to become freely available on the internet. Experts from the cyber security service provider Palo Alto Networks got DeepSeek to generate a script for reading emails and Word files. Such scripts are often the basis for hackers to steal data.

More and more data protection authorities are advising against their use. DeepSeek does not fulfil the minimum standards of the GDPR, meaning that a legal representative in Europe is also required for companies not based in the EU. DeepSeek does not yet have one.

DeepSeek as inspiration

One, if not the biggest advantage of the hype surrounding DeepSeek is that open source models are once again coming more into focus and therefore represent a serious alternative to the top dogs. Depending on the use case, companies and institutions do not need high-performance all-round models. Although these offer enormous possibilities, they also require enormous resources - not least financial resources. It often makes more sense to use applications that are customised to individual requirements.

The issue of ‘legal security plays’ also plays a much greater role for many companies in Germany than in the Middle Kingdom or the USA. The French start-up Mistral AI, for example, fulfils all of these requirements. But the European joint project Open GPT-X also provides competitive solutions with Teuken-7B, for example. The selection of a suitable model is the basis for a successful AI implementation, and our experts are happy to provide support here.

Work with us to maximise your return on intelligence

Artificial intelligence is not a replacement for human dialogue.
We look forward to hearing from you.