What is this?
Artificial intelligence is the gift that continues to give – or, from your point of view, the guest who refuses to leave. Every few months, a new model of AI comes promising to be smarter, faster and more powerful than the previous one.
Enter In depthThe last participant of the AI arms race, from China with great ambitions and an even greater appetite for language treatment.
It is compared to the most advanced AI chatbots, including those built on GPT-4, which is currently one of the most powerful AI models available. But unlike GPT-4, which is controlled by Openai and requires a subscription to access its best features, Deepseek is open source. This means that its underlying technology is freely available, allowing developers and companies to use, modify it and improve it as they wish – provided they have technical skills. In theory, this means that companies can take depth and personalize it for their own needs, such as the construction of smarter chatbots, research tools or assistants supplied by AI. Theoretically.
Deepseek is designed to process and generate human text with precision and consistency. If you have already used Chatgpt, Claude or Gemini, you already have an idea of what you expect: an AI that can answer questions, generate content, summarize items and pretend to be a very good assistant Read and somewhat detached.
However, there are several important weaknesses, led by censorship. Since Deepseek has been developed in China and is subject to monitoring of the Chinese government, its responses are subject to restrictions, in particular with regard to politically sensitive subjects. Although this is not unusual – most AI models have a certain level of content moderation – Deepseek limitations are shaped by the regulations imposed on it.
This means that users encounter filtered answers or outright refusals to engage with certain subjects, such as Taiwan or the Republic of China, which makes it less attractive for those looking for discussions on the IA without restriction.
Deepseek differs in its claim from a higher understanding in Chinese language, associated with a robust understanding of English and other languages. Given that most language models (LLM) are still strongly focused on the West, this could create a gap for a global audience.
The name “Deepseek” suggests a profound pursuit of knowledge, although it looks a bit like a Cyberpunk novel from the 90s. In reality, it is a well -formed machine that has been nourished by a huge set of data , which allows it to predict the text at an impressive level. Theoretically, he should be able to write essays, poems crafts, generate code and debate philosophical questions with you at 3 am when insomnia strikes. In practice, it works well – however, like the whole AI, it always has its quirks and its weaknesses.
During the Cisco Live conference in Amsterdam this week, it was revealed that Deepseek had failed a major security test: 100% of attacks against the model were successful, which shows that it was vulnerable to Jailbreak attacks, to data leaks and the manipulation of the model. It did not mark much worse than the Meta lama and the Chatgpt-4o of Openai, but the test highlighted the widespread security risks of AI systems in commercial and government applications.
Where Deepseek really flexes his muscles, it is in the pure opening of his model. The OPENAI GPT -4, for all its power, remains an enclosed garden – you use it in their terms, in their ecosystem. Deepseek, however, is an open book. Developers can take the model, refine it for specific use cases and integrate it into their own platforms.
For the general consumer, it is fascinating, but not exactly intuitive. Experience does not have the smooth and polished interfaces of Chatgpt or Claude, and it is obvious that it is a tool for those who like to experiment. It is a reminder that the best AI is not a question of raw power; This is accessibility and ease of use.
How much does it cost?
Deepseek is free – in a way.
If you are just an occasional user looking to play with AI, Deepseek in its raw form will not cost you a penny. But if you hope to use it on a large scale – for example, for a commercial chatbot or a generation of content – it could become a continuous expenditure. Deepseek does not invoice by word or by response as some AI services, but via a system of “tokens”, based on the calculation power used. The more you ask to do it, the more it costs in power.
Why is it important?
The current AI landscape is dominated by a handful of technology giants that control access to the most powerful models. By offering a high quality alternative that anyone can use and modify, Deepseek shakes the balance of powers. It also increases the challenges of AI development in the world.
For consumers, this could mean more long -term choice. Instead of counting on a single AI supplier, companies and developers can adapt Deepseek to their specific needs.
What are the greatest negatives?
- Lack of friendly interfaces: Deepseek is powerful, but it is not wrapped in a way that makes it accessible to the average user.
- Censorship concerns: Certain subjects may be limited or moderate due to the regulatory environment in which Deepseek has been developed.
- Safety weaknesses: Deepseek failed 100% of tests in a Cisco analysis.
What are the biggest positive points?
- Open Source flexibility: Unlike proprietary models, Deepseek allows companies and developers to modify it and integrate it freely.
- Strong multilingual capacities: Especially in Chinese, it surpasses many competitors, making it a powerful tool for global applications.
- Contest the status quo of the AI: By providing a viable alternative to dominant players, Deepseek pushes the industry to more competition and innovation.
* Arthur Goldstuck is CEO of World Wide World and editor -in -chief of Gadget.co.za. Follow him on Bluesky on @ art2gee.bsky.social.