What is DeepSeek: The Chinese AI chatbot creating waves in the AI space

If you have been following the AI space for a while, you’d know that the market is dominated by US giants like OpenAI, Google, Meta, and Anthropic to name a few. Now, China-based firm, DeepSeek, has released its very own AI chatbot, and it has been creating waves.

But what is DeepSeek that is arguably disrupting the AI landscape? Let’s take a deep dive.

What is DeepSeek?

It is an AI chatbot, not unlike the popular ChatGPT, which has debuted on the Google PlayStore for Android phones and Apple’s App Store for iPhones.

Like any other generative AI, DeepSeek can give instant answers to your questions, help you search the web, read and summarise documents, and prepare an itinerary for you. Furthermore, DeepSeek can also help you code, solve mathematical problems, and do a lot more.

Basically, DeepSeek can do most things that ChatGPT can do, at about the same speed as ChatGPT. So, what’s different about DeepSeek? And why has it become a hot topic in the Silcon Valley (the hub for AI tech)?

Well, DeepSeek is a comparatively cost-effective AI chatbot that runs on its own open-source model, DeepSeek-V3. What’s special about the model is that it’s built using a fraction of the cost required to train and build models of OpenAI.

While US-based AI firms, like OpenAI, spent close to $100 million to train its GPT-4 model, DeepSeek spent close to $5.6 million to train its DeepSeek-V3 model. DeepSeek recently released its new R1 model that succeeds the V3 model. It also directly rivals OpenAI’s o1 model.

ALSO READ: ChatGPT turns 2: How this AI evolved from concept to conversations

What’s interesting is that DeepSeek’s AI models are powered by Nvidia’s less-advanced H800 chips. This is because China does not have access to the latest Nvidia AI chips, thanks to the US ban. To put this into context, OpenAI and other US tech AI firms used Nvidia’s flagship-grade chips like H100 or its follow-up Blackwell B200 to train their models.

Despite this, DeepSeek was able to build and train its AI models, making its chatbot as good as the likes of ChatGPT. It’s no surprise then that Silicon Valley firms are vary and keeping a close watch on DeepSeek.

For users (like you and me), DeepSeek is just another chatbot. It’s free to use, is quick to respond, and doesn’t feel too different from other chatbots. However, DeepSeek cannot generate images at the moment. Although, the company just released a new image model family called Janus-Pro, which can be accessed via the AI community platform — Hugging Face.