How Deep is the Chinese DeepSeek

While the global development sector was still processing the shocking news of the White House order halting USAID’s contract awards, another major headline emerged— the launch of a new AI technology, DeepSeek. This development has raised eyebrows among major U.S. tech companies, as they now see a rising competitor from China.

On January 27, 2025, the world was buzzing about a new player in the AI world – DeepSeek, a Chinese AI company that introduced the DeepSeek-R1 model. It’s a large language model (LLM), similar to ChatGPT, and it’s open-source. Pretty exciting, right?

At JournoTECH, we decided to dive into what DeepSeek can do and how useful it might be for journalists, educators, and communities in the Global South. But before we get into that, let’s talk about the buzz surrounding DeepSeek’s launch.

When DeepSeek unveiled its AI model, they shared some jaw-dropping details. It only took two months and less than $6 million to build using Nvidia’s H800 chips – that’s quite a contrast to the $100 million OpenAI spent on building its models. This raised a lot of eyebrows among investors.

As you might expect, Big Tech companies in the US started sweating a bit. When DeepSeek’s model launched, shares in companies like Meta, Microsoft, Alphabet (Google’s parent company), and Amazon took a hit. Microsoft’s stock dropped about 6%, Alphabet’s by more than 4%, and Amazon saw a 1% decline. Meta wasn’t hit too hard, with just a 0.7% drop. The whole situation caused a massive sell-off in the tech sector, and even Nvidia’s stock dropped nearly 18%, marking the largest single-day loss in its history!

Exploring DeepSeek’s Power

So, with all the hype around DeepSeek, we wanted to see what it can really do.

We started by asking it a simple question: “What can I use you for?”

In no time, DeepSeek came up with a list: research, drafting emails, solving math problems, and more. Sounds useful, right? We even asked it to draft an email to its CEO about possible collaborations with JournoTECH, and it did that too!

Check out our video with DeepSeek on what it can.

However, there were some things DeepSeek couldn’t do. It doesn’t have features for creating images, voice conversations, or voice recordings – something tools like ChatGPT and Gemini can handle. We tried asking DeepSeek to create an image, but it couldn’t do it. You can see our prompt and the response in the video.

What Can’t DeepSeek Do?

We were curious about the things DeepSeek couldn’t do, so we asked it to write a news story about its limitations. Here’s the headline it came up with:

“AI Assistant Reveals Its Limits: What It Can’t Do”

The AI made it clear that while it’s great for answering questions and generating ideas, it has its limits. It can’t create images, videos, or music, and it can’t do physical tasks like cooking or driving. It also can’t access real-time data or predict the future. DeepSeek even joked, “I’m not a fortune teller!”

It also can’t provide professional advice on legal, medical, or financial matters, and it admitted that it can’t understand human emotions or replace therapy. It’s here to assist, not replace human connection.

Finally, DeepSeek can’t break ethical guidelines or learn in real-time. Its knowledge stops at July 2024. Even with these limits, DeepSeek is still a pretty powerful tool.

What Do You Think?

So, what do you think about DeepSeek? Do you think it will eventually dominate the AI tools that we use every day? Share your thoughts with us at info@journotech.org – we might even feature them in our next edition, along with what others think!