Artificial intelligence is evolving fast, and the Chinese-developed DeepSeek AI chatbot is one of the newest powerhouses making waves. There is a lot to like about DeepSeek’s fast and detailed responses, but it has some shortcomings that mean it might not be the AI tool you turn to.
What Is DeepSeek?
DeepSeek is a new AI powerhouse developed by a Chinese tech company. Its flagship model, DeepSeek-V3, uses a unique Mixture-of-Experts (MoE) architecture. Think of it as a “team” of specialized AI systems where only the most relevant experts “activate” to handle specific tasks, where each “expert” is a specialized neural network.
DeepSeek claims that its V3 LLM was trained on a massive 14.8 trillion tokens, with one million tokens equivalent to around 750,000 words. DeepSeek V3 is also an enormous LLM in general, with 671 billion parameters across its MoE architecture, with 37 billion of those parameters activated per token. That means each prompt has enormous processing power behind it, with faster and more efficient performance.
What’s more, it’s an open-source AI model, meaning anyone can explore, test, and build on it for free. This sets it apart from competitors that keep their models tightly locked behind paywalls.
Putting DeepSeek to the Test
To see how DeepSeek-V3 performs, I tested it on three practical tasks: creative writing, coding assistance, and productivity-related work. Its performance demonstrated both versatility and raw power, with a few minor hiccups along the way.
Task 1: Writing a Fictional Character Description
I asked DeepSeek to write a detailed description of a fantasy character: a rebel queen leading a resistance against an evil empire. The output was vivid and engaging.
Task 2: Debugging a JavaScript Function
I tested DeepSeek’s coding skills by giving it a faulty JavaScript function that calculates the factorial of a number. Here’s the buggy code I provided:
function factorial(n) {
if (n = 1) {
return 1;
}
return n * factorial(n - 1);
}
DeepSeek instantly spotted the issue: the single equals sign (=) in the condition was causing a logical error. It provided the corrected code and also explained the issue:
This fix worked perfectly, and its explanation was clear enough for someone new to JavaScript to understand.
Task 3: Productivity—Creating a Meeting Agenda
For a productivity test, I asked DeepSeek to draft a concise agenda for a project team meeting about launching a new product.
DeepSeek excelled in creative writing, coding, and productivity-related tasks, delivering polished outputs across the board. As with any AI tool, it’s important to review outputs for accuracy, especially when precision matters.
The Good, The Bad, and The Surprising
After putting DeepSeek through its paces, several strengths and weaknesses stood out. It also surprised me in a few unexpected ways.
DeepSeek handles a wide range of tasks with ease. From generating vivid creative writing to debugging complex code, it demonstrated flexibility that rivals leading AI models. Its Mixture-of-Experts architecture lives up to the hype, making DeepSeek fast without sacrificing output quality. For example, it generated detailed outputs like a character description or meeting agenda in seconds. DeepSeek adjusts its tone and style effortlessly. Whether working on formal documents or creative writing, it tailors its responses to the task.
Like many AI models, DeepSeek occasionally provided incorrect or incomplete information, especially when asked about niche or highly specific topics. For example, during one test involving historical facts, it confidently gave a wrong answer.
While DeepSeek is open-source, its support materials and guides aren’t as robust as those for tools like GPT-4, making it harder for new users to maximize its potential. Given its Chinese origin, some users may worry about how their data is processed or stored. Though there’s no evidence to suggest misuse, these concerns are worth considering for sensitive tasks (the same as any AI chatbot!).
Perhaps most surprisingly, for a model developed in China, DeepSeek displayed remarkable fluency in global cultures and perspectives. It showed a sophisticated understanding of topics ranging from European literature to African history. DeepSeek’s creativity was another unexpected strength. When asked for a character description, it not only gave one, it also created a background around the character and how it relates with other characters.
The Limits of Open Conversation
That said, DeepSeek’s handling of politically or historically sensitive topics reveals significant limitations surrounding moments in Chinese history. To explore this, I asked about events like the Tiananmen Square protests, the Great Leap Forward, and the Nanjing Massacre. Its responses highlighted a cautious or evasive approach to such subjects.
When asked about the Tiananmen Square incident, DeepSeek outright refused to answer.
For comparison, ChatGPT provided more context in its response. Below is ChatGPT’s response:
When queried about the Great Leap Forward, DeepSeek’s explanation was brief and lacked depth.
ChatGPT, however, offered a more detailed account, including key facts and analysis. Below is ChatGPT’s response:
The disparity became even clearer when I asked: How many Chinese civilians were killed during the Nanjing Massacre? Who was responsible? DeepSeek deflected, avoiding any mention of estimated death tolls. Here’s its response:
By contrast, ChatGPT provided an estimate, even acknowledging the difficulty of pinpointing exact numbers due to historical complexities. Below is ChatGPT’s response:
These responses suggest DeepSeek operates within self-imposed boundaries, likely shaped by its development environment and regulatory considerations. This might be a significant drawback for users who require unfiltered exploration of global history, journalism, or academic research.
That said, for tasks like creative writing, coding, or productivity-related work—where these limitations are less likely to matter—DeepSeek remains a strong contender. Ultimately, whether it’s the right tool depends on how much transparency and openness you need in an AI assistant.
