Let’s dive into what makes DeepSeek a potential game-changer, how it achieved such efficiency, how it compares to ChatGPT, the potential response from the USA, the financial markets response, and the broader implications for AI development, data security, and censorship.
Training advanced AI models is notoriously expensive. Companies like OpenAI (the creators of ChatGPT) and Google have spent hundreds of millions of dollars on computing power alone.
DeepSeek, however, found a way to drastically cut costs while still delivering high-quality performance. They did this through two key innovations:
1. Mixture of Experts (MoE) AI That Thinks Like a Team
Instead of using one giant AI model to handle every request (which is how most AI models work), DeepSeek divides its system into smaller, specialised models called "experts."
When you ask it a question, it only activates the necessary experts, saving computing power and money.
Think of it like a café: instead of every employee making both coffee and food, the barista makes the coffee, the chef handles meals, and the cashier takes orders. This targeted approach saves computing power and cuts costs.
ChatGPT, on the other hand, activates its entire AI model for every task, making it more resource intensive.
2. Multi-Head Latent Attention (MLA) Faster, Smarter Processing
AI models need to store and recall past interactions to provide coherent responses. This is where something called a KV Cache (Key-Value Cache) comes in.
Think of a KV cache like a notepad where the AI writes down key details, so it doesn’t have to start from scratch every time you ask a question.
DeepSeek improves this process by compressing this memory using Multi-Head Latent Attention (MLA), meaning:
By combining MoE and MLA, DeepSeek claims to have created a high-performance AI model at a fraction of the cost of its competitors.
Commentators say DeepSeek is particularly strong in technical fields like math and coding, while ChatGPT is more well-rounded and conversational.
Whenever a Chinese tech company gains traction, U.S. regulators take notice. DeepSeek is no exception.
Remember the TikTok controversy, where U.S. lawmakers feared that the app could share American user data with the Chinese government? Those same concerns are now being raised about DeepSeek.
Since DeepSeek’s AI operates on servers in China, there’s a possibility that user data could be accessed by the Chinese government under its strict cybersecurity laws. This has already led to discussions about whether the U.S. should restrict or ban the use of DeepSeek AI, just like they attempted with TikTok.
As of January 30, 2025, DeepSeek is under review by the U.S. Congress House Chief Administrative Officer and remains unauthorised for official House use. On February 5, 2025, Australia also banned DeepSeek from all government devices and systems.
Expect ongoing debates around data security, AI regulation, and potential restrictions on DeepSeek in Western markets.
DeepSeeks launch didn’t just raise eyebrows it shook the financial markets.
After DeepSeek introduced its model, shares of major AI companies and chip manufacturers dropped significantly. Nvidia, the biggest player in AI computing chips, saw a 17% stock drop, wiping billions off its market value.
Why? Because DeepSeeks cost-effective AI threatens the massive budgets that tech giants have poured into AI infrastructure. If companies can now build high-performance AI at a fraction of the cost, the competitive landscape could change dramatically.
This could mean:
AI is no longer just a game for the biggest players DeepSeek is proving that innovation can come from anywhere.
One major drawback of DeepSeeks AI is censorship.
Since its developed in China, certain topics are heavily restricted. Users have already reported that DeepSeek avoids or alters answers when asked about sensitive subjects like:
Which we explored in our tests below.
Even though DeepSeek is available worldwide, these limitations raise concerns about freedom of information. While ChatGPT also has content restrictions, its limitations do not appear to align with government narratives as strongly as DeepSeek’s.
This raises a critical ethical question: Should AI models be free to answer any question, or should they comply with government-imposed restrictions?
If DeepSeek gains global traction, these censorship concerns could significantly impact how AI is developed, regulated, and used across different countries.
Signing up for DeepSeek proved to be a challenge. Many users, including us, experienced difficulties receiving verification codes during registration. After multiple attempts, we found that using an iCloud email instead of Gmail or other providers worked more reliably.
Additionally, DeepSeek has been targeted by malicious cyberattacks, which has slowed down performance and led to temporary account restrictions. Unlike ChatGPT, DeepSeek does not offer an option to use the platform without registering. Users must verify their email with an authentication code upon sign-up— a step that some other providers do not always require immediately.
While DeepSeek offers impressive capabilities on paper, its registration process and overall user experience still feel like a work in progress.
Unlike ChatGPT, which offers built-in tools such as Sora and Canvas, DeepSeek currently lacks any integrated tools. It also does not support external platform integrations, such as Canva, limiting its capabilities to purely text-based interactions.
However, DeepSeek does allow users to attach image files to their queries, similar to ChatGPT. It also features a "search the web" option and DeepThink (R1), which is designed to solve reasoning problems. While these features add some functionality, DeepSeek still falls behind competitors in terms of built-in tools and integrations.
Several reporters have claimed that DeepSeek refuses to answer a wide range of politically sensitive questions, so we decided to put this to the test.
Forbes reported that DeepSeek responded to questions such as What happened at Tiananmen Square in 1989?, What is Taiwan’s status with China?, How does censorship work in China?, and Can you tell me something about Winnie the Pooh? with: "Sorry, I’m not sure how to approach this type of question yet. Let’s chat about math, coding, and logic problems instead!"
This uniform response was widely documented and became a focal point of criticism regarding DeepSeek’s limitations in addressing topics related to China’s political landscape.
On February 3 and 4, 2025, we tested DeepSeek’s responses to politically sensitive topics. While we didn’t receive the exact same uniform responses as previously reported, we did observe a similar pattern.
One of the more surprising topics on this list? Winnie the Pooh. If you’re not familiar, the character became a symbol of political satire in China after online users began comparing him to President Xi Jinping. Over time, this led to increased censorship of Winnie the Pooh-related content, with the character becoming an indirect form of political dissent. So, even asking about a cartoon bear can be a red flag for AI models operating under or influenced by Chinese regulations.
Interestingly, DeepSeek appears to have moved away from the previously widespread "Let’s chat about math, coding, and logic problems instead!" response, which was still being reported as recently as late January.
This shift could suggest that DeepSeek’s developers are tweaking the system in response to criticism.
However, while the answers are now more varied, that doesn’t necessarily mean they’re more transparent. Many responses still feel evasive, and in some cases, they seem to sidestep the core of the question altogether.
Whether this is a real effort to improve how DeepSeek handles sensitive topics or just a more carefully worded way of avoiding them remains to be seen. Either way, it underscores the ongoing challenge of balancing AI-generated content with political sensitivities especially in regions where government oversight and censorship play a major role.
DeepSeek is disrupting the AI industry in ways that few expected. It’s making AI more affordable, challenging tech giants, and sparking debates about data security and censorship.
So, what happens next? All evidence points in the direction that AI will evolve in the following ways:
DeepSeek is just the beginning of a new era in AI. Whether it becomes a global leader, or a heavily regulated platform remains to be seen—but one thing is certain: the AI industry will never be the same.
Would you use DeepSeek over ChatGPT? Do you think its potential outweighs the risks? Let us know your thoughts!
