It’s the first day of the Lunar New Year 2025, and already the world of Generative Artificial Intelligence (Gen AI) is already shaking up! If you have not already heard in the last few days, a Chinese AI startup company has captured headlines all over the world and has “woken” the tech world into the cutting-edge developments of Gen AI, but yet highlighted the inefficiencies of the current tech world. The company, known as DeepSeek, has within a remarkably short span of time and limited resources, demonstrated that it has the ability to develpop a highly advanced Gen AI model that is rivaling industry leaders like OpenAI and Google Deepmind. Their model — called DeepSeek-R1 — is generating serious discussions globally about the future of Gen AI, and perhaps this is the impetus the world needs to lead us into the age of “Agentic AI.”

What is DeepSeek?

Founded in 2023 in Hangzhou, China, DeepSeek is making waves with its flagship model — DeepSeek-R1. The company claims that their AI model has great efficiency and capability, and its performance is on par with OpenAI’s GPT-o1! Furthermore, their model is open-sourced, and the company achieved its performance and efficiency at a fraction of the cost of current Gen AIs!

Using an innovative architecture known as “mixture of experts“, DeepSeek’s AI utilizes specialized submodels (called “experts”) to only activate when needed. As an example, although its V3 model has 671 billion parameters, it only uses just 37 billion per task based on this technique, thereby reducing computational costs tremendously. Moreover, by prioritizing Reinforcement Learning techniques for training, DeepSeek was able to enable faster reasoning capabilities at lower resource costs, yet outperforming its competitors, such as ChatGPT, in terms of common tasks, like doing maths calculations. Notably, DeepSeek-R1’s capabilities extend beyond traditional generative AI tasks like coding and natural language understanding. Its ability in complex reasoning and adaptability has set a new benchmark for Gen AI models. Have a look at this video on what DeepSeek is all about!

DeepSeek vs ChatGPT: Who Wins?

Since it made the headlines, there have been many different comparisons between DeepSeek and the current Gen AI applications, e.g. ChatGPT, Gemini, etc. While many of these comparisons are subjective and depend on the interpretations of the authors, it is clear that despite some disadvantages, DeepSeek may potentially be a leader in the industry as it evolves over time… Have a look at a some quick comparisons in the videos below.

The “Marvel” of DeepSeek: A “Vision” of Duality

Now, if the AI world can be likened to the “Marvel” universe, based on my opinion, DeepSeek could potentially be likened to Marvel’s “Vision”, who arose from Tony Stark’s (aka Ironman’s) AI sidekick — Jarvis! Just like Vision, DeepSeek’s model kind of embodies both immense power and complexity, straddling the line between heroism and ambiguity. For some, it can be considered a hero — an innovator democratizing access to AI, while fostering efficiency and promoting eco-friendly practices through its revolutionary “mixture of experts” architecture. By reducing energy consumption and development costs, DeepSeek is setting a new standard for sustainable AI, potentially mitigating the environmental impact of current large-scale Gen AI models. In this light, DeepSeek symbolizes a champion of progress, empowering smaller players in the AI field (e.g. Small-Medium Enterprises, SMEs) to compete and innovate in an industry that has been long dominated by the tech giants.

In contrast, like Vision, DeepSeek also carries with it other questions about its tremendous potential for good and bad alike. Within days of its global discussions, the model suffered a major cyberattack that forced the company to limit user registrations, reigniting concerns about the cybersecurity of AI models and the need for privacy and confidentiality of data — and its potential for abuse and misuse by “evil-doers”. These complex issues surrounding this newly birthed chatbot paint DeepSeek not just as a hero of innovation, but as an enigmatic figure whose power must be closely scrutinized.

DeepSeek is the Talk of the Town

DeepSeek’s impact is definitely being felt across tech industries and financial markets. Leading tech companies have experienced significant stock market losses just a few days ago due to DeepSeek’s rise. For example, Nvidia saw a nearly $600 billion drop in market value — a testament to the disruption caused by this startup’s innovation. At its core, DeepSeek’s seemingly lower development costs is paving the way for smaller players to compete with the “giants” in the tech space, potentially reshaping the Gen AI landscape in the near future.

The Age of Agentic AI

DeepSeek’s innovations may potentially indicate a shift towards a new paradigm in AI — known as “Agentic AI“. Unlike traditional AI, which is primarily used for creating original material based on previous patterns, Agentic AI possesses a degree of autonomy, enabling it to take initiative and make independent decisions based on learning from a feedback loop. As outlined in Oracle’s blog, Agentic AI holds the promise of revolutionizing business processes and user experiences by offering highly personalized, proactive, and context-aware solutions.

Have a quick overview of Agentic AI by watching this video!

Agentic AI builds upon foundational AI technologies and advanced algorithms to:

  1. Recognize Context: Analyze and adapt to user environments and preferences in real time by recognizing items, extracting relevant and appropriate features, or identifying relevant elements in the environment
  2. Reason & Act: By using a technique known as retrieval-augmented generation (RAG) to access data sources and deliver accurate outputs, Agentic AI can process the data to formulate tasks based on its understanding, and then decide and act on it if it is integrated with other applications and software.
  3. Learn Continuously: Improve performance through iterative learning from new data and interactions in feedback loops and through experience.

The differences between Gen AI agents and Agentic AI are summarized in the table below:

ParameterGen AI / AI AgentAgentic AI
Core featuresFocused on creating original material in various forms (e.g. text, images, music) based on what it was previously trainedFocused on autonomous decision-making & action execution, thus able to perform complicated tasks & reasining, streamline processes and fulfil preset objectives
Goal orientationTask-specific & follows instructions that are setGoal-driven & is able to formulate tasks and solve problems on its own
Learning mechanismUses data-driven techniques, e.g. pattern recognition, transformers, generative adversarial networks (GANs)Uses optimization strategies, e.g. reinforcement learning & rules-based approaches to gradually enhance decision-making skills
Learning capabilitiesLearns within certain set rules, or may not learnContinuously learns & improves from feedback and experience
Level of autonomyLimited autonomy, mainly powered by human inputHighly autonomous, can act independently based on its context-awareness
Decision-making processPre-programmed responses to inputs or through previously trained dataBased on complex reasoning and analysis
ComplexityHandles simpler, more structured tasksHandles complex, dynamic environments
Interactions with environmentReacts to set inputs but does not adaptActively adapts to its environment, surrroundings and changes
Responsiveness to changeLimited ability to adapt to new situationsChanges its goals, methods and approaches autonomously
Application areasCreative industries like marketing, visual design, recommendation models, and entertainmentDriverless cars, financial trading, healthcare diagnostics, robotic process automation (RPA)
Sources:
– Lisowski E. AI agents vs agentic AI: What’s the difference and why does it matter? Medium. 18 Dec 2024.
E&ICT Academy, IITK. Agentic AI vs generative AI: Key differences and use cases in 2025. 28 Jan 2025.

Advantages of DeepSeek: Efficiency and Eco-Friendliness

DeepSeek’s “mixture of experts” approach is not only cost-effective but also eco-friendly. By activating only the necessary computing resources for specific tasks, it can dramatically reduce energy consumption compared to traditional Gen AI models, which often require extensive computational power. This energy efficiency definitely aligns with growing global “green” efforts to create sustainable technologies, thus making DeepSeek a potential leader in eco-conscious AI development.

Furthermore, through lower development costs, DeepSeek potentially demonstrates that smaller organizations can also harness powerful AI capabilities without incurring exorbitant expenses. These advantages may potentially position DeepSeek as a transformative force for the tech industry in terms of both technological innovation and environmental sustainability.

Challenges and Opportunities

While the potential of Agentic AI is immense, its path is not without challenges. Many parties are now trying to find out the flaws of this product. Furthermore, ethical considerations, combatting cybersecurity threats, and the need for robust governance frameworks will play a critical role in determining how this technology evolves. DeepSeek’s recent cyberattack underscores the importance of continuously maintaining robust cybersecurity frameworks in AI systems in order to combat against malicious actors. Moreover, fostering international collaboration and transparency in AI development will be essential to mitigate risks and ensure equitable access to these transformative technologies. DeepSeek’s success also poses an important question: How can organizations and nations balance innovation with accountability?

Conclusion

DeepSeek’s emergence marks a turning point in the AI landscape. Its innovations are not just about efficiency and performance but also about redefining what AI can achieve in the broader context of digital and technological advancements. Like Vision in Marvel’s “Vision Quest”, DeepSeek can potentially be a tech anti-hero who embodies the complexity of technological advancement — its power to uplift, and its potential to disrupt.

As we anticipate more revelations in the age of Agentic AI, the lessons unveiled from DeepSeek’s journey and its evolution will be instrumental in shaping a future where AI is not only powerful but also ethical, secure and inclusive. DeepSeek will probably be the first of many who will lead us into the Agentic AI era. How the world responds to this new frontier will define the trajectory of AI technology and its impact on the human society for years to come.