DeepSeek: The Al poster child
DeepSeek emerges as a cost-effective AI innovator, leveraging MoE models and open-source strategies to challenge giants like ChatGPT and Google Gemini. Discover how its 'wait-and-learn' approach is reshaping the AI landscape.
DeepSeek
Out of curiosity, we just asked Chat GPT if there are chances that other tools just like yours will get ahead in the race. We got this result: “DeepSeek, or any future AI model or system, might outperform me in specific ways depending on its focus, scope, and technological advancements.” Surprising or not DeepSeek is now in the spotlight.
The Wait and learn strategy
AI proved that innovation can come from unexpected quarters. Up until January 2025, the West led the AI race, relying on breakthroughs in architecture, publishing research like Google's pivotal "Attention is All You Need" paper, and investing billions in R&D.
DeepSeek has emerged as the latest AI tool to leverage Mixture-of-Experts (MoE) models and transformers, much like its counterparts. What makes DeepSeek unique is its “wait-and-learn” strategy, which involves strategically piecing together existing methods to dramatically reduce costs.
We heard that it uses “knowledge distillation,” where smaller models are trained using insights from larger ones. DeepSeek’s developers, through their DeepSeek R1 AI model, seem to have found a way to achieve this performance at a fraction of the expense. To be precise, it uses an RL-driven approach and a 671B MoE architecture to deliver results.
Cost wars begin
A cost war across the globe has competitors sweating. Until last month, AI companies claimed most spending was allocated to infrastructure for training and data collection. For comparison, OpenAI’s introduction of their $200/month pro plan got many people to use the premium version, but DeepSeek managed to release free and competitive alternative a mere 45 days later. This gave them a standing ovation and they must be appreciated for their impeccable timing. Their ability to undercut and respond swiftly to evolving Ai market is getting competitors like ChatGPT, Google Gemini, Jasper, Claude Ai and Perplexity sweating.
What’s most intriguing is DeepSeek’s strategic decision to open-source its models unlike others who claim to do so but haven’t. Critics might dismiss it as a tactical response to accusations of being a "clone," but in reality, the wait-and-learn strategy worked. Open sourcing is working in DeepSeek’s favour, gaining goodwill, attracting developer communities, and expanding its reach globally, sidelining massive marketing budgets
Security
DeepSeek’s rise reminds us that AI’s future isn’t solely dictated by who spends the most but by who innovates smarter. As this rivalry unfolds, competitors are spotlighting data security. However, the Qodo team who is currently hosting DeepSeek on their servers say that none of the data reaches China.
Ai Bubble popping?
Nvidia was first in line to lose $600B in the market. For open-source enthusiasts and tinkerers, the future looks bright as giants falter. 2025 will be a year to watch.