The recent excitement surrounding DeepSeek, an advanced large language model (LLM), is understandable given the significantly improved efficiency it brings to the space. However, some reactions to its release seem to misinterpret the magnitude of its impact. DeepSeek represents a leap forward in the expected trajectory of LLM development, but it does not signal a revolutionary shift toward artificial general intelligence (AGI), nor does it mark a sudden transformation in the center of gravity of AI innovation.
Rather, DeepSeek’s achievement is a natural progression along a well-charted path—one of exponential growth in AI technology. It is not a disruptive paradigm shift, but a powerful reminder of the accelerating pace of technological change.
DeepSeek’s efficiency gains: A leap along the expected trajectory
The core of the excitement surrounding DeepSeek lies in its impressive efficiency improvements. Its innovations are largely about making LLMs faster and cheaper, which has significant implications for the economics and accessibility of AI models. However, despite the buzz, these advancements are not fundamentally new, but rather refinements of existing approaches.
In the 1990s, high-end computer graphics rendering required supercomputers. Today, smartphones are capable of the same task. Similarly, facial recognition—once a niche, high-cost technology—has now become a ubiquitous, off-the-shelf feature in smartphones. DeepSeek fits within this pattern of technology: an optimization of existing capabilities that delivers efficiency, but not a new, groundbreaking approach.
For those familiar with the principles of technological growth, this rapid progress isn’t unexpected. The theory of Technological Singularity, which posits accelerating progress in key areas like AI, predicts that breakthroughs will become more frequent as we approach the point of singularity. DeepSeek is just one moment in this ongoing trend, and its role is to make existing AI technologies more accessible and efficient, rather than representing a sudden leap into new capabilities.
DeepSeek’s innovations: Architectural tweaks, not a leap to AGI
DeepSeek’s main contribution is in optimizing the efficiency of large language models, particularly through its Mixture of Experts (MoE) architecture. MoE is a well-established ensemble learning technique that has been utilized in AI research for years. What DeepSeek has done particularly well is refine this technique, incorporating other efficiency measures to minimize computational costs and make LLMs more affordable.
- Parameter efficiency: DeepSeek’s MoE design activates only 37 billion of its 671 billion parameters at any given time, reducing the computational requirements to just 1/18th of traditional LLMs.
- Reinforcement learning for reasoning: DeepSeek’s R1 model uses reinforcement learning to enhance chain-of-thought reasoning, a vital aspect of language models.
- Multi-Token training: DeepSeek-V3’s ability to predict multiple pieces of text simultaneously increases the efficiency of training.
These improvements make DeepSeek models dramatically cheaper to train and run when compared to competitors like OpenAI or Anthropic. While this is a significant step forward for the accessibility of LLMs, it remains an engineering refinement rather than a conceptual breakthrough toward AGI.
The impact of open-source AI
One of DeepSeek’s most notable decisions was to make its models open-source—a clear departure from the proprietary, walled-garden approaches of companies like OpenAI, Anthropic, and Google. This open-source approach, championed by AI researchers like Meta’s Yann LeCun, fosters a more decentralized AI ecosystem where innovation can thrive through collective development.
The economic rationale behind DeepSeek’s open-source decision is also clear. Open-source AI is not just a philosophical stance but a business strategy. By making its technology available to a broad range of researchers and developers, DeepSeek is positioning itself to benefit from services, enterprise integration, and scalable hosting rather than relying solely on the sale of proprietary models. This approach gives the global AI community access to competitive tools and reduces the stranglehold of large Western tech giants on the space.
China’s growing role in the AI race
For many, the fact that DeepSeek’s breakthrough came from China might be surprising. However, this development shouldn’t be viewed with shock or as part of a geopolitical contest. Having spent years observing China’s AI landscape, it’s clear that the country has made substantial investments in AI research, resulting in a growing pool of talent and expertise.
Rather than framing this development as a challenge to Western dominance, it should be seen as a sign of the increasingly global nature of AI research. Open collaboration, not nationalistic competition, is the most promising path toward the responsible and ethical development of AGI. A decentralized, globally distributed effort is far more likely to produce an AGI that benefits all of humanity, rather than one that serves the interests of a single nation or corporation.
The broader implications of DeepSeek: Looking beyond LLMs
While much of the excitement around DeepSeek revolves around its efficiency in the LLM space, it’s crucial to step back and consider the broader implications of this development.
Despite their impressive capabilities, transformer-based models like LLMs are still far from achieving AGI. They lack essential qualities such as grounded compositional abstraction and self-directed reasoning, which are necessary for general intelligence. While LLMs can automate a wide range of economic tasks and integrate into various industries, they do not represent the core of AGI development.
If AGI is to emerge in the next decade, it is unlikely to be based purely on transformer architecture. Alternative models, such as OpenCog Hyperon or neuromorphic computing, may be more fundamental in achieving true general intelligence.
The commoditization of LLMs will shift AI investment
DeepSeek’s efficiency gains accelerate the trend toward the commoditization of LLMs. As the costs of these models continue to drop, investors may begin to look beyond traditional LLM architectures for the next big breakthrough in AI. We may see a shift in funding toward AGI architectures that go beyond transformers, as well as investments in alternative AI hardware, such as neuromorphic chips or associative processing units.
Decentralization will shape AI’s future
As DeepSeek’s efficiency improvements make it easier to deploy AI models, they are also contributing to the broader trend of decentralizing AI architecture. With a focus on privacy, interoperability, and user control, decentralized AI will reduce our reliance on large, centralized tech companies. This trend is critical for ensuring that AI serves the needs of a global population, rather than being controlled by a handful of powerful players.
DeepSeek’s place in the AI Cambrian explosion
In conclusion, while DeepSeek is a major milestone in the efficiency of LLMs, it is not a revolutionary shift in the AI landscape. Rather, it accelerates progress along a well-established trajectory. The broader impact of DeepSeek is felt in several areas:
- Pressure on incumbents: DeepSeek challenges companies like OpenAI and Anthropic to rethink their business models and find new ways to compete.
- Accessibility of AI: By making high-quality models more affordable, DeepSeek democratizes access to cutting-edge technology.
- Global competition: China’s increasing role in AI development signals the global nature of innovation, which is not restricted to the West.
- Exponential progress: DeepSeek is a clear example of how rapid progress in AI is becoming the norm.
Most importantly, DeepSeek serves as a reminder that while AI is progressing rapidly, true AGI is likely to emerge through new, foundational approaches rather than optimizing today’s models. As we race toward the Singularity, it is crucial to ensure that AI development remains decentralized, open, and collaborative.
DeepSeek is not AGI, but it represents a significant step forward in the ongoing journey toward transformative AI.