DeepSeek Decoded: Unraveling the myths behind the Chinese AI startup’s rise

DeepSeek has been making waves in the tech industry ever since it released its AI reasoning model, R1. The model’s efficiency and cost-effectiveness have put it in direct competition with OpenAI’s models, sparking both excitement and scrutiny. While many are hailing DeepSeek as a major disruptor, others are questioning how it built its AI and what this means for the global AI race.

Let’s break down five common myths surrounding DeepSeek’s rise and the reality behind them:

Myth #1: DeepSeek’s AI Models Signal the Arrival of AGI

While DeepSeek’s R1 model is an impressive advancement, it does not indicate that Artificial General Intelligence (AGI) is within reach. AGI refers to an AI system capable of reasoning and problem-solving at a human level across various domains. No company, including OpenAI, has yet achieved this milestone.

DeepSeek was initially a research unit of a Chinese hedge fund, High-Flyer, before transitioning into an AI company. Although R1 is a step forward in AI efficiency, experts believe AGI is still several breakthroughs away. NYU professor and AI expert Gary Marcus has noted that achieving AGI requires significant advancements beyond what DeepSeek has demonstrated.

Myth #2: DeepSeek’s Success Proves US Export Controls Are Ineffective

While DeepSeek has managed to innovate despite US restrictions on advanced GPUs, these controls still pose a challenge to China’s AI ambitions.

DeepSeek reportedly stockpiled 10,000 Nvidia A100 GPUs before restrictions took effect, allowing them to develop their AI models. However, experts argue that while the company has found ways to optimize efficiency, it still lacks access to cutting-edge AI hardware. This could slow China’s ability to scale AI models in the long run. AI policy expert Miles Brundage pointed out that while necessity drove DeepSeek to optimize, they would still benefit from access to more advanced chips.

Myth #3: DeepSeek is a Major Threat to Nvidia

While DeepSeek’s efficiency has raised concerns for Nvidia, the chip giant is unlikely to suffer in the long run.

Following DeepSeek’s announcement, Nvidia’s stock initially took a hit, dropping 17% and wiping out nearly $600 billion in market value. However, industry leaders believe DeepSeek’s advancements could actually increase demand for GPUs rather than reduce it. Microsoft CEO Satya Nadella highlighted Jevons Paradox, which suggests that making AI more efficient could lead to greater overall GPU consumption.

Myth #4: DeepSeek R1 is a Fully Open-Source Model

While DeepSeek R1 can be downloaded, modified, and used freely, it does not meet the widely accepted definition of open-source AI.

The model’s architecture and weights have been released under a permissive MIT license, allowing public use. However, true open-source AI requires transparency about training data and complete access to the training code—details DeepSeek has not provided. Experts believe companies hesitate to disclose training data due to potential copyright risks.

Myth #5: DeepSeek’s AI Models Pose Greater Privacy Risks

DeepSeek’s AI poses no greater privacy threat than other large language models (LLMs).

Concerns have arisen due to DeepSeek’s Chinese origins and data policies. The company states that its servers are located in China, raising potential regulatory concerns. However, some experts, including Perplexity AI’s CEO Aravind Srinivas, argue that users can download and run DeepSeek’s R1 model locally, ensuring complete privacy. Perplexity AI has also hosted R1 in US and EU data centers, claiming its version is free from censorship.

Final Thoughts

DeepSeek’s rise marks an important moment in AI development, showcasing China’s ability to produce competitive models. However, the road to AGI remains long, and challenges like hardware access and regulatory scrutiny will continue to shape DeepSeek’s future. Whether it can maintain its momentum and challenge industry leaders like OpenAI and Nvidia remains to be seen.