Introduction

A seismic event has rocked the AI community. DeepSeek, a Chinese company, has ignited a global stir, reshaping the AI landscape dramatically.

Furthermore, DeepSeek has steered autonomous driving technology in a novel direction!

(For reference, click: "Lei Jun's YU7: Taking on Musk with AI and Autonomous Driving, Aspiring to Surpass 'Tesla'?")

I. The 'Cost-Effective Champion' of the AI World

Let me first provide a brief overview of DeepSeek's 'journey'.

Founded on July 17, 2023, with the backing of a Quant Hedge Fund, DeepSeek has been akin to a secretive knight, wielding an unseen manual of martial arts, and making waves quietly but powerfully.

In May 2024, it open-sourced the second-generation MoE large model, DeepSeek-V2, directly competing with GPT-4 Turbo at a mere 1% of the price.

This cost-effectiveness has earned it the moniker 'Cost-Effective Champion of the AI World'.

Following this, it continued to excel. On November 20, the preview version of the inference model DeepSeek-R1-Lite was launched; on December 26, the first version of the model DeepSeek-V3 was launched and open-sourced simultaneously, each step taken with conviction, each move captivating the audience.

II. The 'Disruptor' in Global Large Model Rankings

On the overseas large model ranking platform Arena, DeepSeek-R1's performance has been nothing short of phenomenal.

Its benchmark test soared to third place across all large model categories, and in the StyleCtrl category, it tied for first place with OpenAI o1.

With a score of 1357, it even surpassed OpenAI o1's 1352 points slightly.

Imagine a novice in a martial arts tournament filled with experts, directly advancing to the top three and tying with a martial arts master in an individual event. How could this fail to astonish?

Prior to this, after excluding closed-source models like OpenAI o1, DeepSeek-V3 ranked first in the open-source model category.

These achievements have been like thunderous blows, shaking up the 'status quo' of the global AI market, compelling everyone to reassess this 'disruptor' from China.

III. The 'Unsettling Force' That Rattles U.S. Stocks

DeepSeek-R1's outstanding performance has been akin to a colossal rock tossed into a tranquil lake, sending ripples through the U.S. stock market.

NVIDIA's stock price plummeted by 3.12% on January 24, to $142.62 per share, marking the company's steepest decline since its products fell short of expectations at the CES exhibition earlier in the year.

Behind this drop, DeepSeek's 'contribution' cannot be overlooked.

Some U.S. stock market influencers and renowned media outlets have deemed DeepSeek as 'the biggest threat to the U.S. stock market'.

Why so?

Because DeepSeek's emergence has shattered conventional perceptions of the AI industry.

If defeating OpenAI truly requires only $55 million (including 5,000 H800 GPUs and $5 million in pre-training costs), then the commercialization of this industry might progress much faster than anticipated.

This undoubtedly poses a significant challenge to American companies that have heavily invested in the AI field.

It was once believed that to dominate the AI world, one must possess massive funds and resources. However, DeepSeek has demonstrated through actions that it can compete on equal footing with, or even surpass, these giants without incurring such exorbitant costs.

How could this fail to instill fear in those American AI giants?

How could this fail to prompt investors to reevaluate the investment logic of this industry?

IV. The Controversial 'Rising Star' of AI

Fame invariably attracts controversy, and DeepSeek is no exception. While it has garnered global attention, it has also sparked numerous questions and conspiracy theories.

Some North American AI practitioners find it hard to believe that DeepSeek's pre-training costs amount to a mere $5.5 million.

Alexandr Wang, the founder of AI tech startup Scale AI, even claimed at the Davos Forum that DeepSeek has stockpiled 50,000 NVIDIA H100 GPUs but has not disclosed this due to the chip embargo. Despite being baseless, this claim has gained widespread circulation.

Others suspect that DeepSeek might be overfitting the benchmark tests and might not perform optimally in practical applications.

After all, for those in Silicon Valley accustomed to high-cost, large-scale AI model investments, it's difficult to fathom that DeepSeek can train such a powerful model at such a low cost.

Fortunately, DeepSeek has opted for the MIT license, the most open standard for open-source software. Everyone can study and verify whether the training costs are indeed that low.

Meta has even initiated a project to reproduce its large model using DeepSeek's open-source papers, attempting to unravel its mysteries.

V. Ushering in a New Era for Autonomous Driving

Apart from shining in the realm of large models, DeepSeek also holds considerable potential in autonomous driving.

In autonomous driving, AI models need to swiftly and accurately process and make decisions based on vast amounts of road condition information, placing extremely high demands on the model's inference capabilities and efficiency.

DeepSeek-R1's robust inference capabilities and low inference costs can precisely meet these autonomous driving requirements.

Perhaps one day in the future, all vehicles on the road will be equipped with DeepSeek's AI models. They will function like seasoned drivers, capable of swiftly responding to various complex road conditions, significantly enhancing driving safety and efficiency.

This isn't a scene from a science fiction movie. With the continuous development and application of DeepSeek's technology, all of this may soon become a reality.

In summary, Unmanned Vehicles Are Coming (WeChat official account: Autonomous vehicles have arrived) believes that DeepSeek's emergence undoubtedly brings new vigor and transformation to the global AI industry. It's like a 'rebel' breaking the mold, proclaiming to the world in its unique way that Chinese AI can not only keep pace but also lead the trend and even empower autonomous driving!

What surprises will it bring us in the future? Let's wait and see.

Dear readers, what do you think?