DeepSeek's Janus

The Jindao foreign exchangedomestic artificial intelligence company DeepSeek has recently become the focus of global attention in the AI industry following the release of its open-source multimodal AI model, Janus-Pro. According to Hugging Face's January 28 news, Janus-Pro showcased outstanding performance by defeating OpenAI's DALL-E 3 and Stable Diffusion in the GenEval and DPG-Bench benchmarks with its innovative design.

Meanwhile, DeepSeek issued a statement via its official service status, stating that its online service has been subjected to large-scale malicious attacks recently, which might cause the registration system to be busy. DeepSeek affirmed that registered users could log in normally and expressed gratitude for users' understanding and support.

The Innovative Design of Janus-Pro

Janus-Pro is a novel multimodal autoregressive framework capable of unifying both understanding and generation tasks. DeepSeek explains that the model enhances task adaptability and model flexibility by separating visual encoding into "understanding" and "generation" pathways while still utilizing a single Transformer architecture. This design effectively resolves the role conflict of visual encoders in traditional methods, making Janus-Pro more advantageous in multitasking.

The model is based on DeepSeek-LLM-1.5b-base and DeepSeek-LLM-7b-base, supports image input with a resolution of 384x384, and further optimizes image generation capabilities through a self-developed tokenizer, demonstrating immense potential in multimodal applications.

Far-reaching Impact on Global Market

Due to its low cost and high performance, DeepSeek's Janus-Pro has had significant implications for the market. It is reported that the model's training cost is much lower than existing giants, achievable with limited hardware resources, thus threatening the dependency on high-end GPUs and impacting the valuation of tech stocks, especially chip stocks.

On January 28, the US stock technology sector was generally under pressure, with the Nasdaq index plunging 3.07%, and Nvidia's stock price plummeted 16.86%, marking the largest single-day market value evaporation in history, losing about $588.862 billion. TSMC dropped more than 13%, AMD fell 6.37%, and Broadcom dropped 17%. Additionally, Google's and Microsoft's stock prices fell by 4.03% and 2.14%, respectively. Notably, Apple's stock rose by 3.25%, reclaiming the top spot in global market capitalization with a market value of $3.5 trillion.

Global Popularity of DeepSeek

Not only causing ripples in the capital market, but DeepSeek's new model has also sparked a frenzy in the application market. On January 27, its AI mobile application topped the free chart in Apple's App Store in the US, surpassing ChatGPT and ranking first in the Chinese market as well.

Market analysts point out that DeepSeek's technological innovation has changed the game rules of the AI large model industry. The low-cost, high-efficiency model suggests that future demand for computing power and hardware in the AI industry will significantly decrease, potentially reshaping the global technology landscape.

DeepSeek's ongoing innovation not only establishes it as a dark horse in the AI field but also sparks widespread discussion globally. Whether this domestic AI technology can continue to lead the industry in the future is worth continuous attention.

The market carries risks, and investment should be cautious. This article does not constitute personal investment advice and has not taken into account individual users' specific investment goals, financial situations, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their particular circumstances. Investing based on this is at one's own responsibility.