As we reflect on the final day of 2024, the year that marked significant strides in the world of large-scale models, we witness a transformation where innovation leaped from zero to one, even as the parameter scales saw little expansion.
In the landscape of tech giants and emerging players, performances were varied and notable.
NVIDIA, the GPU powerhouse, saw its market value soar to new heights. Yet, it wasn’t without its hiccups as the B200 chip faced design flaws causing delays in production and delivery. The Rabbit R1, a device that quickly gained fame, served as a stark reminder that practicality and safety cannot be overlooked in the realm of AI hardware.
Domestically, the landscape was vibrant with over 252 large models recorded, more than half of which were embedded in various industries, focusing largely on education and entertainment. The “Six Small Tigers,” including companies like Zhipu AI, Yue Zhi An Mian, and Minimax, carved out distinct niches for themselves, striving to stay afloat in a competitive market. DeepSeek’s open-source V3 model outperformed Meta’s Llama 3.1, earning the title of the strongest open-source model.
China’s tech behemoths also reported impressive achievements. Baidu’s Wenxin Yanyan boasted an average daily model invocation of 1.5 billion, serving a user base of 430 million. Alibaba’s Tongyi Qianwen open-source model Qwen2.5 outperformed Llama 405B. Tencent’s Hunyuan large model made strides in multimodal capabilities, open-sourcing a video generation model with 13 billion parameters. ByteDance’s Doubao large model achieved a daily active user count of 7.5 million, with comprehensive product coverage.
Exciting directions for the future emerged with the demonstration of fully autonomous driving. Elon Musk showcased a completely driverless Robotaxi, heralding the era of L4-level autonomous driving. The robotics field, ranging from the high-end Atlas to household robotic dogs, aligned neatly with the advancements in large models. The releases of products like Sora and Veo highlighted the potential of visually simulating worlds through text adjustments.
In summary, the year 2024 saw the dawn of the large model era, bringing forth a plethora of astonishing innovations. Entrepreneurs pressed forward with determination, confident that great products would eventually see the light of day.
—
Below is the formatted content for a WordPress blog post:
In the year 2024, the landscape of large-scale models was marked by continuous innovation. Despite the challenges faced with the B200 chip, NVIDIA remained the unchallenged leader in the GPU domain. The case of Rabbit R1 was a crucial reminder that the essence of AI hardware lies in its practicality and safety.
Within our domestic market, a surge was witnessed in the registration of large models. The “Six Small Tigers” defined their positions with clarity, fighting to maintain their ground amidst fierce competition. DeepSeek’s V3 model stood out as the most potent open-source model, showcasing China’s strength in the field of large-scale models.
Chinese tech giants made significant strides in AI models and products, with Baidu, Alibaba, Tencent, and ByteDance each reporting remarkable progress.
Applications in fully autonomous driving, robotics, and video generation captured the world’s attention. The unveiling of Elon Musk’s fully autonomous Robotaxi, the robotic dogs from UST, and products like Sora opened our eyes to the boundless possibilities of large model applications.
As 2024 ushers in the beginning of the large model era, it has already given birth to numerous innovations. In this new epoch, entrepreneurs will continue their march forward, ever hopeful that exceptional products are just around the bend.
