Barcelona, Spain – In a significant leap forward for Chinese artificial intelligence, Zhipu AI has unveiled its latest model, GLM-4-Plus, at the prestigious KDD 2024 conference. The model has shown remarkable performance, surpassing GPT-4o in several tasks and establishing China’s growing dominance in the global AI landscape.
The KDD (Knowledge Discovery and Data Mining) conference, held in Barcelona, Spain, is a leading event in the field of data mining and machine learning. This year, it has been a stage for Chinese research teams and tech companies to showcase their latest advancements. Among them, Zhipu AI has made a powerful impression with its 超大杯 (extra-large cup) model family, including the GLM-4-Plus, which has been tested and proven to be a game-changer.
Surpassing GPT-4o
The GLM-4-Plus model, developed entirely in-house by Zhipu AI, has demonstrated impressive capabilities in language understanding, instruction following, and long-text processing. In direct comparisons with GPT-4o, the GLM-4-Plus has managed to not only match but also exceed its performance in most tasks. This is a significant milestone for Zhipu AI, marking a new era of AI development in China.
Dr. Gu Xiaotao, a leading researcher at Zhipu AI, introduced the bilingual conversational robot ChatGLM at the Large Language Model Day on August 29. ChatGLM, which supports both Chinese and English, has become a symbol of China’s technological prowess. Dr. Gu also highlighted the significant upgrade to the company’s base model, the GLM-4-Plus.
New Models and Features
In addition to the GLM-4-Plus, Zhipu AI has also released the CogView-3-Plus, a text-to-image model that boasts performance comparable to the best-in-class MJ-V6 and FLUX models. The company has also introduced the GLM-4V-Plus, an image/video understanding model that offers exceptional image recognition and time-aware video comprehension capabilities. Once launched on an open platform, it will be the first general video understanding model API in China.
Furthermore, Zhipu AI has open-sourced a larger version of its video generation model, CogVideoX 5B, which outperforms existing open-source video generation models and is considered the best choice in its category.
Real-World Testing
The performance of GLM-4-Plus was put to the test with a series of real-world tasks, including general knowledge, logic reasoning, and visual understanding. The model showed a strong grasp of logical reasoning, successfully solving complex problems that have previously stumped other AI models. It also demonstrated an impressive ability to understand and analyze images and videos, often achieving results similar to human comprehension.
For instance, when presented with a comic strip about NVIDIA, GLM-4V-Plus accurately interpreted the metaphor of the AI boom as a gold rush, identifying NVIDIA as the provider of shovels to other AI companies. Additionally, the model was able to describe the attire, expressions, and relationships of multiple characters in a meme, providing a nuanced understanding of the content.
Video Understanding and Generation
In a practical test, GLM-4V-Plus was able to understand complex video content, such as a basketball game clip, and provide detailed summaries, make inferences, and answer time-related questions. This capability positions it as a powerful tool for video analysis and understanding.
The model’s ability to generate HTML code from a website screenshot is another testament to its versatility. It accurately identified and named content modules, such as logos, banners, and news sections, and used modern layout techniques like flex, making it a valuable asset for web development.
Consolidating China’s Position
With these advancements, Zhipu AI is not only consolidating China’s position as a leader in the global AI field but also setting new standards for what AI models can achieve. The company’s dedication to innovation and its commitment to developing cutting-edge technologies are clear, and the world is taking notice.
As China continues to invest heavily in AI research and development, models like GLM-4-Plus are a testament to the nation’s growing expertise and influence in the industry. The future looks promising for Zhipu AI and the broader Chinese AI community, with the potential to shape the global AI landscape for years to come.
Views: 0