Recently, Zhang Peng, the CEO of BrainMap AI, was invited to appear on the program "Dialogue" at the CCTV Financial Program Center. Alongside him were Zhou Hongyi, founder of 360 Group; Xu Zhigen, founder and CEO of Jiuguang Intelligence; Li Chuyuan, Secretary of the Party Committee and Chairman of Guangzhou Pharmaceutical Group; He Yifan, Deputy Editor-in-Chief of China Entrepreneur Magazine; Zhang Zhiwei, Executive Editor-in-Chief of Huaxia Times; Zhu Zhuo, Deputy Editor-in-Chief of "Finance" Magazine; and Joshua Bengio, Turing Award winner and professor at the University of Montreal. They collectively explored and discussed topics such as the technological development and commercialization paths of artificial intelligence under the theme "AI + Future Speculation."
Zhang Peng likens current advancements in artificial intelligence to the "improvement stage of steam engines." He believes that the industry has generally accepted that large models are an effective path toward AGI (Artificial General Intelligence), but there are still many issues worth exploring. Setting the goal at AGI, breakthroughs in each of these issues could potentially lead to leaps in development, with no visible ceiling at the moment.
Zhang Peng also believes that with the advent of large models, communication between humans and machines will become increasingly natural language-oriented, which is the direction that humanity hopes and will inevitably move towards. However, he suggests that the definition of programmers in the future will change, with traditional programmers becoming fewer, and a need for more so-called "new-style programmers" who are based on new communication methods between machines and humans. He emphasizes that when discussing issues, it is easy to overlook another factor — time, leading to an overestimation of the short-term value of technology and an underestimation of its long-term value.
In the realm of empowering industries with large-scale models, Zhang Peng emphasizes, "We have already seen the dawn, and 2024 should be the starting year for launching towards AGI." According to Zhang, large models themselves have had a tremendous impact on improving robotic algorithms, evident in aspects such as planning, rule completeness, and generalization. However, even though large models boast a comprehensive knowledge base, bridging the gap with hardware to address real-world issues, such as those pertaining to daily life, will take some time. By 2024, there is a high probability of seeing applications capable of solving practical problems emerge. Zhang believes that the capabilities derived from the improvement of general-purpose large models enable innovation, potentially leading to what's known as "large model-native applications" or "large model-native scenarios." Throughout this process, both technology and industries need to evolve and find practical applications.
Throughout its history, Zhiku AI has been committed to creating a new generation of cognitive intelligent large-scale models, focusing on innovating large models in China. Zhiku AI collaborated to develop the bilingual billion-scale ultra-large pre-training model GLM-130B, and based on this, launched the conversational model ChatGLM, open-sourced the single-card version model ChatGLM-6B, and the new generation base model GLM-4. Additionally, Zhiku AI has also developed the AIGC model and product matrix, including the AI efficiency assistant Zhiku ClearSpeech, the high-efficiency code model CodeGeeX, the multimodal understanding model CogVLM, and the text-image model CogView.
Zhiku AI has always adhered to the complete independent research and development of large models. Based on the fully self-developed pre-training framework GLM, Zhiku AI has created multiple widely recognized model products. Currently, the billion-scale model GLM-130B has received usage demands from over 70 countries and more than 1,000 research institutions. The open-source model ChatGLM-6B has been downloaded globally over 13 million times.
In 2024, the starting year for the charge towards AGI, Zhiku AI will continue to advance independent innovation in large-scale model technology, boldly striving towards general artificial intelligence with new breakthroughs.
