Academician Zhang Bo: Four Future Development Directions for Large Models

Recently, Academician Zhang Bo, an academician of the Chinese Academy of Sciences and honorary president of the Institute of Artificial Intelligence at Tsinghua University, stated in his speech at the ISC.AI 2024 12th Internet Security Conference that current artificial intelligence lacks theory, only having developed models and algorithms targeted at specific fields. Both software and hardware are specialized, and the market is small, which is why a large-scale artificial intelligence industry has yet to develop; this is the crux of the issue.Academician Zhang Bo: Four Future Development Directions for Large ModelsAcademician Zhang Bo of the Chinese Academy of Sciences and honorary president of the Institute of Artificial Intelligence at Tsinghua UniversityAt 89 years old, Academician Zhang Bo has cultivated a group of artificial intelligence talents at Tsinghua University over the past few decades and is one of the founders of the artificial intelligence discipline in China. Many hot “Tsinghua system” large model companies such as Shenshu Technology, Zhipu AI, Mianbi Intelligent, Kimi, etc., have benefited from the technical foundation laid at Tsinghua, with core technical talents directly or indirectly inheriting from Zhang Bo.In this speech, Academician Zhang not only pointed out the flaws and problems existing in current artificial intelligence technology but also provided directions for future improvements.

When Considering Basic Models, Consider 3 Major Capabilities and 1 Major Flaw

According to Academician Zhang, due to theoretical limitations, the previous stage of the artificial intelligence industry must develop in conjunction with specific application fields, thus the artificial intelligence developed in this stage is considered specialized artificial intelligence, or “weak” artificial intelligence. However, he also pointed out that current basic models have achieved universality in language issues, “When we consider basic models, we need to consider 3 major capabilities and 1 major flaw; this is very important and is the starting point for considering future industrial development.”He explained that the strength of large language models lies in their powerful language generation ability, strong human-machine natural interaction ability, and strong ability to draw inferences. “The language generation of large language models belongs to the open domain, capable of generating diverse results that all outputs can be understood by humans. Even when it is ‘nonsense’, we can still understand what it is saying, which is very important. Humans and machines can engage in natural language dialogue in an open domain, a goal we previously thought would take generations to achieve, but unexpectedly, this goal was reached in 2020.”Academician Zhang stated that the flaw of large models is ‘hallucination’. “Because we require diverse outputs, it inevitably produces errors. This error is very different from the errors produced by machines; machine errors are often controllable, but this error is inherent and will occur, and we cannot control it. Therefore, this is also an issue we need to consider for its applications in the future.”Combining the 3 major capabilities and 1 major flaw, Academician Zhang summarized the current suitable application scenarios for large models: they must have a high tolerance for errors. He stated that from an industrial perspective, the application of large models presents a ‘U’ shape—planning and design at the front require content diversity, while service and recommendations at the back also require diversity, with a high tolerance for errors. However, the middle part needs to be considered based on the situation.Despite the issues, Academician Zhang still stated that regardless, “models must be used,” because with a model foundation, the efficiency and quality of applications will definitely improve. In the past, application scenarios were developed on empty computers providing services; an empty computer is equivalent to being illiterate. Now with large models, the platform is at least a high school student, and development efficiency will certainly improve; future directions will definitely be like this.”Academician Zhang focused on analyzing the root causes of hallucinations, as he believes the fundamental limitation of models is that all the work done by machines is externally driven, taught by humans how to do it, rather than being proactive. Meanwhile, the results they generate are heavily influenced by prompts, which is a significant difference from humans completing work under internal intentions.

Four Future Development Directions for Large Models: Alignment, Multimodal, Agents, Embodied Intelligence

Academician Zhang introduced four future development directions for large models, which are crucial for improving large models.The first is alignment with humans,“Large models lack the ability to judge right from wrong and cannot self-update; they are updated under human drive. If this point is not broken through, machines cannot self-evolve. Large models require external prompts, so correcting the errors of large models under human drive is our first task.”The second is multimodal generation,“Multimodal generation will be very important for industrial development in the future because large models are mainly generating text, but if we use the same methods to generate images, sounds, videos, and code, the level of generation will be close to that of humans. The reason we can generate images so well now is mainly due to linking images with text. Therefore, the essential breakthrough is in text processing.”The third is the concept of AI Agents,“We need to combine large models with surrounding virtual environments, allowing the environment to prompt them about their errors because one only knows right from wrong after doing something. Thus, the concept of agents is very important, allowing the environment to prompt agents and giving them a chance for reflection to correct errors.”The fourth is embodied intelligence,“By adding robots, large models can also work in the physical world. How to develop general-purpose robots in the future? I believe it should be ‘software general, hardware diverse’. While Musk promotes humanoid robots, I believe the future is not limited to humanoid robots.”In Academician Zhang’s view, developing the third generation of artificial intelligence requires first establishing a theory, as the existence of large models lacks theoretical explanations, leading to various confusions and misunderstandings. As machines grow larger in scale, if theories cannot explain them, it will cause panic. Achieving safe, controllable, trustworthy, reliable, and scalable artificial intelligence technology is vital; until this field is fully developed, artificial intelligence will always pose safety issues.

Source: Beijing News Beike Finance

END

Academician Zhang Bo: Four Future Development Directions for Large Models● The grand event is coming! The 2024 Embodied Intelligence Robot Development Conference ● Highlights of 55 listed robot companies’ 2023 annual reports: mixed results, reshuffling intensifies ● Highlights of the Embodied Intelligence Robot Development Conference: thematic advance, high specifications, excellent roadshows

● The wave of embodied intelligent robots is coming, and another industry grand event is about to set sail

● Dialogue with Tsinghua University’s Zhao Mingguo: Accelerating the evolution to replicate Boston Dynamics’ movements; what we need is confidence and innovation!

● The preliminary results of the 2023 National Science and Technology Awards have been announced! 9 985 universities have no winners! Xi’an Jiaotong University and Huazhong University of Science and Technology are in the top three! (Full list attached)

● Unprecedented! American engineers collaborate with ChatGPT4 to design artificial intelligence chips

● Russian President Putin approves the new version of the “2030 National Strategy for Artificial Intelligence Development”

● A small town in Denmark with a population of less than 200,000! How did Odense become a global robot hub?

● The “Year of Commercialization” has begun, and humanoid robots gain new strength as players

● Swiss researchers develop a new type of artificial muscle, lighter, safer, and more robust!

● The EU terminates Amazon’s acquisition of iRobot; what will happen to the once-dominant robot vacuum cleaner giant?

● The top ten news in the robot industry for 2023

● A bipedal robot driven by muscle tissue has been developed, marking a breakthrough in biological hybrid robots!

● Professionals discuss the robot-as-a-service model—the future of automation

● A comprehensive overview of China’s humanoid robot R&D teams

● Under the heat of humanoid robots, the competition between advancement and resistance

● Who is the most outstanding? Highlights from 53 listed robot companies’ 2023 mid-year reports

● Download the half-year reports of 53 listed robot companies (PDF attached)

● Academician Report|Pan Yunhe: The Behavioral Intelligence and Product Intelligence of Artificial Intelligence

● Academician discusses the new momentum for promoting collaborative intelligent manufacturing in robotics

● Academician discusses six key technologies for innovative design in robotics

● Ximu Technology broadens new angles in humanoid robot research

● Academician discusses the dual-driven model for the future development of artificial intelligence

● Academician discusses how institutional intelligence brings “Transformers” from the screen to reality

Academician Zhang Bo: Four Future Development Directions for Large ModelsContact Us· WeChat: myrobot2001· Contact Number: 18100123515

Leave a Comment