AI Empowers Sustainability, Large Models Create "Incalculable Value

The 2024 Cloud Computing Conference concluded as scheduled, with AI large models remaining the main protagonists of this year's event.

At the conference, Alibaba Cloud's flagship model Qwen-Max was once again upgraded, with performance approaching that of GPT-4o.

Alibaba Cloud also announced that it is centering on AI to comprehensively restructure the underlying hardware, computing, storage, networking, databases, and big data, aiming to create the strongest AI infrastructure for the AI era.

The full commitment to AI is driven by Alibaba Cloud's deep belief that generative AI has the vast potential to "gradually permeate the digital world, take over the digital world, and change the physical world."

Currently, the capabilities of large models in various disciplines such as physics, chemistry, and biology are approaching the level of a Ph.D. On the other hand, large models not only have continuous availability but also possess the ability to learn from examples, generalize, summarize complex knowledge and data, and have multimodal capabilities in audio and video.

This allows large models to help humans solve various problems to a certain extent.

At present, Alibaba Cloud is collaborating with various parties to implement AI in various industry scenarios.

An indispensable part of this is to make large models intelligent and benevolent, empowering sustainable development, education, and scientific research to seek more welfare for the people's livelihood.

In the economic and social fields, the Hou Tu large model jointly developed by the Information Center of the Ministry of Natural Resources and Alibaba Cloud is empowering the digital governance of natural resources.

In the field of astronomical research, the Xing Yu 3.0 astronomical large model based on Tong Yi has been connected to the telescope cluster, assisting in astronomical discoveries.

In the field of education, the Higher Education Press is accelerating educational reform and promoting educational equity using AI technology... AI, which is rapidly permeating all walks of life, is having a profound impact and creating incalculable value in places that many people "cannot see."

According to reports, under the close cooperation between the Information Center of the Ministry of Natural Resources (hereinafter referred to as the "Information Center") and Alibaba Cloud, the preliminary progress made by the Hou Tu large model in the natural resources industry includes: constructing the overall technical route of the natural resources industry large model; carrying out the collection and processing of natural resources industry knowledge, collecting more than 700,000 documents of natural resources basic knowledge, policies and regulations, standards and specifications, and other seven major categories; organizing the development of key technologies for the natural resources industry large model.

In the development of rich modal knowledge document parsing and processing technology, the accuracy rate of recognition for text OCR (Optical Character Recognition), paragraphs, tables, images, formulas, etc., has exceeded 90%.

Based on the Qwen series of general large models + RAG (Retrieval-Augmented Generation) and fine-tuning, the large model is infused with professional knowledge in the field of natural resources, jointly creating an intelligent entity for the natural resources industry large model.

Wu Hongtao, Deputy Director and Engineer of the Information Center, introduced at the Cloud Computing Conference that the name Hou Tu large model originates from the deity "Hou Tu" in ancient Chinese mythology, whose duty is the management of natural resources and territorial space.

Currently, the Hou Tu large model has been preliminarily applied to the work of national land change investigation.

For a long time, this work has been of great significance for comprehensively, timely, and accurately grasping the situation of land use in our country, protecting arable land, saving and using land intensively, protecting the ecological environment, and ensuring national food security.

The land change investigation is mainly divided into two stages: investigation and verification.

In the investigation phase, researchers need to go to the investigation site to conduct visits and surveys; in the verification phase, staff need to identify and check a large number of satellite images, surveying data, as well as on-site photos and videos, to verify the type of land use.

The problem is that our country will add a large number of new images and videos of land and resources every day, and it is difficult to complete all manual verifications.

In order to reduce the workload of investigators and improve the accuracy of investigation and verification, last year, the Information Center of the Ministry of Natural Resources and Alibaba Cloud launched a cooperation, jointly creating the Hou Tu natural resources industry large model based on Tong Yi base model.

Taking the investigation of agricultural land change as an example, during the field investigation, researchers can use AI tools developed based on the Hou Tu large model to take photos of the target land, and the large model can accurately identify the area of agricultural land, land use type, and other information, assisting staff in their work, and avoiding overshoot, misshoot, and missed shoots.

In the verification phase, the large model will identify land types and attributes based on existing on-site photos, automatically output verification conclusions, and assist staff in improving the accuracy and efficiency of verification.

The large model centered on people is liberating people from tedious and repetitive work.

It is reported that in the future, the Hou Tu large model will also be applied to other types of land investigation work such as arable land, garden land, forest land, grassland, and commercial land, helping the sustainable development of natural resources.

Wu Hongtao also mentioned that over the past year, the Information Center has also been exploring the application of the Hou Tu large model in scenarios such as spatial planning and land use control, resource protection, and geological disaster prevention and control, "the two-way approach of the large model base and application scenarios" will "fully support the improvement of digital governance capabilities of natural resources, to answer the must-answer questions in the process of digital governance of natural resources."

Looking at the sky, the large model measures the earth while the AI large model is also looking at the sky.

In April this year, the Artificial Intelligence Working Group of the National Astronomical Observatories of the Chinese Academy of Sciences released the new generation of astronomical large model Xing Yu 3.0, which is the first application of the large model in the field of astronomical observation.

Xing Yu 3.0 is built on Alibaba Cloud Tong Yi Qianwen open source model, and has been successfully connected to the telescope array Mini Si Tian at the Xinglong Observation Station of the National Astronomical Observatories, which can realize autonomous control of the telescope for observation and analysis of observation results, and intelligently give the next observation suggestions.

Astronomical telescopes are the "cameras" for humans to explore the universe.

Compared with single telescopes, large telescope arrays can more effectively integrate high-performance telescope resources, with lower costs and higher observation efficiency.

However, with the continuous increase in the number of telescopes, how to control large telescope arrays has become a common challenge faced by the astronomical community today.

In the past, astronomical observation mainly depended on the cooperation of observation assistants and scientific researchers, who often needed to modify observation plans according to the weather conditions of the observation site, wait for data to return in the observation room, and analyze data in real-time, and then modify observation plans according to the results.

Each link of important observation targets requires manual participation, with low efficiency and difficulty in controlling multiple telescopes at the same time.

Xing Yu 3.0 attempts to solve this problem.

Relying on the Alibaba Cloud Tong Yi Qianwen base model and Bai Lian platform, Xing Yu 3.0 has been trained based on more than 300,000 expert-corrected data, with outstanding performance in professional capabilities such as astrophysics.

When observing celestial bodies, Xing Yu 3.0 will first query the coordinate information of a certain celestial body and feedback to the observation personnel; when confirmed, Xing Yu 3.0 will automatically observe according to the plan input by the observation personnel; after a single exposure is completed, the large model will recover and process the data, judge the observation value of the target source according to the results, and recommend the next observation plan.

Compared with the past, Xing Yu 3.0 has independently completed some tasks in astronomical observation that originally required manpower, improving the efficiency of astronomical observation and allowing researchers to return to astronomical research.

In addition to assisting in astronomical observation, the National Astronomical Observatory has also cooperated with Alibaba Cloud to develop the world's first solar vertical Sora model, which can be used to detect solar flares, as well as for real-time segmentation of magnetic fields and evolution and other related research.

On the other hand, the ability of large models to process massive data is also needed in the field of astronomy.

According to Luo Ali, a researcher at the National Astronomical Observatory of the Chinese Academy of Sciences, the amount of data from astronomical observation is increasing very fast, with dozens of TB of data being generated every day, and in the next 10 years, the total amount of data in the field of astronomy will go from PB to EB level.

In order to better mine these astronomical data, the National Astronomical Observatory signed a strategic cooperation agreement with Alibaba Cloud in 2016.

Luo Ali revealed that before the methods of big data and machine learning, the global results added annually in the field of astronomy were about more than 1,000, and after having these new tools, data, and methods, the annual results have doubled.

In the future, the Xing Yu large model will be connected to the large telescope array Si Tian of the National Astronomical Observatory (Mini Si Tian is part of Si Tian), which is a national-level major astronomical infrastructure project proposed by Chinese astronomers for time-domain astronomy.

The project is expected to reach 72 meter-class telescopes by 2030, generating about 140 TB of processed data per night, becoming the most efficient sky survey project in the world.

AI's "Education for All" 2500 years ago, the thinker and educator Confucius proposed his ideal education - "education for all" and "teaching students in accordance with their abilities," the former talks about the universality of education, and the latter hopes to further achieve personalized education.

The emergence of large models is gradually promoting this ideal education to become a reality.

At present, there is an imbalance in the distribution of educational resources between urban and rural areas and regions in our country, and there is a big difference in the quality of education.

How to apply advanced technology to promote the co-construction and sharing of high-quality educational resources is an important task to promote educational equity.

Liu Chao, the president of the Higher Education Press (hereinafter referred to as "Gao Jiao She") mentioned at the Cloud Computing Conference that the practice of Gao Jiao She in educational equity 1.0 was to empower the popularization of educational resources through information technology, and to achieve the sharing of high-quality educational resources.

It is understood that Gao Jiao She has built the world's largest online open course platform Ai Course, the largest vocational education digital teaching resource sharing platform, and the online teaching service platform Smart Vocational Education and other educational resource platforms.

More than 100 million users of various teaching platforms of Gao Jiao She can directly connect to the classrooms of famous schools and teachers through a network cable, sharing high-quality resources.

Entering a new era, Liu Chao believes that talent cultivation is shifting from knowledge-based to ability-based, and people's practical needs for autonomous learning, personalized learning, and lifelong learning are becoming increasingly strong.

Natural language models can achieve dialogue-based learning and have unlimited potential and broad prospects in the field of education.

He firmly believes that educational equity 2.0 must be education empowered by artificial intelligence.Based on this concept, the Higher Education Press is building an AI foundation using large models such as the Alibaba Cloud Tongyi Qianwen.

The Higher Education Press has accumulated 4,500TB of high-quality educational materials over 70 years, including excellent textbook texts, rich multimedia resources, and exercise libraries.

Alibaba Cloud's Tongyi large model performance is among the forefront of domestic large models.

Previously, the Higher Education Press, Alibaba Cloud, and Zhejiang University have jointly released the "Zhihai Sanle" educational vertical large model trained based on the Tongyi Qianwen 7B (7 billion parameters) model, and applied it to the core courses of the 101 plan in 14 universities, "Introduction to Artificial Intelligence".

It is reported that the model can provide intelligent, personalized learning for related courses, serving talent cultivation.

In Liu Chao's view, the ideal educational large model needs to integrate four characteristics: more correct answers, distributed output, emotional provision, and resource recommendations.

Among them, the distributed output is called Socratic heuristic teaching.

Liu Chao said that Professor Wu Fei's team at Zhejiang University has made a lot of work in the development of Socratic teaching intelligence this year, allowing AI to break down a complex problem into several related simple problems, and ask them to students in a distributed manner, so as to inspire students to truly master the logic and reasoning process of learning.

Different students often have different learning progress even if they are in the same grade, and the mastery of knowledge points is also inconsistent.

The large model with memory and generalization ability can achieve "thousands of people, thousands of faces", promoting the realization of personalized learning, and making education truly "education for all".

The value that cannot be calculated is not only in the above fields.

Alibaba Cloud has always insisted on open source large models, serving as the foundation for the application development of thousands of industries, and an "AI application tropical rainforest" has grown on top of Tongyi.

As of mid-September 2024, the download volume of the Tongyi Qianwen open source model has exceeded 40 million, and the total number of Qwen series derivative models has exceeded 50,000, becoming the world-class model group second only to Llama.

At the 2024 Yunqi Conference, Alibaba Cloud CTO Zhou Jingren released the new generation of Tongyi Qianwen open source model Qwen2.5, and the flagship model Qwen2.5-72B performance surpasses Llama 405B, once again ascending to the throne of the world's open source large model.

Based on the vision of becoming the "most open cloud in the AI era", Alibaba Cloud hopes to empower sustainable social development with AI technology, as Liu Xiangwen, vice president of Alibaba Cloud and director of the China Sustainable Development Research Association, said, the combination of artificial intelligence and cloud computing, from the digital world into the physical world, will bring immeasurable value to the entire society.

Li Meng, chairman of the China Sustainable Development Research Association, believes that artificial intelligence will empower sustainable development and become a key support for promoting the green transformation of the economy and society.

As an emerging technological force, artificial intelligence will play an irreplaceable role in helping China optimize resources, improve efficiency, promote industrial transformation, environmental protection, eliminate poverty, educational equity, and accelerate the achievement of sustainable development goals.