With the continuous development of artificial intelligence, the scientific research and industrial sectors are demanding more advanced large language models for specific vertical domains.
On the afternoon of January 25th, Shanghai Jiao Tong University officially released Optics GPT, a vertical large language model for the field of optics, marking a significant breakthrough in the intersection of artificial intelligence and optical technology in my country.
On January 25th, the Optics GPT vertical large language model release conference was held at the Wenbo Building Conference Center on the Minhang Campus of Shanghai Jiao Tong University, officially launching a professional large language model deeply customized for the field of optics. Academician Fan Dianyuan of the Chinese Academy of Engineering, Academician Zhuang Songlin of the Chinese Academy of Engineering, Wu Qiang, Deputy Secretary of the Minhang District Committee and District Head of Shanghai, Pan Yan, Deputy Director of the Shanghai Municipal Commission of Economy and Information Technology, Li Zuxing, Second-level Inspector of the Shanghai Municipal Science and Technology Commission, Tan Ruicong, Deputy District Head of Minhang District, Zhang Xian, Deputy District Head of Minhang District, Pan Qing, former Deputy Director of the Interdisciplinary Department of the National Natural Science Foundation of China; Ding Kuiling, Academician of the Chinese Academy of Sciences and President of Shanghai Jiao Tong University, Guan Haibing, Member of the University Party Committee and Vice President, Lu Qi, Deputy Dean of the University's Science and Technology Development Institute and Director of the Academic Development and Achievements Department, experts and scholars from universities and research institutions, enterprise representatives, and heads of relevant colleges, functional departments, and project teams attended the conference. The meeting was chaired by Guan Haibing and Lu Qi.
In his speech, Ding Kuiling stated that artificial intelligence is profoundly disrupting the paradigm of knowledge production and reshaping the underlying logic of scientific research and engineering innovation. As a core source of talent cultivation and innovation, universities must proactively plan and prepare for this. He pointed out that the release of the optical large language model is an important practice for the university to address major national needs and promote the intersection and integration of artificial intelligence and cutting-edge disciplines, which is of great significance to promoting independent innovation and intelligent transformation of the optical industry.
In his speech, Pan Yan pointed out that the release of the optical large language model demonstrates the innovative vitality of the intersection and integration of artificial intelligence and optical engineering. In recent years, Shanghai has taken the lead in deploying scientific intelligence, accelerating the development of scientific intelligence and strengthening systematic promotion through a series of measures such as implementing the "Hundred Teams and Hundred Projects" program, building open communities, and cultivating an innovative ecosystem. In the next step, Shanghai will accelerate the creation of an open community for scientific intelligence and promote the deep integration of technological innovation and industrial innovation.
Professor Yi Lilin, project leader and professor at the School of Integrated Circuits (School of Information and Electronic Engineering) of Shanghai Jiao Tong University, delivered a keynote speech on Optics GPT, a vertical large language model for the field of optics, and demonstrated the large model system on-site.
At the press conference, Shanghai Jiao Tong University initiated the establishment of the "Optical Large Language Model Academic Ecosystem Alliance," with representatives from several domestic universities and research institutes participating in the launch ceremony. The academic ecosystem alliance aims to promote the collaborative development of basic theories, key algorithms, and data resources related to optical large language models, build an open and shared scientific research cooperation network, and promote high-level talent training and academic innovation.
The "Optical Large Language Model Industrial Ecosystem Alliance" was also launched simultaneously, with several leading domestic industry units joining the cooperation. The industrial alliance will focus on key areas such as intelligent operation and maintenance of optical communication networks, intelligent measurement instruments, laser processing equipment, and optoelectronic system design, promoting the deep integration of large language model technology with industrial needs and exploring replicable and scalable application models.
At the meeting, Shanghai Jiao Tong University launched a strategic cooperation with Shenzhen Wanliyan Technology Co., Ltd. and Jiaqiang (Shanghai) Intelligent Technology Co., Ltd. to jointly promote the continuous research and development and industrialization of intelligent measurement equipment, as well as the large-scale application of optical large language models in intelligent manufacturing scenarios.
To further integrate the model toolchain and hardware infrastructure, and to provide open services to universities, research institutions, and enterprises, creating a comprehensive innovation platform integrating R&D services, design services, processing services, and cutting-edge research in the fields of intelligent optical communication and computing, intelligent laser design and processing, Shanghai Jiao Tong University and Minhang District jointly launched the "Artificial Intelligence and Optics Cross-Integration Platform" project. Relevant leaders from the Minhang District government and representatives from the university participated in the launch ceremony, providing policy and resource support for the platform's construction.
The official release of Optics GPT (Optical Large Language Model) marks a significant breakthrough for my country in the field of artificial intelligence and optical technology integration. This model is like a "virtual optical expert," capable of deeply understanding optical principles and providing intelligent support for scientific research, design, and teaching.
Currently, while general artificial intelligence models are powerful, they struggle to deeply understand hard science and technology fields such as optics, which require profound professional knowledge and precise calculations. The "Optical Future" project team at Shanghai Jiao Tong University, after dedicated research and development, has successfully created this "optics-native" professional model. It doesn't simply modify a general-purpose model, but rather "grows" from specialized optical data, systematically learning core knowledge and design logic in fields such as optical communication and optical design.
In simple terms, if general-purpose large models like ChatGPT are "knowledgeable generalists," then Optics GPT is a "seasoned specialist." It focuses its efforts on deeply learning all the knowledge in a specific field, thereby answering more professional questions and solving more specific problems within that field, with greater reliability. This is like training a doctoral student in optics to become an AI expert in that field.
To objectively evaluate the actual level of Optics GPT's professional capabilities in optics, the team built a professional evaluation set covering six major areas: optical physics, quantum optics, optical design, nonlinear optics, optical computing, and optical communication. They then conducted systematic comparative tests of Optics GPT against several mainstream general-purpose large models and open-source large models. The evaluation results show that Optics GPT achieved leading results in all core dimensions, fully validating its professional depth and engineering cognitive abilities in the optical vertical field. This marks the validation of a brand-new technical path: through specialized and structured training, smaller models can also surpass giant general-purpose models in vertical fields.
As a completely independently developed domestic model, the optical large model possesses four distinct characteristics:
Lightweight Deployment: The model has a parameter size of 8B, supporting efficient deployment on edge devices, significantly lowering the application threshold for the optical industry.
High Cognitive Ability: Through systematic and structured injection of knowledge in the field of optics, it forms profound "optical literacy" and accurate physical intuition.
Strong Applications: It demonstrates comprehensive leading performance in core scenarios such as algorithm generation, system diagnosis, simulation design, and experimental assistance.
Full Controllability: The entire process from data construction and model training to deployment and operation is independently controllable, ensuring industrial security and data privacy.
Application Prospects: Empowering Education, Research, and Industry
In terms of education and teaching: As a new generation of intelligent teaching tools, the model can transform abstract optical theories and complex formulas into intuitive visual demonstrations and interactive Q&A, and can automatically generate rich teaching cases and virtual experiments, thereby revolutionizing traditional teaching methods and significantly improving teaching efficiency and learning experience.
In terms of scientific research: In basic research and cutting-edge exploration, this model can serve as an all-weather intelligent research assistant, helping researchers quickly review literature, inspire innovative ideas, complete complex simulation calculations, and assist in designing experimental schemes, thereby accelerating the research process from theory to verification.
In terms of industrial design, it deeply empowers key links in the optical industry chain:
In the field of domestic high-end instruments, it improves the user experience of instruments and significantly increases the intelligence level of high-end instruments.
In the field of computing infrastructure, it provides decision support for intelligent fault diagnosis and operation and maintenance of data center optical interconnection systems, and performs rapid and accurate factory calibration of optical modules.
In the field of laser manufacturing, it promotes the upgrading of core components such as lasers towards intelligent directions such as autonomous parameter optimization, intelligent state prediction, and automatic fault operation and maintenance.
Through its application in the above scenarios, the large-scale optical model will powerfully promote the comprehensive intelligent upgrading of my country's optical industry in design and development, production and manufacturing, and operation and maintenance services.
Looking to the future, Shanghai Jiao Tong University will always be guided by major national needs, focusing on building an "AI for Science" innovation paradigm, and leveraging the innovative power of universities to drive industrial breakthroughs in technological bottlenecks and achieve intelligent upgrading, continuously providing "Jiao Tong University solutions" for the construction of a strong scientific and technological nation and the development of new quality productive forces.
