Although the big language model has achieved remarkable achievements, its sustainable development is worrying.Researchers at the Frontier Artificial Intelligence Research Center are exploring the sustainable next -generation large language model and reducing the plan for carbon footprints.In the next few years, practical applications will continue to expand, but people's concerns about privacy protection, personalized experience and sustainable development will also increase.We hope that the government's efforts in artificial intelligence strategies and research and development will help deal with these problems.
ChatGPT has only launching one year, which has significantly changed the way people work, create and communicate.It and other large language model tools play an important role in helping people understand what genetic artificial intelligence (General AI).Generating artificial intelligence is to generate or create new content by learning the mode and structure of input data.
ChatGPT's developer Openai said its user base continues to expand, with more than 100 million active users per week, which is the fastest -growing consumer application.People will definitely be interested in the huge potential of generating artificial intelligence.According to reports, Microsoft, technology giants, has invested in OPENAI for a total of $ 13 billion and joined the board of directors, while Google uses its own production artificial intelligence tools in products such as e -mail and documents.
Singapore is also seeking to develop and use effective artificial intelligence solutions, including generating artificial intelligence and large language model tools.The government announced the national artificial intelligence strategy 2.0 in early December. This is an updated version of the national artificial intelligence strategy in 2019. It has implemented an important artificial intelligence plan in various industries, governments and scientific research fields.The strategy aims to promote the development of artificial intelligence through community, infrastructure and environment.At the same time, the Information and Communication Media Development Bureau announced the cooperation with the National Artificial Intelligence Core (AISG) and the Singapore Science and Technology Research Bureau (A*Star) to spend 70 million yuan to launch a national multi -modal large language model plan to promote artificial intelligence in Singapore and the regionInnovation, enhance the ability of local artificial intelligence research and development, and cultivate artificial intelligence expertise.
Big language models are crucial, and can bring changes to all walks of life.Financial talents can use this learning algorithm for risk analysis and strengthen fraud testing.In the field of digital economy, large language models can also write code to assist programmers to develop software.
Although the large language model has huge potential and is likely to be widely used, before that, we have to deal with some problems.Specifically, we should enhance the training and data protection measures of large language models in professional fields.
Take into account the privacy protection and personalized experience of the big language model
Privacy protection is the main problem of using large language models.When the user enters the detailed information or ask the question (also known as the prompt), the large language model will collect confidential information.In addition, many large language model designers have guided and improved the ability of large language models through user prompts, especially in terms of generating results.This means that some large language models may be stored in the confidential information that the user entered in accidentally, and the database for training large language models, and this information may be revealed when responding to other user prompts.
In May this year, after Samsung Electronics found that some employees uploaded confidential code, they were prohibited from using tools such as ChatGPT.Data transmitted to such platforms will be stored on the external server, which is difficult to retrieve and delete, and may be leaked to other users.
In order to solve the problem of privacy protection, the Information and Communication Research Institute (I2R) under the new scientific research has been committed to exploring privacy protection technology and allowing large language models to handle confidential data safely in the environment of protecting privacy.In addition, the new scientific research set up a cutting -edge artificial intelligence research center last year to explore how to ensure user privacy while enhancing the personalized content of large language models.For example, researchers at the Research Center use a large language model with a special field prompt by fine -tuning, so as to respond to related issues, anonymous treatment of user interaction and prompts.
Fine -tune can improve the performance of large language models in other ways.When an enterprise interacts with customers with a general chat robot, the "sound" may seem unpopular or cannot reflect the corporate brand.Examples to use employees to communicate with customers to train and chat robots can help ensure accurate voice in place.The work of cutting -edge artificial intelligence research centers covers the personalized response and privacy protection of large language models.
Personalized large language models can promote education development
Big language models bring huge potential to personalized experience.Taking education applications as an example, when students studying aloud, teachers often need timely feedback from one -to -one, which is extremely difficult for large class teaching.To deal with this problem, teachers can use artificial intelligence to give students individual guidance.
Dr. Chen Fangyu and her information and communication research institute team developed an artificial intelligence counseling system to help students learn themselves.
Education Technology Corporation Commontown has deployed Malay and Tamilian versions in the school.The English and Chinese version derived the commercial product NOMOPAI, assisting students to prepare for oral test and improve the sales training of the enterprise.
When students read aloud, the artificial intelligence speech evaluation and counseling system will detect and score their pronunciation, tone and fluency.The system also provides personalized feedback to assist students to improve speech skills and improve the overall language level.Teachers can also use feedback from artificial intelligence systems to tailor -made teaching strategies and solutions for students.
In the challenge of artificial intelligence education hosted by the Ministry of Education and the Ministry of Education, the scientific researchers of the Institute of Information and Communication Design Multi -mode artificial intelligence mentors, guide children's mother tongue (Chinese, Malay, and Tamili)Communicate.It uses visual elements such as pictures to stimulate students who just go to elementary school.These design and research and development help maintain the diversification and bilingual advantages of Singapore.
Next Frontier Field: Long -term sustainable development of large language models
Although the big language model has made remarkable achievements, sustainable development is worrying.To keep large language models accurately through a large number of training parameters, it means that data centers and servers must consume more energy, thereby increasing carbon emissions.
Researchers at the Frontier Artificial Intelligence Research Center are exploring the sustainable next -generation large -language models, and through innovative technology, the development of carbon footprints in the training and deployment phase is developed.For example, reduce the data required by the training model, or use good performance and small energy consumption in the deployment stage to effectively adjust the existing model.
In the next few years, the technology of generating artificial intelligence and large language models will undoubtedly continue to develop.Practical applications will continue to expand, but people's concerns about privacy protection, personalized experience and sustainable development will also increase.We hope that the government's efforts in artificial intelligence strategies and R & D will help deal with these problems, so that generating artificial intelligence and large language models will continue to become a tool for changing our lives, work and creation.
The author is the director and professor of the New Scientific Research Frontier Artificial Intelligence Research Center
Focus Publishing Translation
The "Scientific Star language" column is launched by the United Morning Post and the Singapore Science and Technology Research Bureau in the last week of each month.
