Take stock and look forward to generative AI.
Editor's note. Recently, the National Language Resources Monitoring and Research Center released the "Top Ten New Words" in 2023, among which "generative artificial intelligence" ranks first and is highly eye-catching. As a high-profile entry throughout the year, generative artificial intelligence has benefited from the emergence of ChatGPT since the beginning of 2023. As a result, it has also led to the rapid development of AI algorithms, large models and big data, and many major technology companies at home and abroad have left the AIGC track. After a whole year of competition, AIGC has now presented a market boom in major language model products and applications, and was once considered to promote artificial intelligence from 10 to 2The key variables and important signs of 0 progress, its development prospects in the new year, are worth looking forward to.
Digital intelligence forward-looking. China Economic Times reporter Hu Pan.
2023 is destined to be a turbulent year for AI technology. This year, the interactive AI application "a hundred flowers bloomed" and became a topic of repeated discussion and reporting by the public and **. Among the "Top Ten New Words" in 2023 released by the National Language Resources Monitoring and Research Center a few days ago, "Generative Artificial Intelligence (AIGC)" topped the list, which on the one hand reveals that AIGC has brought a non-negligible impact in China throughout the year, and on the other hand, it also shows that the whole society is looking forward to the development trend of AIGC in the new year.
A hundred flowers bloom", AIGC has entered a new stage.
Throughout 2023, from the "100 Model War" competition to the "Hundred Flowers Blooming" of the application layer ecology, the AIGC industry will usher in an explosive period this year.
In the international field, AI models are showing their edge, and they are using technology to control the future. This year, OpenAI officially announced the multi-modal large model GPT-4, adding an image function (multi-modal), making it from a conversational demo to a super app that can process text, images, voice, and other information carriers on multiple lines. At the same time, a number of tech giants such as Google and Microsoft have launched their own AI models, intending to compete in market share.
In China, the "100 model war" is also being staged. Wenxin Yiyan, Alibaba's Tongyi Qianwen, 360's 360 Zhibrain, Tencent's Mixed Yuan, Huawei's Pangu ......This year, the domestic large model also showed a scene of hundreds of battles. According to the data, as of October, China had released 238 large models, a threefold increase from 79 in June.
It is worth mentioning that in August this year, the "Interim Measures for the Management of Generative Artificial Intelligence Services" issued by seven departments including the Cyberspace Administration of China was officially implemented, setting the entry threshold from the policy level, and the standard system covers multiple industries, covering the capabilities of the model industry, the engineering performance of the model, the model computing power network, the safety and reliability of the model, etc., and the importance of data, algorithms, models, and security factors is highlighted. This also means that the "100 model war" has returned to rationality, and the AIGC industry has entered the stage of integration.
In this process, how to draw on the principles of biology and psychology to make AI systems adapt more quickly to new tasks and environments through meta-learning and other methods?How to reduce the dependence on large amounts of annotated data so that AI systems can learn knowledge in an unsupervised or self-supervised mannerHow to enable AI systems to process and integrate information from different modalities (e.g., images, voice, text, etc.) to improve their understanding and application capabilitiesHow to reduce the size and compute resource requirements of AI models to be deployed and run on edge devices;How to prevent and respond to attacks on AI systems, such as adversarial samples, model theft, and data poisoning, etc., to ensure the security and reliability of AI systems, are all urgent problems in the development of AIGC technology. Shen Yang, a professor at the School of Journalism at Tsinghua University and director of the Metaverse Culture Laboratory, said bluntly in an interview with a reporter from the China Economic Times.
is also huge and small", the risk points of AI development still need to be paid attention to.
With the rapid update and iteration of AI large models, what trends will the AIGC industry show in the future, and what sparks will collide?
In this regard, Shenyang said that the future development of large language models may show two opposite trends, namely giant mode and micro mode. Among them, the gigantic mode is all-modal, massive parameters, that is, drawing wisdom from a broader field of knowledge, providing deeper insights into problems, emphasizing the infinity and diversity of knowledge and the role of AI technology in the pursuit of truth and understanding the world. The micro pattern is model compression and optimization, striving to achieve approximate performance with limited resources. That is, while maintaining high accuracy, the size and computational cost of the model are greatly reduced, the computational efficiency is optimized, and the model is efficient, which can be run on smaller devices.
In view of global language and cultural differences, data security, and compliance requirements, the localization of large language models is a rigid need. Localization improves the accuracy and usefulness of models in specific language, culture, and industry contexts to meet the needs of more users. At the same time, China will not have only one large model, and each enterprise and individual will have their own GPT, which will present a pattern of pluralism and symbiosis. Shenyang said.
It should also be noted that in the development process of AIGC, there are still some risk points worthy of our attention.
Shenyang said that AIGC's strong involvement in the global industrial chain may fully replace programmers, graphic design, and legal AI customer service, setting an upper limit for labor costs, and the demographic dividend of third world countries no longer exists. As a result, the industrial chain of the third world will suffer a huge impact, and the low-end industrial chain will no longer be transferred. AIGC with the support of large computing power may become a sharp blade to divide the global industrial chain of multinational companies.
At the same time, under the influence of the technological Matthew effect, a stable society can predictably form a useless class. The mediocre mental work of most people will lose their value.
In addition, the ethical risks of AI will also be magnified exponentially as AI technology becomes powerful, thus affecting more people.
In view of this, Shen Yang suggested that in the future, the development of AI also needs to introduce ethics and humanistic care, develop personalized AI that can reflect the diverse values and needs of users, design AI education that can improve people's information literacy and critical thinking skills, and establish laws and regulations that can prevent AI abuse and protect users' rights and interests, so as to avoid AI risks from multiple dimensions.
The copyright of this *** belongs to China Economic Times. If the content of this article is quoted or quoted, permission must be obtained and indicated**China Economic Times.