2023 is the year of the concentrated outbreak of global artificial intelligence large models, and the generative artificial intelligence (AIGC) application led by ChatGPT has swept the world, opening up a new track for the global artificial intelligence industry.
Zhao Zhiyun, director of the China Institute of Science and Technology Information, said that various technical routes of China's large models are also making breakthroughs in parallel, especially in natural language understanding, machine vision, multi-modality, etc., and a number of influential large models in the industry have emerged. According to incomplete statistics, 79 large models with a scale of more than 1 billion parameters have been published in China.
Let's take stock of the artificial intelligence models and their derivatives that have been released in China.
MOSS
Released by the team of Professor Qiu Xipeng of the Natural Language Processing Laboratory of Fudan University, MOSS is the first conversational large-scale language model developed by a university team in China, which can perform a series of tasks such as dialogue generation, programming, and fact answering, opening up all technical paths for generative language models to understand human intentions and have dialogue capabilities.
As the earliest ChatGPT-like product released in China, the MOSS team said that it is still a very immature model, and there is still a long way to go from ChatGPT.
Wen Xin large model
The first domestic company to come forward to "fight" ChatGPT is Baidu. Baidu's Wenxin inherits its technical advantages accumulated over the years in the field of search engines and AI. Wenxin Yiyan's training adopts a different method from the traditional search engine, which is based on Baidu's self-developed deep learning framework PaddlePaddle, Wenxin large model technology research and development, by introducing multi-layer Transformer network to learn language knowledge. Compared with traditional rule- and model-based methods, Wen Xin Yiyan can better adapt to the complexity of natural language, and can be trained on a large-scale corpus, resulting in better performance.
Although the Wen Xin Yiyan conference was criticized for showing the product in the form of screen recording, and the early performance of the product was not satisfactory, after a period of public beta iteration, coupled with Baidu's unique data resource advantages in the field of Chinese search engines, the current Wen Xin Yiyan is close to the ability of ChatGPT 3.0 in terms of Chinese processing, and the future can be expected.
360 GPT Large Model
360 has launched two products based on a series of self-developed large models, namely the ChatBot product "360 Intelligent Brain" and the AI image generation tool "360 Hongtu". 360 Brain integrates the technical capabilities of 360CV large model, 360GPT large model, 360GLM large model, and 360 multimodal large model, and realizes in-depth applications in the fields of language understanding, image recognition, natural language processing, and question answering systems.
At present, the landing application of "360 Intelligent Brain" combined with 360 search, intelligent hardware and other advantageous scenarios has gradually matured, and open internal testing. Among them, "Tuchacha" was awarded the "Excellent Case of Generative Artificial Intelligence Technology and Application" by the Key Laboratory of the Ministry of Industry and Information Technology, and 360 Group was also invited to participate in the preparation of China's large model application standards.
Nissin New Model
SenseNova big model is a large model system developed by SenseTime, which includes functions such as text diagrams, perceptual model annotation, and model development in addition to the natural language processing model with hundreds of billions of parameters. Based on these different capabilities, SenseTime has launched Chinese language processing ChatBot product "Shang", Wenshengtu product "Second Painting", digital human video generation product "Ruying", 3D scene generation product "Qiongyu", and 3D model generation product "Gewu", respectively, to enter the field of AIGC in an all-round way.
Tongyi large model
Developed by Damo Academy, a scientific research institute owned by Alibaba, the Tongyi model uses deep learning techniques, recurrent neural networks (RNNs) and long-short-term memory networks (LSTMs), attention mechanisms, and transfer learning. Its training data contains a large number of language and text data, including but not limited to: text data in Chinese, English, Japanese, French, Spanish, German and other languages; text data on various topics such as literature, history, science, art, etc.; and a wide range of expertise and technical documentation. Alibaba Cloud CEO Daniel Zhang said that all Alibaba products will be connected to the "Tongyi Qianwen" model in the future and undergo a comprehensive transformation, including Tmall, Taobao, DingTalk, Tmall Genius, Xianyu, Hema and other products.
On June 1, Alibaba Cloud announced the official launch of the new AI product "Tongyi Listening Understanding", which focuses on audio and video content, becoming the first large-model application product in China to open public testing. Tongyi Listening Access is connected to the understanding and summarization capabilities of the Tongyi Qianwen model, which can help users efficiently complete the transcription, retrieval, abstraction and organization of audio and video content anytime, anywhere, such as automatically taking notes, organizing interviews, and extracting PPT with the large model.
Tiangong big model
"Tiangong" is jointly developed by Kunlun Wanwei and Singularity Zhiyuan, and is another generative AI product of Kunlun Wanwei after the AI painting product "Tiangong Qiaohua". "Tiangong" interacts with users in question and answer style through natural language, and AI generation capabilities can meet diversified needs such as copywriting, knowledge Q&A, code programming, logic deduction, and mathematical calculation.
iFLYTEK Spark cognitive large model
As an artificial intelligence listed company deeply engaged in the fields of intelligent speech, natural language processing and computer vision in China, the primary goal of the iFLYTEK Spark cognitive model launched by iFLYTEK is to benchmark ChatGPT and GPT-4. iFLYTEK Xinghuo focuses on multi-role, multi-style long text generation, conversational understanding of arbitrary tasks, knowledge questions and answers of massive information, logical reasoning based on thinking chain, mathematical ability, code understanding and generation ability to achieve technological breakthroughs. It is reported that iFLYTEK will further realize the release of multiple product-level applications such as the Spark APP on June 9.
In addition to the above-mentioned products in the form of chatbots, dozens of large models such as Huawei's "Pangu" large model, which focuses on industrial AI-empowered, and Tencent's "mixed element" large model, which focuses on cross-modal video retrieval capabilities, have also been released this year.
epilogue
Although domestic large-model products have sprung up, the problems faced by the industry are also more prominent. The "Research Report on Chinese Intelligent Large Model Map" released by the China Institute of Science and Technology Information pointed out that natural language processing is still the most active key area for the research and development of large models, followed by the multimodal field, and there are still few large models in the fields of computer vision and intelligent speech. From the perspective of the distribution of R&D subjects, different innovation subjects such as universities, scientific research institutions, and enterprises are actively participating in the research and development of large models, but the joint research and development between academia and industry is still insufficient.
In view of the current development of large models in China, Zhao Zhiyun suggested that in the future, the mainland needs to strengthen the coordination of resources and R&D forces, encourage the open source and opening of large models, strengthen international cooperation, actively participate in global AI governance, and accelerate the implementation of AI governance principles and ethical norms in the whole chain of large model research and development, so as to further promote the orderly development of large models.
Welcome to follow the "New Digital Tide" public account and learn about the digital economy with us!
News of the week | The Politburo meeting of the Central Committee pointed out that attention should be paid to the development of general artificial intelligence
ChatGPT has set off a wave of AI big models