On August 31, the large-scale model products of 8 enterprises/institutions, including Baidu, Byte, SenseTime, Zidong Taichu under the Chinese Academy of Sciences, Baichuan Intelligent, and Zhipu Huazhang, have passed the "Interim Measures for the Management of Generative Artificial Intelligence Services" for the first time . , can be officially launched to provide services to the public.
Specifically include:
-
Five Beijing companies: Baidu ( Wenxin Yiyan ), Douyin ( Skylark model ), Zhipu AI ( GLM model ), Chinese Academy of Sciences ( Zidong Taichu model ), Baichuan Intelligent ( Baichuan model )
-
Three Shanghai enterprises and organizations: SenseTime ( Daily New large-scale model ), MiniMax ( ABAB large-scale model ), Shanghai Artificial Intelligence Laboratory ( general-purpose large-scale model for scholars )
The following are the introductions and links of the 8 AI models (and products):
Baidu (Wen Xin Yi Yan)
Wenxin Yiyan (English name: ERNIE Bot) is Baidu's new generation of knowledge-enhanced large-scale language model, a new member of the Wenxin large-scale model family. It can interact with people, answer questions, assist in creation, and help people obtain information efficiently and conveniently. knowledge and inspiration.
Wenxin Yiyan is a large-scale language model for knowledge enhancement. Based on the Flying Paddle deep learning platform and Wenxin knowledge-enhanced large-scale model, it continuously integrates and learns from massive data and large-scale knowledge, and has the technical characteristics of knowledge enhancement, retrieval enhancement, and dialogue enhancement.
On March 20, 2023, Baidu's official WeChat news Wenxin Yiyan cloud service will be launched on March 27. In June, version 3.5 of Baidu Wenxin Large Model was available for internal testing. On August 31, Wenxin Yiyan was fully opened.
Wenxin Yiyan online experience: https://yiyan.baidu.com/
Douyin (Skylark model)
In mid-August, ByteDance’s Douyin, an AI robot “Doubao” developed based on the large model of the skylark, began a small-scale invitation test. Users can log in through mobile phone number, Douyin or Apple ID.
"Doubao" is an AI project previously code-named "Grace" in Byte, and currently has the functions of Wenshengwen and Wenshengtu. Enter the default page, and you can find some chat interfaces of intelligent agents, including the AI robot "Doubao", English teacher Elaine, all-round writing assistant, Xiaoning who loves chatting, etc. These agents can provide users with multilingual and multifunctional AIGC services, including but not limited to Q&A, intelligent creation, chat, etc. In addition, Doubao also supports voice playback capabilities, that is, the answers are not only listed in the form of text, but also broadcast in voice at the same time.
Users can also train their own intelligent body by themselves. The specific operation method is: choose to create a new dialogue, and give the intelligent body of the dialogue a new name, and then can ask questions to the intelligent body according to the user's interests. , this model is somewhat similar to the more popular "AI Companion", which is to create an artificial intelligence that can chat and talk.
Doubao online experience: https://www.doubao.com/chat/
Wisdom Spectrum AI (GLM Large Model)
On August 31, Zhipu Huazhang officially launched the first generative AI assistant - "Zhipu Qingyan", which is now available in major app stores. , vivo, Xiaomi, etc.) to download, or search for "Zhipu Qingyan" in the WeChat applet to experience its functions.
Based on ChatGLM2, a Chinese-English bilingual dialogue model independently developed by Zhipu AI, the assistant has been pre-trained with trillions of characters of text and code, and uses supervised fine-tuning technology to provide users with intelligent services in the form of general dialogue. "Zhipu Qingyan" as the user's intelligent assistant can answer various questions and complete various tasks for users in work, study and daily life.
At present, "Zhipu Qingyan" has rich capabilities such as "general question and answer, multi-round dialogue, creative writing, code generation, and virtual dialogue", and will open up multi-modal and other generation capabilities in the future.
Zhipu AI, established in 2019, is one of the first batch of companies in China to develop large-scale models, transformed by the technological achievements of the Knowledge Engineering Laboratory (KEG) of Tsinghua University.
Zhipu Qingyan online experience: https://chatglm.cn/
Baichuan Smart (Baichuan Large Model)
On August 31, Baichuan Intelligent, founded by former Sogou CEO Wang Xiaochuan, announced that it was the first to pass the filing of the "Interim Measures for the Management of Generative Artificial Intelligence Services" and open its services to the whole society. Users can log in to www.baichuan-ai.com to experience the brand-new experience brought by the Baichuan large model in the fields of knowledge question answering and text creation.
Only four months after its establishment, Baichuan Intelligent released three general-purpose large language models one after another. According to Baichuan Intelligent, as of today, the downloads of the two open source models of Baichuan-7B/13B on Huggingface have exceeded 3 million.
Baichuan-53B online experience: https://www.baichuan-ai.com/home
Sense Soup (Daily New Large Model)
SenseTime's large-scale model application "SenseChat" also announced on August 31 that it will fully open its services to users. Users can experience it through the URL
https://chat.sensetime.com . After the user's mobile phone is registered, he can have a conversation with "Consultation SenseChat".
On April 10, SenseTime launched the large-scale model "Ri Ri Xin", including the natural language processing model "Shang Shang", the Vincent graph model "Miao Hua" and the digital human video generation platform "Ru Ying". Among them, discussing SenseChat1.0 is also one of the earliest chat robot products in China based on a large language model with hundreds of billions of parameters.
Chat with SenseChat online experience https://chat.sensetime.com/
MiniMax (ABAB large model)
The ABAB large model is an advanced general-purpose large language model with powerful language processing capabilities and the ability to understand and generate natural language text.
MiniMax was founded in December 2021 by Yan Junjie, former vice president of SenseTime and head of general intelligence technology, known as All in AGI.
In March 2023, MiniMax launched an API open platform for enterprise users, supporting service calls for text and speech models. At the same time, MiniMax's other generative dialogue AI "Inspo" started internal testing. Its specific abilities include writing, obtaining information, providing suggestions, and remembering the previous text, and completing continuous tasks.
It is said that the MiniMax "abab" large model capability is suitable for most text processing-related scenarios, such as logical reasoning, text continuation, copy generation, text expansion, text rewriting, content summarization, code generation, knowledge retrieval, etc. In addition, the MiniMax open platform also provides large-scale sound models, which can be applied to scenarios such as dubbing and audio interaction. At present, the MiniMax open platform has access to more than 100 paying customers, providing more than ten industry scenarios such as office collaboration, interactive entertainment, customer service, search, and education.
MiniMax "abab" large model online experience: https://api.minimax.chat/
Chinese Academy of Sciences (Zidong Taichu Large Model)
Established in October 1956, the Institute of Automation of the Chinese Academy of Sciences is the earliest national automation research institution in China. "Zidong Taichu" is positioned as a cross-modal general artificial intelligence platform and will be officially released in 2021.
"Zidong Taichu" currently has a parameter of 100 billion, has multiple modes of vision, text, and voice, and is built on the basis of a full-stack localized basic software and hardware platform. In June this year, Zidong Taichu released version 2.0, adding new modal data such as video, sensor signals, and 3D point clouds. At present, Zidong Taichu has implemented applications in many industries such as text creation, medical care, logistics, and government affairs.
Zidong Taichu online experience: https://taichu-web.wair.ac.cn/
Shanghai Artificial Intelligence Laboratory (general model for scholars)
Shanghai Artificial Intelligence Laboratory (also known as Pujiang Laboratory) was established in 2020, initiated by the Shanghai Artificial Intelligence Innovation Center.
In 2021, Pujiang Lab, together with SenseTime, the Chinese University of Hong Kong, and Shanghai Jiaotong University, will release the general vision technology system "Scholar" (INTERN) . Up to now, the parameters of Shusheng's large model system have reached hundreds of billions, including three basic models of "Scholar Multimodal", "Scholar Puyu" and "Scholar Tianji", as well as open source models for the development and application of large models. system.
Scholar·Pu language online experience https://internlm.intern-ai.org.cn/