Wuhan Institute of Artificial Intelligence in conjunction with the Institute of Automation of the Chinese Academy of Sciences independently developed "Zidong Taichu (idiom); refers to the Korean Three Kingdoms period (475-221 BC)"The big model has been iterated to version 2.0, and according to the latest official news, "Zidong Taichu 3.0" will be released in the first half of this year.
According to Hubei Daily, Wang Jinqiao, director of the Wuhan Institute of Artificial Intelligence, introduced:"In the first half of the year, we will release 'Zidong Taichu' 3.0, which will further enhance the ability of big models to empower thousands of industries. Leveraging the relevant research, the institute's cooperation with Dongfeng Motor and Guangting Information in the field of intelligent driving has been in full swing."
Zidong Taichu First Generation AI Big ModelReleased in 2021 and jointly developed by the Institute of Automation of the Chinese Academy of Sciences (IAAS) and Huawei, it is claimed to be "the world's first 100 billion-parameter multimodal large model".
Unlike most of the current text-based language models, the "Zidong Taichu" model was developed with multimodal technology as the core, using more types of data such as graphic-audio-text for cross-modal unified characterization and learning, and the "Zidong Taichu" model has been developed to be a multimodal model.Realized "unified representation" and "mutual generation" among image, text and speech data..
Released in June 2023, Zidong Taichu 2.0 is based on Huawei's full-stack localized hardware and software platforms, Rise AI and Rise MindSpore, and compared to the first generation, it focuses on improving decision-making and judgment capabilities, based on text, images, and audio."Zidong Taichu 2.0 can incorporate more modal data such as 3D, video, and sensor signals., and optimize features such as fused cognition of speech, video, and text, and common sense computing.
According to the report, upgraded to version 3.0, the big model will have the ability to independently select and use tools based on various performance optimizations to meet the needs of deeper logical interactions.
Zidong Taichu will significantly shorten and optimize the training process for intelligent driving.. Through the big language model and multimodal capabilities, the big model can replace the massive manual labor, complete the automatic labeling of data such as cars, signal lights, road signs, etc., and dramatically improve the efficiency of intelligent cars to perceive the world.
At the same time, "Zidong Taichu", when connected to the car, is able to learn and understand driving goals and intentions at a deeper level, quickly translate environmental information into a language that can be read and understood by the device, predict the situation in the next few seconds, and give appropriate instructions to the car in advance.