The large model booth was crowded with people Gao Chang Photo
Zhongxin Wanghai, July 8 (Li Qiuying, Gao Chang) The 2023 World Artificial Intelligence Conference recently kicked off in Shanghai. The theme of this conference is "Intelligent Connected World Generates the Future", "Big Model" is one of the highlights of this Artificial Intelligence Conference, and the reporter walked into the hall and experienced a variety of "explosive" large models on the spot.
In Huawei's exhibition area, the Pangu model attracted many visitors. In front of the booth "Pangu Basic Model – Drawing a New World", an exhibitor was interacting with the narrator, "I chose the keywords '3D rendering' 'tree'. In less than 5 seconds, three unique paintings were presented on the screen in front of them, and participants chose the most satisfactory one to generate personalized works to achieve "instant" generation of everything. In addition to this booth, the Pangu models unveiled at the scene also include "Pangu Manufacturing Model", "Pangu Mine Model", "Pangu Meteorological Model" and "Pangu Drug Molecule Model".
Pangu drug molecule large model Gao Chang photo
Huawei's "Pre-trained Large Model White Paper" pointed out that pre-trained large models came into being to solve the fragmentation dilemma of AI algorithm implementation. "It collects a large amount of images, text, and other data, uses unsupervised or self-supervised learning methods to extract the knowledge contained in the data, and stores it in a neural network model with a large number of parameters." When faced with a specific task, simply invoking a common process can unleash this knowledge and combine it with industry experience to solve real problems.
The Natural Language Processing Laboratory of Fudan University developed a large-scale language model MOSS Gao Changxing
Zhou Tangjun, a docent at the "Pangu Mine Big Model" booth, mentioned that the big model is not only the question and answer language model such as AIGC and ChatGPT that the public understands, in essence, the logic of the large model can be applied in various fields, and the generalization ability is strong, which can cover more scenarios and industries than the small model. In the application scenario of the "Pangu Mine Model", artificial intelligence replaces people to engage in dangerous work and repetitive labor, covering 1000+ business scenarios such as mining, mining, machinery, transportation, and communication, building a full-scenario ecology of intelligent mines, and promoting the double improvement of safety and efficiency through small-sample tuning.
At the exhibition site, the booth of language intelligence technology company Midu is also one of the "explosive models". In the "Poems for You" experience area, the reporter entered the word "fireworks" with the assistance of the staff, and the AI immediately automatically generated a modern poem with beautiful wording and neat workmanship, and printed it as a "collection". At the same time, a number of AIGC immersive interactive experiences such as "AI Postcard" and "AI Tour China" are also being carried out. It is understood that the generative experience of this poetic work is based on the Honey Manuscript Tongtong empowered by the large language model. According to the staff, Midu Manuscript can not only "compose poetry", but also has the ability to write specific genres including announcements, press releases, etc., which speeds up and improves the efficiency of office in all walks of life.
Related display of large model applications Gao Chang Photo
It is reported that on the 6th, Midu released the first knowledge Q&A and content generation large-language model that supports the localized software and hardware operating environment - honeynest. Liu Yidong, chief technology officer of Midu, said, "The general big model should solve the problem of allowing people who do not have the capacity building of the general model to enjoy the convenience of the general model. "Honeynest is built on hundreds of billions of high-quality Chinese multimodal data training, which not only enables customized content generation of "thousands of texts and thousands of faces", but also provides users with more efficient, safe, and stable guarantees for daily use based on Huawei's Ascend AI ecosystem.
In addition, focusing on large models, the World Artificial Intelligence Conference set up a theme exhibition area of "Towards General Artificial Intelligence" for the first time, focusing on displaying more than 30 large models at home and abroad. Mobvoi launched the large-language model "Sequence Monkey" with multimodal generation capabilities, empowering finance, law, medical care, media and other fields through model, image, voice customization and other services; 360 self-developed cognitive general large model "360 Intelligent Brain", reshaping a new paradigm of human-machine collaboration; The Natural Language Processing Laboratory of Fudan University developed a large-scale language model MOSS, which opened up all technical paths for generative language models to understand human intentions and have dialogue capabilities for the first time in China. SenseTime launches the "SenseNova" big model... Through forward-looking vision and high-standard display of large models with Chinese characteristics and Shanghai representatives, the conference fully demonstrates the disruptive impact of large models on the industry.
Visitors visit the booth related to the large model Gao Chang Photo
In front of the Pangu Large Model booth, an exhibitor working in the market field asked the specific application of the industry equipped with the large model, "I think it is very attractive to use AI to generate posters, but I will also consider whether there is a commercial copyright issue?" The docent smiled and replied, "There is really no standard at present, and it will continue to be improved in the future." "The large model has brought a change in thinking to the landing scheme of artificial intelligence, and the pre-trained large model has taken a small step forward in the direction of artificial intelligence algorithm automation, at the same time, the large model empowers diversified specific scenario applications, and also provides a boost for the improvement of the industry and related regulations, and the industry large model will continue to help scene innovation in the "spiral". (End)