"It is the essential feature of a large model that it can provide generalization capabilities on a model and solve the diverse needs of a series of scenarios and applications, so as to solve the problem of balance between costs and benefits."
On July 4, at the main forum of industrial development of the World Artificial Intelligence Conference, Zhang Peng, CEO of Zhipu AI, said that the current AI boom due to large models is different from before, in the past, AI technology has solved some practical problems, but today's development of large models has brought more important human-like cognitive capabilities.
Zhang Peng said that in the past, AI was not versatile enough and the cost was too high. However, the large model brings a new opportunity, which can provide generalization capabilities on a model, which is also the main direction of empowering the real economy with a new generation of large model technology - turning the original structure with a large base investment but small returns into an inverted pyramid structure to truly amplify its value.
01 GLM-New Generation Pedestal Large Model Technology Frontier and Industrial Application Forum was held
On July 5th, at the GLM-New Generation Pedestal Large Model Technology Frontier and Industrial Application Forum, hosted by the Knowledge Engineering Laboratory of the Department of Computer Science of Tsinghua University, undertaken by AI TIME, and co-organized by Donghao Lansheng (Group) Co., Ltd. and Zhipu AI, the guests focused on the GLM-4 model, shared the latest research results and theoretical breakthroughs of the GLM-4 model, and explored the technical frontier, industrial ecology and application of GLM-4.
论坛上,智谱AI CodeGeeX技术负责人郑勤锴发布了第4代CodeGeeX代码大模型CodeGeeX4-ALL-9B。
As an open-source version of the latest generation of CodeGeeX4 series models, CodeGeeX4-ALL-9B continues to iterate on the basis of GLM-4's powerful language capabilities and greatly enhances code generation capabilities. Using the single model of CodeGeeX4-ALL-9B, it can support comprehensive functions such as code completion and generation, code interpreter, network search, tool call, warehouse-level long code Q&A and generation, covering various scenarios of programming development. The performance of multiple authoritative code capability evaluation sets is the strongest model with less than 10 billion parameters, and even more than several times the scale of the general model, which achieves the best balance between inference performance and model effect.
At present, the number of individual users of CodeGeeX has exceeded 1 million, and CodeGeeX is completely free for individual users, and can be downloaded and used for free in various mainstream IDEs.
In addition to the release of the 4th generation CodeGeeX, at the forum, Huang Minlie, tenured professor of the Department of Computer Science and Technology of Tsinghua University, Zhang Jing, associate professor of the Department of Computer Science, School of Information, Chinese Renmin University, Yang Yang, associate professor of the School of Computer Science and Technology of Zhejiang University, Dai Guohao, tenured associate professor of the School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University, and Tu Cunchao, founder and CEO of Power Law Intelligence, discussed in depth the impact of GLM large models on the industry and industrial development.
Zhang Peng shared a number of innovative cases of GLM-4 in applications, especially breakthroughs in intelligent content generation, industry automation, and user personalized customization services. Demonstrates the value of GLM-4 in complex business environments.
"In the past few years, the business practice of Zhipu has accumulated a lot of experience for us, and I dare not say that it is the best practice, but it is a better practice." Zhang Peng said in his speech. GLM-4's rapid advancement in capabilities such as agent and tool invocation makes it possible to implement native AI architectures within the enterprise.
02 The GLM pedestal model was unveiled at WAIC 2024 with application results
The WAIC 2024 Zhipu AI booth showcased a series of innovative achievements with the Zhipu Large Model Open Platform bigmodel.cn and the Zhipu Large Model Product Matrix as the core.
As the treasure of this year's WAIC Pavilion, the Zhipu Large Model Open Platform bigmodel.cn is the best way to experience the Zhipu GLM series large models. The newly upgraded bigmodel.cn has been connected to the latest GLM large model family bucket, and new functions such as one-click fine-tuning and All Tools API calls have also been launched.
Whether you are a technology geek, a professional engineer, or a company looking for large-scale model capabilities, you can find products and services that suit you on the platform. At present bigmodel.cn there are more than 400,000 enterprise customers and developers, with an average daily call volume of 60 billion tokens, and the daily consumption of APIs has increased by more than 90 times in the past four months.
As a concentrated display of bigmodel.cn application achievements, "Zhipu Town" appeared in booth C. Based on the capabilities of the Zhipu large model, the town brings together typical cases of public affairs, consumption, cultural tourism, medical care, insurance, education, automobile, finance, industry and other industry scenarios, and provides users and enterprises with a variety of intelligent services such as intelligent driving, intelligent investment advisory assistant and financial report Copilot, provident fund consulting assistant, intelligent consultation, and travel intelligent body.
Zhipu AI also created a digital human live broadcast platform for CGTN, and AI Lao Luo also appeared in the Qingyan exhibition area in the form of a digital human, and the audience could ask AI Lao Luo about the cheats of bringing goods. In addition, the first issue of the AI creative work "Qingyan Picture Book" was also unveiled at the scene, which is an AI picture album with the theme of "Qing", showing the powerful drawing ability of Qingyan App. The audience can watch the picture scroll through the Qingyan Album applet, and can draw the same style with one click.
Creating products that can fully expand the imagination of users, and turning imagination into real productivity based on large-scale model technology, is the answer to the must-answer question of "from imagination to productivity".