Domestic foundationLarge ModelStartupsStep StarRecently officially launched, founded by Jiang Daxin, former global vice president of Microsoft. The company focuses on multimodal fusion and scaling law, has developed a ToC product with hundreds of billions of parameters, and has made significant progress in the research and development of trillion models. Step Star's products include two applications for the C-end market:Jump to askandBubbling Duck, are now fully open for use.
Yuewen product entrance: https://www.1ai.net/6958.html
Yuewen is a chat application that is positioned as a personal efficiency assistant. It can perform basic question-and-answer, online information retrieval, file processing, and multimodal understanding. In the test, Yuewen showed excellent performance. It can not only comprehensively compare the differences between ChatGPT and Claude, but also accurately retrieveup to dateTechnological progress, such as major events at Musk’s brain-computer interface company.
In addition, Yuewen can handle complex logical reasoning tasks, such as converting images to tables and calculating net growth rates, and excels in multimodal understanding. Currently, Yuewen is free to use, without worrying about usage restrictions or model overload.
Bubbling Duck Product Entrance: https://www.1ai.net/6956.html
Bubble Duck is an AI open world platform that provides a rich set of intelligent dialogues and interactive games. Users can have interesting conversations with these intelligent agents, explore plot-based interactive games, and even create new dialogues themselves. Bubble Duck has both an app version and a web version, and the official website provides detailed getting started documentation to ensure that users can easily get started.
The "soul" behind Step Star is two 100 billion parameter large models, namely the Step-1 100 billion parameter language large model and the Step-1V 100 billion parameter multimodal large model. The Step-1V large model performs outstandingly in multimodal understanding, and can accurately describe and understand the information in images and videos, and perform tasks such as content creation and logical reasoning. In the "Sinan" multimodal model evaluation list, Step-1V ranksFirst, the performance is comparable to GPT-4V.
Jiang Daxin is full of confidence in the development of large models. He believes that the GPT-3.5 model with hundreds of billions of parameters is a watershed, and the training of trillion-parameter models requires huge computing power, systems, data and algorithm support. Step Star has rich experience and resources in this regard, and team members have deep backgrounds and practical experience in the field of AI.
The team of Jieyuexingchen consists of more than 150 people, including Jiang Daxin, Jiao Binxing and Zhu Yibo, who have rich research and practical experience in the field of AI. Jiang Daxin once said that he did not want to just be messed up in the wind of the times, but to actively participate in the development of AI technology.
Recently, Yuewen has launched a new feature called "One Picture to Understand". Users can upload documents, and Yuewen will generate a long picture and text with the key points of the analysis with one click. This is a feature that ChatBot did not provide before, providing users with a brand new experience.