On July 6th, in WAIC 2024, the World Artificial Intelligence Conference and AI Global Governance High-level Meeting, Step Star released three new products of the Step series: the official version of Step-2 trillion parameter language model, Step-1.5V multimodal model, and Step-1X image generation model.

Since its release in March this year, the Step series of universal large models has made comprehensive progress from hundreds of billions of parameters to trillions of parameters, from language models to multi-modal models, from understanding to generation.

At the same time, Step Star also focuses on exhibiting self-developed large model application products for C-end users, and discloses the latest developments and plans in the field of large model ecological cooperation.

During the conference, Li Jing, Vice President of Step Star, was interviewed by reporters from Kechuangban Daily in Caixin's live broadcast room.

He introduced that the latest released Step-2 trillion parameter language model officially adopts the innovative MoE mixed expert model architecture. The newly upgraded Step-1.5V hundred billion parameter multimodal model has excellent video understanding ability. It can accurately identify objects, characters and environments in videos, understand the overall atmosphere and emotions of the characters in the video, and even conduct descriptions and interactive Questions and answers.

In addition, Step Star also released the Step-1X image generation model, and collaborated with Shanghai Film to demonstrate the video generation capability for Chinese animation IP.

Step Star is also actively developing large-model applications for C-end users. The intelligent assistant, 'Jumping Question' and the AI open world platform, 'Bubble Duck', are two self-developed products launched by Step Star for C-end users. Li Jing introduced that 'Jumping Question' is positioned as an efficiency tool that fully utilizes the characteristics of the large model of Step Star's multimodal, accurately describing and understanding information such as text, data, and charts in images, and completing tasks such as content creation, logical reasoning, and data analysis. 'For example, in a photo taken at a scenic spot, you can ask Jumping Question to get information about that spot.'

'Bubble Duck' has created an AI open world. Here, users can explore stories, create characters, and immerse themselves in their own open world. 'We hope to help creators create interactive content consumption scenes with users through the unlimited expansion and personalized generation capabilities of AI. Currently, Bubble Duck is iterating and optimizing with immersion and user interaction experience as the core.'

The large-model of artificial intelligence has gone through a year and a half of 'technical frenzy'. Now, the price competition in large-models has brought about an acceleration of commercial competition. Domestic large model manufacturers such as ByteDance, Alibaba, Baidu, and Tencent have been competing for price reduction in the past two months.

In response to this, Li Jing said that he has predicted the fierce competition in the market early on, and in the future, Step Star will persist in exploring C-end and also pay attention to high-value vertical fields in the B-end.

On July 5th, Caiyue Step Star, Step Star, and GTJA Securities held a strategic cooperation signing ceremony. The three parties will work hard in aspects such as data, computing power, models, and applications to enhance the intelligent application capabilities of the securities industry and build the most financial savvy large model. Caiyue Step Star was jointly established by interface Caixin and the universal large model company Step Star, focusing on the technological research and development and scene application exploration of large models in the financial field.

'This cooperation is based on Step Star's universal large model base and combined with the trillion-level financial language materials accumulated by interface Caixin, on which we cooperate deeply with GTJA for the promotion of large models in intelligent investment advisory scenarios,' Li Jing said.

In addition, Step Star and Shanghai Film jointly explore innovation in the AI+IP field. The two parties launched a 'Measure Which Star You Are in Journey to the West' AI interactive experience, which called the Step series large model and integrated multiple capabilities such as image understanding, style transfer, image generation, and plot creation. Step Star also demonstrated the video generation capability for the IP of 'Calabash Brothers' in combination with Shanghai Film.

'This year is the 60th anniversary of Journey to the West. As a childhood classic IP that has accompanied several generations to grow up, how can we have better interactions with young people? Therefore, we will integrate AI text generation, AI plot generation, AI personality tests, and other tools to activate the classic IP to the fullest extent through interesting AI interactions, and use technology to pay tribute to the classics and tell good Chinese stories. This is our vision,' Li Jing emphasized.