Apply for Trial
Newsroom

SenseTime Recognized as Industry Leader in AGI Infrastructure, Showcases Latest Progress in Large Models

2023-06-05

June 5, 2023, Shanghai - SenseTime participated in the 2023 Lingang New Area Intelligent Computing Conference in Shanghai last Friday, where it was recognized as the leader of Lingang’s intelligent computing industry. SenseTime also showcased its latest progress in pushing the boundaries of artificial general intelligence (AGI) with AI large models powered by its SenseCore AI infrastructure and SenseNova foundation model sets.

 

Themed “Lead the Future”, the conference brought together experts and industry representatives to discuss development opportunities and promote collaboration in the Lingang New Area with the aim to foster a vibrant AGI ecosystem. The goal was further supported by the establishment of the “Lingang New Area Intelligent Computing Alliance” at the conference.


 The “Lingang New Area Intelligent Computing Alliance” was established at the 2023 Lingang New Area Intelligent Computing Conference in Shanghai.jpg

The “Lingang New Area Intelligent Computing Alliance” was established at the 2023 Lingang New Area Intelligent Computing Conference in Shanghai

 

Dr. Xu Li, Chairman and CEO of SenseTime, shared SenseTime’s latest progress in the field of large models during his keynote speech empowered by SenseCore. SenseTime's open-source multimodal AI model "INTERN 2.5" which was unveiled earlier this year, demonstrated superior performance across three major visual tasks and over 20 authoritative task datasets. SenseCore also played a crucial role in the development of "FengWu", the global medium-range weather forecasting AI large model launched in April. "FengWu" achieved superior performance in forecasting core atmospheric variables with a high degree of accuracy beyond 10 days, allowing for accurate prediction of typhoon tracks and other extreme weather conditions.

 

For autonomous driving, SenseTime introduced the industry's first end-to-end autonomous driving solution, “UniAD,” which integrates perception and decision-making. "UniAD" surpassed the state-of-the-art (SOTA) methods in various key datasets and metrics, achieving a 30% improvement in lane line prediction accuracy, nearly 40% reduction in motion displacement prediction errors, and almost 30% reduction in planning errors.

 

Additionally, SenseTime recently co-released the versatile AI intelligent agent GITM, enabling it to simulate human-like behavior in the highly realistic virtual world of "Minecraft."  “SenseChat”, the latest large-scale language model (LLM) developed by SenseTime, achieved significant improvements in various aspects within just over a month since its release in early April. It demonstrated enhanced capabilities in multilingual skills, table generation and processing, historical knowledge and induction, multi-turn dialogue, and long-text comprehension. SenseChat also secured a leading position in the industry across multiple authoritative test datasets and evaluation rankings. Furthermore, SenseTime's AI Code Assistant demonstrated an impressive 619% improvement in inference efficiency, enabling it to support six times the number of users with the same computational resources thereby, significantly reducing usage costs.

 

Dr. Xu Li, Chairman and CEO of SenseTime, shared SenseTime’s latest progress in the field of large models.jpg

Dr. Xu Li, Chairman and CEO of SenseTime, shared SenseTime’s latest progress in the field of large models

 

"SenseTime is dedicated to developing premium AGI infrastructure, as it constantly improves the iteration speed and capabilities of large models to unlock the vast potential of AGI, “said Dr. Xu Li. “We are thrilled to deepen collaborations with industry partners, explore new opportunities and future development directions."

 

As of May this year, SenseCore has served over 40 core clients including more than 10 clients for large models, covering cutting-edge fields including intelligent driving, biopharmaceuticals, smart business and academia. SenseTime’s large-scale models have been successfully deployed in more than 20 application scenarios, showcasing their versatility and effectiveness. Leveraging SenseCore, SenseTime has established a compressive MaaS (Model-as-a-Service) solution, providing customers with high-efficiency, low-cost and scalable AI infrastructure and services. Through close collaboration with both upstream and downstream partners, SenseTime aims to build a prosperous, open, and diversified AGI industry ecosystem, embracing the AI 2.0 era.