SenseTime won the first place in the SuperCLUE September evaluation list, attracting attention in the direction of AI intelligence

Recently, SuperCLUE released the overall ranking list and each category task list for September. SenseChat 3.0 by SenseTime ranked first in the overall list of Chinese large models. In the newly added AI Agent sub-list, SenseChat 3.0 also ranked first, surpassing all domestic Chinese large models as well as GPT-3.5 and Claude 2, and second only to the performance of GPT-4. This demonstrates SenseTime’s advantages in innovative development and productivity release in the field of large models, as well as its accumulation and potential in exploring AGI

Note: Foreign representative models (GPT4.0/Claude2/gpt-3.5) do not participate in the ranking.

SuperCLUE is a comprehensive evaluation benchmark for Chinese general large models. It aims to comprehensively evaluate the performance of large models in various capability dimensions. It is one of the most professional and representative Chinese large model evaluation benchmarks in China. . This evaluation selected the 20 most representative general language models at home and abroad

Ranked first in the overall discussion list, and the objective score exceeded GPT-3.5

The latest SuperCLUE overall ranking and each classification task ranking mainly focus on the four capability quadrants of large models, namely language understanding and generation, including language understanding and extraction, context pairing, generation and creation, and role playing; Professional skills and knowledge, including knowledge and encyclopedia, calculation, code, logic and reasoning; Agent intelligence, including tool use, task planning; Security, including system security, command attack, a total of 12 basic abilities

SenseChat 3.0 of SenseTime ranks first in the overall rankings with a total score of 62.75 points. In the OPT objective part, SenseChat 3.0’s score even exceeds GPT-3.5, demonstrating its ability to perform large-scale Chinese models. Strong comprehensive competitiveness

SenseTime discussed that SenseChat will be officially launched in April 2023. It is one of the earliest large language models based on hundreds of billions of parameters in China and is continuously iteratively updated. It relies on SenseCore, a large-scale AI device from SenseTime. Currently, there are about 30,000 online GPUs and the computing power scale has been increased to 6 ExaFLOPS, effectively supporting the training, upgrade iteration and services of large language models.

Accelerate the development of artificial intelligence agents and accelerate the move towards strong artificial intelligence

With the development of large models, "chat" is far from meeting people's requirements. The ability to accurately use tools has become the key to liberating the productivity of large models. SuperCLUE's new AI Agent sub-list is the industry's first AI Agent list. It focuses on evaluating the performance of AI Agents in two key capabilities: "tool use" and "task planning." Evaluations show that SenseTime considers that SenseChat 3.0 has the potential to serve as a human super assistant. It can complete tasks autonomously according to human needs, thereby fully releasing the productivity of large models, making it second only to GPT-4 on the AI ​​Agent list, and ahead of the rest. Participate in the evaluation of large models.

Currently, the world's leading AI agents are almost all driven by the leading large model GPT-4. With the help of powerful tool usage capabilities, they can decompose complex problems into achievable sub-tasks and human-like natural Language interaction and other abilities. SenseChat 3.0 is discussed as a leading Chinese large-scale model. It uses three common tools: code interpreter, API call and search to solve complex tasks, flexibly build AI agent applications, and support enterprise productivity innovation.

Currently, SenseTime has established close cooperative relationships with more than 500 customers in multiple vertical industries such as finance, mobile phones, medical care, automobiles, real estate, energy, media, and industrial manufacturing. As a powerful large-scale model, SenseChat continues to improve rapidly, providing SenseTime with the opportunity to develop more powerful artificial intelligence agents that can interact like humans, and to explore the road to artificial general intelligence. important foundation and support

The above is the detailed content of SenseTime won the first place in the SuperCLUE September evaluation list, attracting attention in the direction of AI intelligence. For more information, please follow other related articles on the PHP Chinese website!