• wx
    分享到微信

    用微信“扫一扫”,点击右上角分享按钮,
    即可将网页分享给您的微信好友或朋友圈。

SenseTime, Shanghai AI Lab Debut New-Generation LLM

(Yicai) Jan. 18, 2024 -- SenseTime Group, one of China’s biggest artificial intelligence firms, and Shanghai Artificial Intelligence Laboratory have released InternLM2, a new-generation large language model they developed together with the Chinese University of Hong Kong and Fudan University.

 

The updated version of InternLM, which was released on June 7, can simultaneously accept and process up to 300,000 Chinese characters, Shanghai-based SenseTime revealed yesterday.

 

InternLM2 can correctly extract key information from materials, expanding its application scenarios, Lin Dahua, a leading scientist at Shanghai AI Lab, told Yicai. It can quickly 'read' and summarize financial reports of hundreds of pages and form abstracts or take out key information from meetings lasting several hours, Lin added.

 

InternLM2 has stronger numerical aptitude, a more powerful ability to invoke tools, and a notably improved conversational skill, Lin said. All the improvements can greatly help its application, and media, education, finance, and traditional manufacturing can all benefit from the progress of LLMs, Lin pointed out.

 

InternLM2 can be used for commercial purposes for free with Shanghai AI Lab’s authorization, the same as the previous version.

 

There is still a gap between LLMs and trusted artificial general intelligence that can match humans in all intellectual aspects, according to Qiao Yu, Shanghai AI Lab’s leading scientist and assistant to its director.

 

More data and computing power, known as the scale path, has been at the heart of the rise of AI technology, but this cannot solve the problem of illusion, trust, and controllability, Qiao said. In the future, scale growth will remain the main method of developing LLMs, but other technological routes should be fully utilized, Qiao added.

 

LLMs will develop faster this year than in 2023, and seeking basic technological improvement and industrial applications are the two main future themes, Lin noted. Institutions, businesses, and teams all had catching up with GPT-3.5 or GPT-4's performance are their goal last year, but their future development will be more diverse, Lin added.


Read Also
AWS to invest $15bn in cloud computing in Japan
Pony.ai’s Sienna-based Robotaxi approved for paid services in Guangzhou
JD.Com Staff Say Chinese E-Retailer’s IP Address Has Been Partly Blocked by Rival Pinduoduo

Research