
AI performance tasks are increasing, but human attention is the bottleneck... Pursuing management system automation
AI-generated summary
Deepseek announced 'Super Agent', a new architecture that mobilizes AI agents to perform multiple tasks simultaneously.
“The number of missions that AI can perform is increasing… Human attention is the bottleneck.”
(Shanghai = Yonhap News) Correspondent Cha Byeong-seop = Chinese artificial intelligence (AI) company Deepseek announced a new architecture called 'Super Agent' that mobilizes AI agents to perform multiple tasks simultaneously.
According to the Global Times and Yonhap News Agency on the 1st (local time), Deepseek disclosed this the previous day, saying that super agents are not limited to one chat window and can continuously perform multiple tasks simultaneously and proactively respond to environmental changes.
Deepseek pointed out that although the number of tasks that AI can perform is increasing, the attention span of humans in charge of management tasks has not increased accordingly, becoming a 'bottleneck' in productivity. The explanation is that one person can follow only 3 to 5 conversations at the same time.
The purpose of developing a super agent is to have the system take over conversation management and context synchronization instead of people, so that people are responsible for goal setting and inspection, and AI is responsible for coordination and execution.
A super agent consists of a 'coordinator' (router) and an 'executor' (worker) who act like a virtual team. The router receives information on the platform, such as chat and email, and distributes tasks, and workers perform individual tasks such as code review and feature development.
Accordingly, the user deals with a single AI agent, but in reality, multiple agents work simultaneously.
In addition, Deepseek explains that the super agent has initiative and can monitor conversations in a group chat room and proactively recognize problems raised without any instructions and suggest solutions.
However, DeepSeek added that this did not present a new theory and merely applied ideas from existing computing systems to the agent architecture.

Japan's GMO Internet Group identified a North Korean-linked hacker organization as the mastermind behind the leak of its customer information and reported it to the authorities. The hacker demanded a ransom of $1 million, and GMO offered a reward of the same amount to any informant who could solve the case.

Samsung Electronics Executive Chair Lee Jae-yong visited the company's advanced semiconductor packaging research facility in Yokohama, Japan, on Sunday to bolster AI chip competitiveness and supply chain cooperation.

UNIST Professor Seung-yeol Han's team developed 'MARBO', a learning method that helps artificial intelligence (AI) infer the opponent's hidden intentions and role and respond to deceptive situations. In the mafia game experiment, the winning rate improved significantly.

The Ministry of Public Administration and Security and the Personal Information Protection Committee held the 3rd 'Everyone's Debate' to discuss ways to protect and utilize personal information in the AI era. At the discussion, various opinions were raised about maintaining a prior consent-centered system and introducing a refusal method after prior notice.

HD Hyundai Robotics has partnered with Palantir Technologies to develop physical AI solutions, aiming to transform workplace operations and industrial site oversight.

Apple has decided to integrate Alibaba's AI model 'Q1' into its AI system 'Apple Intelligence' in China. It will be tested in the developer beta version starting on the 4th of this month and officially released next month.