Redirecting to original paper in 30 seconds...
Click below to go immediately or wait for automatic redirect
📄 Abstract
Abstract: We envision a new era of AI, termed agentic organization, where agents solve
complex problems by working collaboratively and concurrently, enabling outcomes
beyond individual intelligence. To realize this vision, we introduce
asynchronous thinking (AsyncThink) as a new paradigm of reasoning with large
language models, which organizes the internal thinking process into
concurrently executable structures. Specifically, we propose a thinking
protocol where an organizer dynamically assigns sub-queries to workers, merges
intermediate knowledge, and produces coherent solutions. More importantly, the
thinking structure in this protocol can be further optimized through
reinforcement learning. Experiments demonstrate that AsyncThink achieves 28%
lower inference latency compared to parallel thinking while improving accuracy
on mathematical reasoning. Moreover, AsyncThink generalizes its learned
asynchronous thinking capabilities, effectively tackling unseen tasks without
additional training.
Authors (7)
Zewen Chi
Li Dong
Qingxiu Dong
Yaru Hao
Xun Wu
Shaohan Huang
+1 more
Submitted
October 30, 2025
Key Contributions
Introduces 'asynchronous thinking' (AsyncThink) as a paradigm for LLMs to solve complex problems collaboratively. AsyncThink organizes internal thinking into concurrently executable structures, with an organizer assigning sub-queries to workers and merging results. This approach achieves 28% lower inference latency and improves accuracy on mathematical reasoning tasks, generalizing to unseen tasks.
Business Value
Accelerates complex problem-solving by AI, making advanced AI applications more responsive and efficient. This can lead to faster innovation cycles and improved user experiences in AI-powered services.