RESEARCH ARTICLE

Real-time critical transition discoveries with large language models

¹ School of Artificial Intelligence and Automation, Huazhong University of Science and Technology, Wuhan 430074, China
² State Key Laboratory of Intelligent Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan 430074, China
³ Department of Computer Science, Brunel University of London, Uxbridge UB8 3PH, UK
⁴ School of Artificial Intelligence and Automation, Wuhan University of Science and Technology, Wuhan 430081, China
⁵ Shanghai Artificial Intelligence Research Institute, Shanghai Jiao Tong University, Shanghai 200030, China
⁶ School of Mechanical Science and Engineering, Huazhong University of Science and Technology, Wuhan 430074, China

^* Corresponding author (email: This email address is being protected from spambots. You need JavaScript enabled to view it. )

Received: 28 September 2025
Revised: 8 December 2025
Accepted: 10 December 2025

Abstract

Real-time prediction of critical transitions in complex systems is critical for preventing catastrophic failures and prolonging system lifespans. Existing approaches have extensively focused on qualitative early-warning indicators, but fall short of delivering quantitative, real-time predictions that could guide timely interventions to adjust system states. Here, we present CT-eProber, a large language model-based framework for efficiently probing critical transitions that enables both quantitative and qualitative early warnings. CT-eProber is a general framework that processes prompt data derived from either time-series sensor signals or discrete features, and rapidly adapts to diverse application domains via low-rank adaptation. We demonstrate the effectiveness of the framework on four representative datasets spanning chemistry, finance and robot systems. Results reveal that CT-eProber consistently achieves high predictive accuracy in both real-time quantitative prediction and qualitative classification of critical transitions. Our findings highlight the feasibility of large language model-driven critical transition discovery, establishing a generalizable pathway for real-time prediction and risk prevention in diverse complex systems.