Researchers are pursuing mathematical proofs to ensure AI safety, moving beyond empirical testing. Mike Winer uses statistical physics to predict neural network behavior, while a new study argues perfect AI control is mathematically impossible, suggesting diverse AI ecosystems as a safer alternative.
A team led by the Chinese Academy of Sciences has confirmed the existence of 'fairy circles' in northern China's arid zones, marking the first such record in the country's drylands. Published in PNAS, the study uses phase transition theory to explain how these ring-shaped vegetation patterns form through resource competition, offering a new framework for predicting ecosystem collapse.
Apurba Biswas and Thomas Guérin of the University of Bordeaux have developed the first fully analytical theory explaining how long-term memory in stochastic processes creates correlations between rare events. Published in Physical Review Letters, their work challenges the standard Arrhenius-Kramers paradigm, which assumes rare events are independent and follow an exponential waiting-time distribution. By modeling systems with power-law autocorrelations, the researchers demonstrate that memory effects cause rare events to cluster in time, deviating from standard statistical predictions. This framework provides a parameter-free method to predict event correlations using only equilibrium measurements, with potential applications in seismology, climate science, and single-molecule biology.