Think like a scientist: physics-guided LLM Agent for equation discovery
Jianke Yang, Ohm Venkatachalam, Mohammad Kianezhad, Sharvaree Vadgama, Rose Yu
This paper presents a framework for a physics-guided LLM agent designed to automate the discovery of symbolic equations from observed scientific phenomena. The agent mimics scientific reasoning, integrating domain knowledge and physical constraints to guide the generation of hypotheses. This approach advances the field of symbolic regression, demonstrating that LLMs can act as structured scientific thinkers and produce equations that are both human-interpretable and aligned with established physical principles.
Artificial intelligence is creating a new global linguistic hierarchy
Giulia Occhini, Kumiko Tanaka-Ishii, Anna Barford, Refael Tikochinski, Songbo Hu et al.
This paper presents a longitudinal analysis of the social, economic and infrastructural conditions of 6,003 languages, in order to evaluate the systemic inequalities in access to AI language technologies. Despite community efforts to broaden linguistic coverage, it finds that the dominance of a small number of languages is intensifying, leaving the vast majority of the world's linguistic communities in a state of persistent digital marginalisation.
Digital ecosystems: enabling collaboration in a fragmented world
Marc Schmitt
This paper introduces a spectrum framework for polycentric digital ecosystems, conceptualising them as nested socio-technical systems that span personal, organisational, inter-organisational and global layers. Extending platform theory, the framework redefines digital ecosystems as distributed adaptive networks, offering concrete pathways for cross-border coordination and innovation in an increasingly fragmented world.
Roberto Balestri
This study examines demographic bias in two leading commercial image generators by creating 3,200 photorealistic images using semantically neutral prompts, then measuring the results using rigorous colourimetric and facial analysis pipelines. The study provides a large-scale, comparative audit methodology that distinguishes aesthetic rendering from underlying representational bias. It highlights the persistent fairness failures of state-of-the-art generative AI systems.
Ali Raza & Fareeha Hanif
This paper provides a chronological review of YOLO object detection models, focusing on their application in remote sensing and geospatial environments. It analyses architectural innovations, evaluation benchmarks, hardware deployment considerations and post-processing strategies across all major YOLO versions. The review outlines the main challenges and future research directions for real-time geospatial object detection.