ForecastingWhen quantum meets AI: NUS brings together international experts to turn frontier technologies into real-world solutionsNational University of Singapore — News · yesterdayDigital twinЦифровые двойники, единая интеграционная платформа и роботы. Правительство утвердило программу развития угольной промышленности РоссииTAdviser · 20hDigital twinAI-Moderated Interviews for Digital Twin CalibrationarXiv API — полнотекстовый поиск "digital twin" · 23hDigital twinHuman-AI Collaboration for Multi-Line Task Adjustment Using Local Large Language Models and a Digital TwinarXiv API — полнотекстовый поиск "digital twin" · yesterdayModellingAgent-Based Modeling of Systems of SystemsarXiv — Multiagent Systems (cs.MA) · yesterdayModellingHybrid Coevolutionary Opinion Game with Language AgentsarXiv — Multiagent Systems (cs.MA) · yesterdayDigital twinLLM and GraphRAG-Driven DSM Generation for Cyber-Physical SystemsarXiv — Systems and Control (eess.SY) · yesterdayModellingDeterministic Construction of Typical Networks in Network ModelsarXiv — Physics and Society (physics.soc-ph) · yesterdayForecastingGray-Box MPC for Articulated Dump TrucksarXiv — Systems and Control (eess.SY) · yesterdayModellingHigher-Order Approximation of Exit Functionals in Sampling-Based Stochastic Model Predictive ControlarXiv — Systems and Control (eess.SY) · 2dDigital twinPhysics-Informed Neural Networks for Climate-Aware Digital TwinsarXiv API — полнотекстовый поиск "digital twin" · 2dModellingРоссийские компании могут оценить готовность собственной инфраструктуры к агентной разработкеCNews · yesterday
Hybrid Coevolutionary Opinion Game with Language Agents
US & West ·
The Hybrid Coevolutionary Opinion Game H-COG combines cost-minimizing Friedkin-Johnsen agents with Phi-4 language agents in a dynamically rewired network. Using 50 agents seeded with opinions from 5,199 Reddit comments, it studies convergence and collective efficiency.
Why it matters
It shows how agent-based models with language components capture opinion dynamics in large networks, relevant to scenario analysis of social systems.
A pre-registered study with 317 participants compared AI-moderated, human-moderated and static interviews. AI moderation matched human depth, covered more themes and, at fixed budget, surfaced more customer needs; digital twins built from interview data were validated against held-out responses to marketing stimuli.
A generic formalism is proposed to represent and control large-scale systems of systems through agent-based simulation. Organizational aspects use the Agent-Group-Role model, functional aspects a goal specification, and multilevel aspects the IRM4MLS meta-model, covering both static and dynamic properties.
Conditions are identified to define the most typical network state in grand canonical ensembles and their random mixtures. A deterministic construction is proposed that converges to this typical state in the thermodynamic limit, addressing how well a real network matches a chosen model.