Multimodal large language models (MLLMs) generate responses autoregressively, integrating visual and linguistic information in an evolving context. Prior work on interpretability has focused on individual layers and circuits (where), leaving the token-level dynamics of multimodal…
This paper proposes a hybrid collaborative framework to optimize technology selection in Small and Medium-sized Enterprises (SMEs) by integrating machine learning (ML) predictions with Planning Poker, consensus-based estimation technique used in agile software development. Addres…