According to the paper published on arXiv [1], the researchers have developed a hierarchical tree-based approach to create Configurable and Static Deep Research Agents (Static-DRA) that can overcome the limitations of static Retrieval Augmented Generation (RAG) pipelines in handling complex, multi-turn research tasks.
The paper also discusses research on evaluating the generalization capabilities of LLM-based agents in mixed-motive scenarios using a framework called Concordia [2]. This work aims to assess how LLM agents perform in social interactions with both human and artificial agents.
Additionally, the researchers propose a two-step paradigm called Context-Aware Hierarchical Learning to address critical vulnerabilities in instruction handling for Large Language Models (LLMs), particularly when exposed to adversarial scenarios [3]. This approach aims to enhance the safety and robustness of LLMs.
[1] arXiv:2512.03887v1 - A Hierarchical Tree-based approach for creating Configurable and Static Deep Research Agent (Static-DRA) [2] arXiv:2512.03318v1 - Evaluating Generalization Capabilities of LLM-Based Agents in Mixed-Motive Scenarios Using Concordia [3] arXiv:2512.03720v1 - Context-Aware Hierarchical Learning: A Two-Step Paradigm towards Safer LLMs