Public articles linked to the same research event.
arXiv The work reframes LLM agent systems built as a loop of Planning, Memory, Tools, and Control Flow as a trajectory-level "agentic cognitive depth" profile across five criteria—context sensitivity, temporal continuity, multimodal coordination, adaptive interaction, and metacognitive monitoring—gives operational proxies and a perturbation procedure for each, links the profile to the agent's world model, proposes measuring it through controlled perturbation, deployment-time dynamic observation, and mechanistic interpretability, and identifies neurosymbolic design patterns such as symbolic verifiers, structured memory, planner coupling, and tool constraints for building and testing these capacities.
The work reframes LLM agent systems built as a loop of Planning, Memory, Tools, and Control Flow as a trajectory-level "agentic cognitive depth" profile across five criteria—context sensitivity, temporal continuity, multimodal coordination, adaptive interaction, and metacognitive monitoring—gives operational proxies and a perturbation procedure for each, links the profile to the agent's world model, proposes measuring it through controlled perturbation, deployment-time dynamic observation, and mechanistic interpretability, and identifies neurosymbolic design patterns such as symbolic verifiers, structured memory, planner coupling, and tool constraints for building and testing these capacities.
The work reframes LLM agent systems built as a loop of Planning, Memory, Tools, and Control Flow as a trajectory-level "agentic cognitive depth" profile across five criteria—context sensitivity, temporal continuity, multimodal coordination, adaptive interaction, and metacognitive monitoring—gives operational proxies and a perturbation procedure for each, links the profile to the agent's world model, proposes measuring it through controlled perturbation, deployment-time dynamic observation, and mechanistic interpretability, and identifies neurosymbolic design patterns such as symbolic verifiers, structured memory, planner coupling, and tool constraints for building and testing these capacities.
The work reframes LLM agent systems built as a loop of Planning, Memory, Tools, and Control Flow as a trajectory-level "agentic cognitive depth" profile across five criteria—context sensitivity, temporal continuity, multimodal coordination, adaptive interaction, and metacognitive monitoring—gives operational proxies and a perturbation procedure for each, links the profile to the agent's world model, proposes measuring it through controlled perturbation, deployment-time dynamic observation, and mechanistic interpretability, and identifies neurosymbolic design patterns such as symbolic verifiers, structured memory, planner coupling, and tool constraints for building and testing these capacities.