NVIDIA AI Introduces SpatialClaw: A Training-Free Agent That Treats Code as the Action Interface for Spatial Reasoning⭐8

NVIDIA AI launches SpatialClaw, a training-free agent using code as an action interface for spatial reasoning tasks without environment-specific training.

A Definition of Good Explanations and the Challenges of Explaining LLM Outputs⭐9

This article defines a "good explanation" using counterfactuals and prior beliefs, then discusses key challenges in applying this definition to large language m...

ToolSense: A Diagnostic Framework for Auditing Parametric Tool Knowledge in LLMs⭐8

ToolSense diagnoses whether LLMs truly understand tools in agent systems, revealing a 50-64% performance gap between benchmark and real-world retrieval.

Position: Hippocampal Explicit Memory Is the Cornerstone for AGI⭐8

A position paper argues hippocampal explicit memory is essential for AGI, as LLMs rely on implicit memory and lack higher-order reasoning like planning.