Researchers have developed a metacognition framework to improve Large Language Models' (LLMs) ability to assess their own uncertainty and knowledge limitations. The system computes a Metacognitive State Vector (MSV) across five dimensions derived from cognitive psychology, allowing it to switch between fast and deliberative processing modes based on query complexity. A proof-of-concept demo showcases the framework's feasibility in creating metacognitive self-awareness and self-regulation in LLM systems. --- Why it matters: This work matters because it addresses a significant limitation of current LLMs: their inability to assess their own uncertainty and knowledge limitations, which undermines reliability and trust. This framework could enable more robust and trustworthy AI systems by providing them with metacognitive capabilities. Source: https://arxiv.org/abs/2608.15400