Uncertainty estimation for language models often treats reliability as a property of the answer distribution: high confidence, stable samples, or low semantic entropy are taken as signs that an answer can be trusted. Table reasoning exposes a fundamentally different failure mode. A model may consistently return the same answer while linking the wrong entity, selecting the wrong column, applying the wrong operation, or relying on incomplete or insufficient evidence. We introduce MIRAGE—Model Inference Reliability Assessment via Grounded Evidence, a structure-grounded reliability framework for selective table reasoning. MIRAGE estimates residual structural uncertainty from test-time traces, combining outcome consistency with grounding and computation signals, and produces calibrated answer-or-abstain decisions with lightweight reliability certificates. On a controlled four-family diagnostic benchmark, structural reliability signals substantially reduce selective risk compared with generic confidence, sampling, semantic-entropy, and hallucination-detection baselines.

Open-world planning poses a significant challenge for general artificial intelligence due to environmental complexity and task diversity, especially in long-term tasks and lifelong learning. Inspired by cognitive theories, we propose M2PA, an open-world multi-memory planning agent. M2PA innovates by combining Large Language Models (LLMs) with human-like multi-memory systems, aiming to fully leverage the strengths of both while mitigating their respective limitations. By integrating the expansive world knowledge and language processing capabilities of LLMs with the perception and experience accumulation abilities of the human memory system, M2PA exhibits situation awareness, and experience generalization capabilities, as well as the potential for lifelong learning. In experiments, M2PA significantly outperforms current state-of-the-art agents across 50 Minecraft tasks in zero-shot learning. In exploratory lifelong learning experiments, M2PA demonstrates its continuous learning ability, achieving a 38.33% success rate in the “ObtainDiamond” task. Our findings provide a novel paradigm for constructing more effective agents in open-world environments.

The challenge of developing agents capable of open-world planning remains fundamental to artificial general intelligence (AGI). While large language models (LLMs) have made progress with their vast world knowledge, their limitations in perception, memory, and reliable reasoning still hinder LLM-based agents from achieving human-level performance in long-term tasks. Drawing inspiration from human cognitive-metacognitive collaboration, we propose Metagent-P, integrating the world knowledge of LLMs, the symbolic reasoning capabilities of cognitive architectures, and the self-reflection characteristic of metacognition to construct a “planning-verification-execution-reflection” framework. Metagent-P improves experience utilization through multimodal memory integration. It uses a neural-symbolic hierarchical representation structure to ensure the plan’s reasoning correctness in advance. Finally, it actively adapts the agent to dynamic environments through monitoring, evaluation, and regulation mechanisms. Experimental results show Metagent-P significantly outperforms current state-of-the-art methods in Minecraft. In long-term tasks, Metagent-P reduces the average replanning counts by 34% and exceeds the average human success rate by 18.96%. Additionally, Metagent-P also demonstrates self-evolution through step-by-step open-world exploration.