Training tool-use agents typically relies on outcome-based filtering: Supervised Fine-Tuning (SFT) on successful trajectories and Reinforcement Learning (RL) on pass-rate-selected tasks. However, this paradigm ignores interaction dynamics—successful trajectories may lack error recovery or exhibit redundancy, while pass rates fail to distinguish structurally informative tasks from trivial ones. To bridge this gap, we propose TopoCurate, an interaction-aware framework that shifts the paradigm from linear outcome filtering to topological interaction modeling. By projecting multi-trial rollouts into a unified semantic quotient topology within the same task, we construct a structured manifold that explicitly captures the divergence between success and failure pathways across tool invocations and environmental responses. Leveraging this representation, we introduce a dual-selection mechanism: for SFT, we prioritize trajectories demonstrating reflective recovery, semantic efficiency, and strategic diversity to mitigate covariate shift and mode collapse; for RL, we select tasks with high error branch ratios and strategic heterogeneity, maximizing gradient Signal-to-Noise Ratio to address vanishing signals in sparse-reward settings. Evaluations on BFCLv3 and Tau2 Bench show that TopoCurate achieves consistent gains of 4.2% (SFT) and 6.9% (RL) over state-of-the-art baselines.

Multimodal large language models (MLLMs) have shown strong capabilities in perception, reasoning, and action generation. However, their ability to sustain exploration in dynamic open worlds remains unclear. Existing embodied and game-based benchmarks often compress interaction into short-horizon tasks or entangle success with domain-specific game mechanics. In this paper, we introduce MineExplorer benchmark for evaluating open-world exploration capabilities of MLLM agents in Minecraft. We first filter atomic tasks whose solutions rely heavily on Minecraft-specific knowledge to better reflect general open-world reasoning. Then we organize the benchmark around a ReAct-style capability formulation and compose atomic tasks into implicit multi-hop tasks. To further construct reliable instances, MineExplorer uses a multi-agent synthesis workflow that jointly designs task graphs, sandbox scenes, and rule-based milestone evaluators. Human evaluation shows that the multi-agent synthesis workflow produces significantly more reliable instances than a single-agent baseline. Experiments with advanced MLLM agents show that open-world exploration remains challenging, as strong models can handle many single-hop tasks but degrade sharply when hidden prerequisites must be coordinated over longer trajectories. Further analysis finds that task difficulty tracks agent completion, and larger models or thinking modes do not consistently translate into better performance. Code is available at anonymity.

Reinforcement learning with verifiable rewards (RLVR) has become a standard paradigm for post-training large language models (LLMs). While Group Relative Policy Optimization (GRPO) is widely adopted, it suffers from sparse reward signals and loses gradients entirely when all responses within a group receive identical rewards. On-policy distillation (OPD) offers a natural remedy by providing dense, token-level supervision from a teacher model. However, naively combining GRPO with OPD leads to degraded performance, due to three underlying causes: not all samples benefit from distillation; fitting too quickly to the teacher undermines the exploratory capacity of RL; and OPD’s advantages are asymmetric, suppressing most tokens. To address these challenges, we propose RSTG (Recovering Learning Signals via Adaptive Teacher Guidance), which applies distillation selectively and precisely where it matters most. At the sample level, OPD is restricted to negative zero-variance prompts with each sample weighted by the teacher’s confidence score. At the token level, distillation targets only tokens with high student entropy or large teacher-student divergence. We further augment training with SFT on correct trajectories generated by the teacher model, injecting positive gradient signals where RL yields none. Experiments demonstrate that RSTG substantially outperforms naive GRPO+OPD by +4.02% on math and +3.05% on code.

Existing KG-augmented models for commonsense question answering primarily focus on designing elaborate Graph Neural Networks (GNNs) to model knowledge graphs (KGs). However, they ignore (i) the effectively fusing and reasoning over question context representations and the KG representations, and (ii) automatically selecting relevant nodes from the noisy KGs during reasoning. In this paper, we propose a novel model, JointLK, which solves the above limitations through the joint reasoning of LM and GNN and the dynamic KGs pruning mechanism. Specifically, JointLK performs joint reasoning between LM and GNN through a novel dense bidirectional attention module, in which each question token attends on KG nodes and each KG node attends on question tokens, and the two modal representations fuse and update mutually by multi-step interactions. Then, the dynamic pruning module uses the attention weights generated by joint reasoning to prune irrelevant KG nodes recursively. We evaluate JointLK on the CommonsenseQA and OpenBookQA datasets, and demonstrate its improvements to the existing LM and LM+KG models, as well as its capability to perform interpretable reasoning.

Without training on labeled task data, unsupervised commonsense question answering seems challenging since it requires commonsense knowledge beyond the context of questions. Previous methods typically retrieved from traditional knowledge bases or used pre-trained language models (PrLMs) to generate fixed types of knowledge, which have poor generalization ability.In this paper, we aim to address the above limitation by leveraging the implicit knowledge stored in PrLMs and propose a two-stage prompt-based unsupervised commonsense question answering framework (TSGP). We first use knowledge generation prompts to generate the knowledge required for questions with unlimited types and possible candidate answers independent of specified choices. Then, we further utilize answer generation prompts to generate possible candidate answers independent of specified choices. Experimental results and analysis on three different commonsense reasoning tasks, CommonsenseQA, OpenBookQA, and SocialIQA, demonstrate that TSGP significantly improves the reasoning ability of language models in unsupervised settings.