Calibration evaluates whether a model confidence aligns with its empirical accuracy. Existing studies often compare the calibration of different large language models using global calibration metrics such as Expected Calibration Error and Brier Score. We begin by showing, both theoretically and empirically, that such comparisons are confounded by differences in model accuracy. For fairer cross-model comparison, we then propose ACE, an accuracy-controlled evaluation framework with three complementary views: Instance-Aligned, Distribution-Aligned, and Candidate-Aligned calibration. Across multiple benchmarks, model families, and confidence elicitation methods, we use ACE to study two practically important comparison axes, small versus large models and thinking versus non-thinking models. We find that many previously reported calibration advantages under raw global metrics weaken substantially after accuracy control. We also find that ranking reversal is frequent: models favored by raw metrics often cease to be favored once accuracy is controlled. Our results show that raw global calibration metrics are not robust for cross-model comparison, and that fair calibration comparison requires accuracy-aware evaluation.

While confidence estimation is a promising direction for mitigating hallucinations in Large Language Models (LLMs), current research dominantly focuses on single-turn settings. The dynamics of model confidence in multi-turn conversations, where context accumulates and ambiguity is progressively resolved, remain largely unexplored. Reliable confidence estimation in multi-turn settings is critical for many downstream applications, such as autonomous agents and human-in-the-loop systems. This work presents the first systematic study of confidence estimation in multi-turn interactions, establishing a formal evaluation framework grounded in two key desiderata: per-turn calibration and monotonicity of confidence as more information becomes available. To facilitate this, we introduce novel metrics, including a length-normalized Expected Calibration Error (InfoECE), and a new "Hinter-Guesser" paradigm for generating controlled evaluation datasets. Our experiments reveal that widely-used confidence techniques struggle with calibration and monotonicity in multi-turn dialogues. We propose P(Sufficient), a logit-based probe that achieves comparatively better performance, although the task remains far from solved. Our work provides a foundational methodology for developing more reliable and trustworthy conversational agents.

Information retrieval (IR) systems play a critical role in navigating information overload across various applications. Existing IR benchmarks primarily focus on simple queries that are semantically analogous to single- and multi-hop relations, overlooking complex logical queries involving first-order logic operations such as conjunction (∧), disjunction (∨), and negation (¬). Thus, these benchmarks can not be used to sufficiently evaluate the performance of IR models on complex queries in real-world scenarios. To address this problem, we propose a novel method leveraging large language models (LLMs) to construct a new IR dataset ComLQ for Complex Logical Queries, which comprises 2,909 queries and 11,251 candidate passages. A key challenge in constructing the dataset lies in capturing the underlying logical structures within unstructured text. Therefore, by designing the subgraph-guided prompt with the subgraph indicator, an LLM (such as GPT-4o) is guided to generate queries with specific logical structures based on selected passages. All query-passage pairs in ComLQ are ensured structure conformity and evidence distribution through expert annotation. To better evaluate whether retrievers can handle queries with negation, we further propose a new evaluation metric, Log-Scaled Negation Consistency (LSNC@K). As a supplement to standard relevance-based metrics (such as nDCG and mAP), LSNC@K measures whether top-K retrieved passages violate negation conditions in queries. Our experimental results under zero-shot settings demonstrate existing retrieval models' limited performance on complex logical queries, especially on queries with negation, exposing their inferior capabilities of modeling exclusion. In summary, our ComLQ offers a comprehensive and fine-grained exploration, paving the way for future research on complex logical queries in IR.

Group Relative Policy Optimization (GRPO) has emerged as a promising critic-free reinforcement learning paradigm for reasoning tasks. However, standard GRPO employs a coarse-grained credit assignment mechanism that propagates group-level rewards uniformly to to every token in a sequence, neglecting the varying contribution of individual reasoning steps. We address this limitation by introducing Outcome-grounded Advantage Reshaping (OAR), a fine-grained credit assignment mechanism that redistributes advantages based on how much each token influences the model's final answer. We instantiate OAR via two complementary strategies: (1) OAR-P, which estimates outcome sensitivity through counterfactual token perturbations, serving as a high-fidelity attribution signal; (2) OAR-G, which uses an input-gradient sensitivity proxy to approximate the influence signal with a single backward pass. These importance signals are integrated with a conservative Bi-Level advantage reshaping scheme that suppresses low-impact tokens and boosts pivotal ones while preserving the overall advantage mass. Empirical results on extensive mathematical reasoning benchmarks demonstrate that while OAR-P sets the performance upper bound, OAR-G achieves comparable gains with negligible computational overhead, both significantly outperforming a strong GRPO baseline, pushing the boundaries of critic-free LLM reasoning.

Reinforcement Learning with Verifiable Rewards (RLVR) improves reasoning in large language models but treats all correct solutions equally, potentially reinforcing flawed traces that get correct answers by chance. We observe that better reasoning are better teachers: high-quality solutions serve as more effective demonstrations than low-quality ones. We term this teaching ability Demonstration Utility, and show that the policy model's own in-context learning ability provides an efficient way to measure it, yielding a quality signal termed Evidence Gain. To employ this signal during training, we introduce In-Context RLVR. By Bayesian analysis, we show that this objective implicitly reweights rewards by Evidence Gain, assigning higher weights to high-quality traces and lower weights to low-quality ones, without requiring costly computation or external evaluators. Experiments on mathematical benchmarks show improvements in both accuracy and reasoning quality over standard RLVR.

Large language models (LLMs) are increasingly deployed as autonomous agents for multi-turn decision-making tasks. However, current agents typically rely on fixed cognitive patterns: non-thinking models generate immediate responses, while thinking models engage in deep reasoning uniformly. This rigidity is inefficient for long-horizon tasks, where cognitive demands vary significantly from step to step, with some requiring strategic planning and others only routine execution. In this paper, we introduce CogRouter, a framework that trains agents to dynamically adapt cognitive depth at each step. Grounded in ACT-R theory, we design four hierarchical cognitive levels ranging from instinctive responses to strategic planning. Our two-stage training approach includes Cognition-aware Supervised Fine-tuning (CogSFT) to instill stable level-specific patterns, and Cognition-aware Policy Optimization (CoPO) for step-level credit assignment via confidence-aware advantage reweighting. The key insight is that appropriate cognitive depth should maximize the confidence of the resulting action. Experiments on ALFWorld and ScienceWorld demonstrate that CogRouter achieves state-of-the-art performance with superior efficiency.

Agent harnesses shape how language-model agents use instructions, tools, and runtime components, but adapting these harnesses requires costly verification. Existing propose-and-verify methods typically score every candidate on a fixed task set, wasting rollouts on unrelated behaviors and allowing aggregate scores to obscure specific regressions. We introduce HarnessLens, a budget-aware framework for automated harness evolution. HarnessLens jointly explores the task space and user-configurable components, derives candidate modifications from execution trajectories, and selectively verifies each candidate on behavior-relevant tasks using an attributable-evidence gate. Across three agent harnesses and four benchmarks, HarnessLens improves average held-out performance by 7.6-13.6

Proactive Recommender Systems (PRSs) aim to guide user preference shift toward target items by generating paths of intermediate recommendations. Reinforcement learning (RL) provides a principled framework for optimizing such sequential decision tasks, as path rewards can naturally capture both short-term acceptance and long-term guidance effectiveness. However, naively applying policy gradients to PRS results in deficient gradient estimation. We identify two deficiencies: (1) path-level rewards decompose into step-level rewards with positive mean, creating a length-dependent bias that causes gradients to favor path extension over meaningful exploration; (2) weighting each step by the entire path-level reward ignores the decomposition structure, leading to high gradient variance. To rectify these two deficiencies, we propose an effective RL framework ProRL with two novel mechanisms for proactive recommendation. First, Stepwise Reward Centering subtracts expected rewards to neutralize length-dependent bias, ensuring that path extension yields zero expected gradient signal. Second, Position-Specific Advantage Estimation leverages the reward decomposition structure to compute step-dependent baselines, reducing gradient variance. Together, these mechanisms yield policy gradients that precisely target path quality. Our experiments on three real-world datasets demonstrate that ProRL significantly outperforms state-of-the-art PRSs. Our code is available at https://github.com/hongruhou89/ProRL.

Quotation recommendation enriches writing by suggesting quotations that fit a given context, but prior systems largely focus on topical relevance and overlook what makes quotes memorable. Based on a user study, we find that preferred quotations are often unexpected yet rational, motivating the goal of selecting quotes that are contextually novel while semantically coherent. We propose NovelQR, which (1) uses a generative label agent to map quotations and contexts into multi-dimensional deep-meaning labels for label-enhanced retrieval, and (2) reranks candidates with a token-level novelty estimator that mitigates auto-regressive continuation bias. Experiments on bilingual datasets across diverse domains show that NovelQR is preferred by human judges and improves overall recommendation quality over strong baselines, while achieving competitive novelty estimation.

Retrieval-Augmented Generation (RAG) effectively grounds large language models (LLMs) in external knowledge but struggles with exploratory reasoning problems (ERPs) that are the complex queries involving high uncertainty and ambiguity. Resolving ERPs requires complex reasoning with unclear paths, tending to result in retrieval noise and error accumulation. Furthermore, the absence of an end-to-end planning mechanism makes it difficult to generate effective trajectories for ERPs. Motivated by database query planning, we introduce PlanRAG, an RAG framework that models ERPs of natural language as logical query trees (LQTs). However, translating ERPs into LQTs is non-trivial due to representation and optimization gaps between structured SQL and unstructured natural language, making it highly challenging to construct high-quality LQTs. To address these problems, we first decompose ERPs into atomic queries and then organize them into LQTs using dynamic programming guided by a cost model involving multiple complementary dimensions. Finally, we execute iterative aggregation, rewriting, retrieval, and generation over LQTs, processing nodes concurrently and propagating intermediate results upward, with further parallelization across multiple threads for efficiency. Our experimental results show that PlanRAG outperforms state-of-the-art iteration-based and graph-based RAG systems on our newly constructed dataset, WikiWeb-ERP, thereby providing a new formulation for optimizing natural language queries. Our source code and dataset are available at https://anonymous.4open.science/r/PlanRAG-main-B2C8/.

Open-ended reasoning and long-form generation tasks lack reliable automatic verification signals for reward-based policy optimization. Rubrics offer a promising alternative, but existing approaches treat them as given artifacts – either hand-crafted or prompt-generated – and often miss the task-specific, knowledge-intensive dimensions that matter most, distorting the reward signal. Our key observation is that rubric construction is itself a research problem: identifying what makes a response correct or insightful requires discovering and synthesizing external knowledge. We propose Deep Research as Rubric (DR-rubric), a two-stage framework for constructing such rubrics. Stage I elicits domain facts, structural constraints, and failure modes through iterative multi-turn agentic search; Stage II distills this evidence into atomic, independently verifiable constraints for GRPO-based policy optimization. Because the model under training can serve as its own rubric generator, DR-rubric-8B supports bootstrap rubric generation without frontier-model assistance. We evaluate on 6 benchmarks spanning agentic research and expert reasoning. Experiments show that DR-Rubric achieves strong competitive performance with only 1K – 3K training instances, where GPT-5-generated rubrics particularly benefit breadth coverage on agentic tasks, Gemini-generated rubrics yield the most balanced performance across agentic and expert reasoning tasks, and bootstrap rubrics exhibit a specialization-to-rebalancing evolution achieving the best overall performance at the third iteration. Results demonstrate that reframing rubric construction from static evaluation templates into an evidence-driven research process yields more scalable, fine-grained reward signals for open-ended tasks.

In-Context Learning (ICL) has emerged as a new paradigm in large language models (LLMs), enabling them to perform novel tasks by conditioning on a few examples embedded in the prompt. Yet, the highly nonlinear behavior of ICL for NLP tasks remains poorly understood. To shed light on its underlying mechanisms, this paper investigates whether LLMs can solve ordinary differential equations (ODEs) under the ICL setting. We formulate standard ODE problems and their solutions as sequential prompts and evaluate GPT-2 models on these tasks. Experiments on two types of ODEs show that GPT-2 can effectively learn a meta-ODE algorithm, with convergence behavior comparable to, or better than, the Euler method, and achieve exponential accuracy gains with increasing numbers of demonstrations. Moreover, the model generalizes to out-of-distribution (OOD) problems, demonstrating robust extrapolation capabilities. These empirical findings provide new insights into the mechanisms of ICL in NLP and its potential for solving nonlinear numerical problems.

In recent years, the research of employing knowledge graphs (KGs) in sequential recommendation (SR) has received a lot of attention, since the side information extracted from KGs, especially the information of the correlations between items, indeed helps the SR models achieve better performance. However, many previous KG-based SR models tend to introduce some noise information when learning item embeddings, or insufficiently fuse item-item correlations into their sequential modeling, thus limiting their performance improvements. In this paper, we propose a D istance-Aware K nowledge-based S equential R ecommendation model ( DAKSR ), which exploits the explicit item-item correlations from KGs to achieve enhanced SR. Specifically, as one critical component in our DAKSR, the distance score matrix (DSM) is first obtained to indicate the correlations between items, and then leveraged in the following three major modules of DAKSR. First, in the Item-Set Embedding layer (ISE) all item embeddings are learned based on DSM, in which the noise information is eliminated effectively. Meanwhile, the Knowledge-Infused Transformer (KIT) incorporates DSM into its attention mechanism to improve the feature extraction. Furthermore, the Knowledge Contrastive Learning module (KCL) also leverages the item-item correlations presented in DSM to generate two credible sequence views, which are used to refine sample representations through a contrastive learning strategy, and thus improve the model's robustness. Our extensive experiments on three SR benchmarks obviously demonstrate our DAKSR's superior performance over the state-of-the-art (SOTA) KG-based recommendation models. The implementation of our DAKSR is available at https://github.com/Easonsi/DAKSR for reproducing our experiment results conveniently.

Recent advances in large language models (LLMs) have enabled social simulation through multi-agent systems. Prior efforts focus on agent societies created from scratch, assigning agents with newly defined personas. However, simulating established fictional worlds and characters remain largely underexplored, despite its significant practical value. In this paper, we introduce BookWorld, a comprehensive system for constructing and simulating book-based multi-agent societies. BookWorld’s design covers comprehensive real-world intricacies, including diverse and dynamic characters, fictional worldviews, geographical constraints and changes, e.t.c. BookWorld enables diverse applications including story generation, interactive games and social simulation, offering novel ways to extend and explore beloved fictional works. Through extensive experiments, we demonstrate that BookWorld generates creative, high-quality stories while maintaining fidelity to the source books, surpassing previous methods with a win rate of 75.36%. The code and demo of this paper can be found at the project page: https://bookworld2025.github.io/.

Although large language models (LLMs) have shown great potential in recommender systems, the prohibitive computational costs for fine-tuning LLMs on entire datasets hinder their successful deployment in real-world scenarios. To develop affordable and effective LLM-based recommender systems, we focus on the task of coreset selection which identifies a small subset of fine-tuning data to optimize the test loss, thereby facilitating efficient LLMs' fine-tuning. Although there exist some intuitive solutions of subset selection, including distribution-based and importance-based approaches, they often lead to suboptimal performance due to the misalignment with downstream fine-tuning objectives or weak generalization ability caused by individual-level sample selection. To overcome these challenges, we propose GORACS, which is a novel Group-level Optimal tRAnsport-guided Coreset Selection framework for LLM-based recommender systems. GORACS is designed based on two key principles for coreset selection: 1) selecting the subsets that minimize the test loss to align with fine-tuning objectives, and 2) enhancing model generalization through group-level data selection. Corresponding to these two principles, GORACS has two key components: 1) a Proxy Optimization Objective (POO) leveraging optimal transport and gradient information to bound the intractable test loss, thus reducing computational costs by avoiding repeated LLM retraining, and 2) a two-stage Initialization-Then-Refinement Algorithm (ITRA) for efficient group-level selection. Our extensive experiments across diverse recommendation datasets and tasks validate that GORACS significantly reduces fine-tuning costs of LLMs while achieving superior performance over the state-of-the-art baselines and full data training.

Large Language Models (LLMs) are prone to hallucination, particularly in long-form generations. A promising direction to mitigate hallucination is to teach LLMs to express uncertainty explicitly when they lack sufficient knowledge. However, existing work lacks direct and fair evaluation of LLMs' ability to express uncertainty effectively in long-form generation. To address this gap, we first introduce UNCLE, a benchmark designed to evaluate uncertainty expression in both long- and short-form question answering (QA). UNCLE covers five domains and includes more than 1,000 entities, each with paired short- and long-form QA items. Our dataset is the first to directly link short- and long-form QA through aligned questions and gold-standard answers. Along with UNCLE, we propose a suite of new metrics to assess the models' capabilities to selectively express uncertainty. We then demonstrate that current models fail to convey uncertainty appropriately in long-form generation. We further explore both prompt-based and training-based methods to improve models' performance, with the training-based methods yielding greater gains. Further analysis of alignment gaps between short- and long-form uncertainty expression highlights promising directions for future research using UNCLE.

Large language models (LLMs) have recently transformed from text-based assistants to autonomous agents capable of planning, reasoning, and iteratively improving their actions. While numerical reward signals and verifiers can effectively rank candidate actions, they often provide limited contextual guidance. In contrast, natural language feedback better aligns with the generative capabilities of LLMs, providing richer and more actionable suggestions. However, parsing and implementing this feedback effectively can be challenging for LLM-based agents. In this work, we introduce Critique-Guided Improvement (CGI), a novel two-player framework, comprising an actor model that explores an environment and a critic model that generates detailed nature language feedback. By training the critic to produce fine-grained assessments and actionable revisions, and the actor to utilize these critiques, our approach promotes more robust exploration of alternative strategies while avoiding local optima. Experiments in three interactive environments show that CGI outperforms existing baselines by a substantial margin. Notably, even a small critic model surpasses GPT-4 in feedback quality. The resulting actor achieves state-of-the-art performance, demonstrating the power of explicit iterative guidance to enhance decision-making in LLM-based agents.

The integration of workflows with large language models (LLMs) enables LLM-based agents to execute predefined procedures, enhancing automation in real-world applications. Traditional rule-based methods tend to limit the inherent flexibility of LLMs, as their predefined execution paths restrict the models' action space, particularly when the unexpected, out-of-workflow (OOW) queries are encountered. Conversely, prompt-based methods allow LLMs to fully control the flow, which can lead to diminished enforcement of procedural compliance. To address these challenges, we introduce FlowAgent, a novel agent framework designed to maintain both compliance and flexibility. We propose the Procedure Description Language (PDL), which combines the adaptability of natural language with the precision of code to formulate workflows. Building on PDL, we develop a comprehensive framework that empowers LLMs to manage OOW queries effectively, while keeping the execution path under the supervision of a set of controllers. Additionally, we present a new evaluation methodology to rigorously assess an LLM agent's ability to handle OOW scenarios, going beyond routine flow compliance tested in existing benchmarks. Experiments on three datasets demonstrate that FlowAgent not only adheres to workflows but also effectively manages OOW queries, highlighting its dual strengths in compliance and flexibility. The code is available at https://github.com/Lightblues/FlowAgent.

This work proposes a novel framework based on nested evolving set processes to accelerate Personalized PageRank (PPR) computation. At each stage of the process, we employ a localized inexact proximal point iteration to solve a simplified linear system. We show that the time complexity of such localized methods is upper bounded by $\min\{\tilde{\mathcal{O}}(R^2/\epsilon^2), \tilde{\mathcal{O}}(m)\}$ to obtain an $\epsilon$-approximation of the PPR vector, where $m$ denotes the number of edges in the graph and $R$ is a constant defined via nested evolving set processes. Furthermore, the algorithms induced by our framework require solving only $\tilde{\mathcal{O}}(1/\sqrt{\alpha})$ such linear systems, where $\alpha$ is the damping factor. When $1/\epsilon^2\ll m$, this implies the existence of an algorithm that computes an $\epsilon$-approximation of the PPR vector with an overall time complexity of $\tilde{\mathcal{O}}(R^2 / (\sqrt{\alpha}\epsilon^2))$, independent of the underlying graph size. Our result resolves an open conjecture from existing literature. Experimental results on real-world graphs validate the efficiency of our methods, demonstrating significant convergence in the early stages.

Teaching scientific concepts is essential but challenging, and analogies help students connect new concepts to familiar ideas. Advancements in large language models (LLMs) enable generating analogies, yet their effectiveness in education remains underexplored. In this paper, we first conducted a two-stage study involving high school students and teachers to assess the effectiveness of LLM-generated analogies in biology and physics through a controlled in-class test and a classroom field study. Test results suggested that LLM-generated analogies could enhance student understanding particularly in biology, but require teachers' guidance to prevent over-reliance and overconfidence. Classroom experiments suggested that teachers could refine LLM-generated analogies to their satisfaction and inspire new analogies from generated ones, encouraged by positive classroom feedback and homework performance boosts. Based on findings, we developed and evaluated a practical system to help teachers generate and refine teaching analogies. We discussed future directions for developing and evaluating LLM-supported teaching and learning by analogy.