Computation and Language
HiNS: Hierarchical Negative Sampling for More Comprehensive Memory Retrieval Embedding Model
Motong Tian, Allen P. Wong, Mingjun Mao et al. (+1 more)
Memory-augmented
language agents rely on
embedding models for effective
memory retrieval. However, existing training data construction overlooks a critical limitation: the hierarchical difficulty of negative samples and their natural distribution in human-
agent interactions. In practice, some negatives are semantically close distractors while others are trivially irrelevant, and natural dialogue exhibits structured proportions of these types. Current approaches using synthetic or uniformly sampled negatives fail to reflect this diversity, limiting
embedding models' ability to learn nuanced discrimination essential for robust
memory retrieval. In this work, we propose a principled data construction framework HiNS that explicitly models negative sample difficulty tiers and incorporates empirically grounded negative ratios derived from conversational data, enabling the training of
embedding models with substantially improved
retrieval fidelity and generalization in
memory-intensive tasks. Experiments show significant improvements: on LoCoMo, F1/BLEU-1 gains of 3.27%/3.30%(MemoryOS) and 1.95%/1.78% (Mem0); on PERSONAMEM, total score improvements of 1.19% (MemoryOS) and 2.55% (Mem0).
Read the full abstract ▾
All authors Motong Tian, Allen P. Wong, Mingjun Mao, Wangchunshu Zhou
Computation and Language
Comparative Study of Large Language Models on Chinese Film Script Continuation: An Empirical Analysis Based on GPT-5.2 and Qwen-Max
Yuxuan Cao, Zida Yang and Ye Wang
As large
language models (LLMs) are increasingly applied to creative writing, their performance on culturally specific narrative tasks warrants systematic investigation. This study constructs the first Chinese film script continuation
benchmark comprising 53 classic films, and designs a multi-dimensional evaluation framework comparing GPT-5.2 and Qwen-Max-Latest. Using a "first half to second half" continuation paradigm with 3 samples per film, we obtained 303 valid samples (GPT-5.2: 157, 98.7% validity; Qwen-Max: 146, 91.8% validity). Evaluation integrates ROUGE-L, Structural Similarity, and LLM-as-Judge scoring (DeepSeek-Reasoner). Statistical analysis of 144 paired samples reveals: Qwen-Max achieves marginally higher ROUGE-L (0.2230 vs 0.2114, d=-0.43); however, GPT-5.2 significantly outperforms in structural preservation (0.93 vs 0.75, d=0.46), overall quality (44.79 vs 25.72, d=1.04), and composite scores (0.50 vs 0.39, d=0.84). The overall quality effect size reaches large effect level (d>0.8). GPT-5.2 excels in character consistency, tone-style matching, and format preservation, while Qwen-Max shows deficiencies in generation stability. This study provides a reproducible framework for LLM evaluation in Chinese creative writing.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.LG
Reflecting in the Reflection: Integrating a Socratic Questioning Framework into Automated AI-Based Question Generation
Ondřej Holub, Essi Ryymin and Rodrigo Alves
Designing good reflection questions is pedagogically important but time-consuming and unevenly supported across teachers. This paper introduces a reflection-in-reflection framework for automated generation of reflection questions with large
language models (LLMs). Our approach coordinates two role-specialized
agents, a Student-Teacher and a Teacher-Educator, that engage in a Socratic multi-turn dialogue to iteratively refine a single question given a teacher-specified topic, key concepts, student level, and optional instructional materials. The Student-Teacher proposes candidate questions with brief rationales, while the Teacher-Educator evaluates them along clarity, depth, relevance, engagement, and conceptual interconnections, responding only with targeted coaching questions or a fixed signal to stop the dialogue. We evaluate the framework in an authentic lower-secondary ICT setting on the topic, using GPT-4o-mini as the backbone model and a stronger GPT- 4-class LLM as an external evaluator in pairwise comparisons of clarity, relevance, depth, and overall quality. First, we study how interaction design and context (dynamic vs.fixed iteration counts; presence or absence of student level and materials) affect question quality. Dynamic stopping combined with contextual information consistently outperforms fixed 5- or 10-step refinement, with very long dialogues prone to drift or over-complication. Second, we show that our two-
agent protocol produces questions that are judged substantially more relevant and deeper, and better overall, than a one-shot baseline using the same backbone model.
Read the full abstract ▾
Computation and Language
RECAP: Resistance Capture in Text-based Mental Health Counseling with Large Language Models
Anqi Li, Yuqian Chen, Yu Lu et al. (+3 more)
Recognizing and navigating client resistance is critical for effective mental health counseling, yet detecting such behaviors is particularly challenging in text-based interactions. Existing NLP approaches oversimplify resistance categories, ignore the sequential dynamics of therapeutic interventions, and offer limited interpretability. To address these limitations, we propose PsyFIRE, a theoretically grounded framework capturing 13 fine-grained resistance behaviors alongside collaborative interactions. Based on PsyFIRE, we construct the ClientResistance corpus with 23,930 annotated utterances from real-world Chinese text-based counseling, each supported by context-specific rationales. Leveraging this
dataset, we develop RECAP, a two-stage framework that detects resistance and fine-grained resistance types with explanations. RECAP achieves 91.25% F1 for distinguishing collaboration and resistance and 66.58% macro-F1 for fine-grained resistance categories classification, outperforming leading
prompt-based LLM baselines by over 20 points. Applied to a separate counseling
dataset and a pilot study with 62 counselors, RECAP reveals the prevalence of resistance, its negative impact on therapeutic relationships and demonstrates its potential to improve counselors' understanding and intervention strategies.
Read the full abstract ▾
All authors Anqi Li, Yuqian Chen, Yu Lu, Zhaoming Chen, Yuan Xie, Zhenzhong Lan
Computation and Language · Cross-listed from cs.LG
Mechanism Shift During Post-training from Autoregressive to Masked Diffusion Language Models
Injin Kong, Hyoungjoon Lee and Yohan Jo
Post-training pretrained autoregressive models (ARMs) into masked
diffusion models (MDMs) provides an efficient route to
diffusion language modeling, but it remains unclear whether the resulting models reuse inherited autoregressive computation or reorganize it for non-autoregressive generation. We compare two 7B ARM-MDM families across four controlled diagnostic tasks and find a task-dependent mechanism shift. On prefix-dominant tasks, MDMs largely preserve inherited high-attribution pathways or exhibit only modest changes in where computation occurs. On globally constrained tasks, the reorganization is substantially stronger, with task-relevant computation shifting toward earlier layers. This depth-wise pattern persists across
prompt resampling,
circuit budgets, and tested inference budgets, while targeted ablations support the functional importance of the identified structures under the tested intervention protocols. At the component level, diagnostic probes suggest that ARMs rely more strongly on sharply specialized components, whereas MDMs exhibit weaker single-component specialization and more diffuse output-space
alignment. Together, these results suggest that
diffusion post-training selectively preserves or reorganizes inherited computation according to task structure, rather than uniformly replacing autoregressive mechanisms.
Read the full abstract ▾
Computation and Language
Render-of-Thought: Rendering Textual Chain-of-Thought as Images for Visual Latent Reasoning
Yifan Wang, Shiyu Li, Peiming Li et al. (+3 more)
Chain-of-Thought (CoT) prompting has achieved remarkable success in unlocking the
reasoning capabilities of Large
Language Models (LLMs). Although CoT prompting enhances
reasoning, its verbosity imposes substantial computational overhead. Recent works often focus exclusively on outcome
alignment and lack supervision on the intermediate
reasoning process. These deficiencies obscure the analyzability of the latent
reasoning chain. To address these challenges, we introduce Render-of-Thought (RoT), the first framework to reify the
reasoning chain by rendering textual steps into images, making the latent rationale explicit and traceable. Specifically, we leverage the
vision encoders of existing
Vision Language Models (VLMs) as semantic anchors to align the
vision embeddings with the textual space. This design ensures plug-and-play implementation without incurring additional pre-training overhead. Extensive experiments on mathematical and logical
reasoning benchmarks demonstrate that our method achieves 3-4x
token compression and substantial inference acceleration compared to explicit CoT. Furthermore, it maintains competitive performance against other methods, validating the feasibility of this paradigm. Our code is available at https://github.com/TencentBAC/RoT
Read the full abstract ▾
All authors Yifan Wang, Shiyu Li, Peiming Li, Xiaochen Yang, Yang Tang, Zheng Wei
Computation and Language · Cross-listed from cs.CV
DeepMoLM: Leveraging Visual and Geometric Structural Information for Molecule-Text Modeling
Jing Lan, Hexiao Ding, Hongzhao Chen et al. (+8 more)
AI models for drug discovery and chemical literature mining must interpret molecular images and generate outputs consistent with 3D geometry and stereochemistry. Most molecular
language models rely on strings or graphs, while
vision-
language models often miss stereochemical details and struggle to map continuous 3D structures into discrete
tokens. We propose DeepMoLM: Deep Molecular
Language M odeling, a dual-view framework that grounds high-resolution molecular images in
geometric invariants derived from molecular conformations. DeepMoLM preserves high-frequency evidence from 1024 $\times$ 1024 inputs, encodes conformer neighborhoods as discrete Extended 3-Dimensional Fingerprints, and fuses visual and
geometric streams with cross-
attention, enabling physically grounded generation without atom coordinates. DeepMoLM improves PubChem captioning with a 12.3% relative METEOR gain over the strongest generalist baseline while staying competitive with specialist methods. It produces valid numeric outputs for all property queries and attains MAE 13.64 g/mol on Molecular Weight and 37.89 on Complexity in the specialist setting. On ChEBI-20 description generation from images, it exceeds generalist baselines and matches state-of-the-art
vision-
language models. Code is available at https://github.com/1anj/DeepMoLM.
Read the full abstract ▾
All authors Jing Lan, Hexiao Ding, Hongzhao Chen, Yufeng Jiang, Nga-Chun Ng, Gwing Kei Yip, Gerald W.Y. Cheng, Yunlin Mao, Jing Cai, Liang-ting Lin, Jung Sun Yoo
Computation and Language · Cross-listed from eess.AS
AQAScore: Evaluating Semantic Alignment in Text-to-Audio Generation via Audio Question Answering
Chun-Yi Kuan, Kai-Wei Chang and Hung-yi Lee
Although text-to-audio generation has made remarkable progress in realism and diversity, the development of evaluation metrics has not kept pace. Widely-adopted approaches, typically based on
embedding similarity like CLAPScore, effectively measure general relevance but remain limited in fine-grained semantic
alignment and compositional
reasoning. To address this, we introduce AQAScore, a backbone-agnostic evaluation framework that leverages the
reasoning capabilities of audio-aware large
language models (ALLMs). AQAScore reformulates assessment as a probabilistic semantic verification task; rather than relying on open-ended text generation, it estimates
alignment by computing the exact log-probability of a "Yes" answer to targeted semantic queries. We evaluate AQAScore across multiple
benchmarks, including human-rated relevance, pairwise comparison, and compositional
reasoning tasks. Experimental results show that AQAScore consistently achieves higher correlation with human judgments than similarity-based metrics and generative prompting baselines, showing its effectiveness in capturing subtle semantic inconsistencies and scaling with the capability of underlying ALLMs.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.CV
HERMES: KV Cache as Hierarchical Memory for Efficient Streaming Video Understanding
Haowei Zhang, Shudong Yang, Jinlan Fu et al. (+2 more)
Recent advancements in
Multimodal Large
Language Models (MLLMs) have demonstrated significant improvement in offline
video understanding. However, extending these capabilities to streaming
video inputs, remains challenging, as existing models struggle to simultaneously maintain stable understanding performance, real-time responses, and low GPU
memory overhead. To address this challenge, we propose HERMES, a novel training-free architecture for real-time and accurate understanding of
video streams. Based on a mechanistic
attention investigation, we conceptualize KV cache as a hierarchical
memory framework that encapsulates
video information across multiple granularities. During inference, HERMES reuses a compact KV cache, enabling efficient streaming understanding under resource constraints. Notably, HERMES requires no auxiliary computations upon the arrival of user queries, thereby guaranteeing real-time responses for continuous
video stream interactions, which achieves 10$\times$ faster TTFT compared to prior SOTA. Even when reducing
video tokens by up to 68% compared with uniform sampling, HERMES achieves superior or comparable accuracy across all
benchmarks, with up to 11.4% gains on streaming
datasets.
Read the full abstract ▾
All authors Haowei Zhang, Shudong Yang, Jinlan Fu, See-Kiong Ng, Xipeng Qiu
Computation and Language
Typhoon OCR: Open Vision-Language Model For Thai Document Extraction
Surapon Nonesung, Natapong Nitarach, Teetouch Jaknamon et al. (+2 more)
Document extraction is a core component of digital workflows, yet existing
vision-
language models (VLMs) predominantly favor high-resource
languages. Thai presents additional challenges due to script complexity from non-latin letters, the absence of explicit word boundaries, and the prevalence of highly unstructured real-world documents, limiting the effectiveness of current open-source models. This paper presents Typhoon OCR, an open VLM for document extraction tailored for Thai and English. The model is fine-tuned from
vision-
language backbones using a Thai-focused training
dataset. The
dataset is developed using a multi-stage data construction pipeline that combines traditional OCR, VLM-based restructuring, and curated synthetic data. Typhoon OCR is a unified framework capable of text transcription, layout reconstruction, and document-level structural consistency. The latest iteration of our model, Typhoon OCR V1.5, is a compact and inference-efficient model designed to reduce reliance on metadata and simplify deployment. Comprehensive evaluations across diverse Thai document categories, including financial reports, government forms, books, infographics, and handwritten documents, show that Typhoon OCR achieves performance comparable to or exceeding larger frontier proprietary models, despite substantially lower computational cost. The results demonstrate that open
vision-
language OCR models can achieve accurate text extraction and layout reconstruction for Thai documents, reaching performance comparable to proprietary systems while remaining lightweight and deployable.
Read the full abstract ▾
All authors Surapon Nonesung, Natapong Nitarach, Teetouch Jaknamon, Pittawat Taveekitworachai, Kunat Pipatanakul
Computation and Language · Cross-listed from cs.LG
PCL-Reasoner-V1.5: Advancing Math Reasoning with Offline Reinforcement Learning
Yao Lu, Dengdong Fan, Jianzheng Nie et al. (+4 more)
We present PCL-Reasoner-V1.5, a 32-billion-parameter large
language model (LLM) for mathematical
reasoning. The model is built upon Qwen2.5-32B and refined via supervised fine-tuning (SFT) followed by reinforcement learning (RL). A central innovation is our proposed offline RL method, which provides superior training stability and efficiency over standard online RL methods such as GRPO. Our model achieves state-of-the-art performance among models post-trained on Qwen2.5-32B, attaining average accuracies of 90.9% on AIME 2024 and 85.6% on AIME 2025. Our work demonstrates offline RL as a stable and efficient paradigm for advancing
reasoning in LLMs. All experiments were conducted on Huawei Ascend 910C NPUs.
Read the full abstract ▾
All authors Yao Lu, Dengdong Fan, Jianzheng Nie, Fan Xu, Jie Chen, Bin Zhou, Yonghong Tian
Computation and Language
DARL: Encouraging Diverse Answers for General Reasoning without Verifiers
Chongxuan Huang, Lei Lin, Xiaodong Shi et al. (+2 more)
Reinforcement Learning with Verifiable Rewards (RLVR) has demonstrated promising gains in enhancing the
reasoning capabilities of large
language models. However, its dependence on domain-specific verifiers significantly restricts its applicability to open and general domains. Recent efforts such as RLPR have extended RLVR to general domains, enabling training on broader
datasets and achieving improvements over RLVR. However, a notable limitation of these methods is their tendency to overfit to reference answers, which constrains the model's ability to generate diverse outputs. This limitation is particularly pronounced in open-ended tasks such as writing, where multiple plausible answers exist. To address this, we propose DARL, a simple yet effective reinforcement learning framework that encourages the generation of diverse answers within a controlled deviation range from the reference while preserving
alignment with it. Our framework is fully compatible with existing general reinforcement learning methods and can be seamlessly integrated without additional verifiers. Extensive experiments on thirteen
benchmarks demonstrate consistent improvements in
reasoning performance. Notably, DARL surpasses RLPR, achieving average gains of 1.3 points on six
reasoning benchmarks and 9.5 points on seven general
benchmarks, highlighting its effectiveness in improving both
reasoning accuracy and output diversity.
Read the full abstract ▾
All authors Chongxuan Huang, Lei Lin, Xiaodong Shi, Wenping Hu, Ruiming Tang
Computation and Language
ClaimDB: A Fact Verification Benchmark over Large Structured Data
Michael Theologitis, Preetam Prabhu Srikar Dammu, Chirag Shah et al. (+1 more)
Real-world fact-checking often involves verifying claims grounded in structured data at scale. Despite substantial progress in fact-verification
benchmarks, this setting remains largely underexplored. In this work, we introduce ClaimDB, a fact-verification
benchmark where the evidence for claims is derived from compositions of millions of records and multiple tables. ClaimDB consists of 80 unique real-life databases covering a wide range of domains, from governance and healthcare to media, education and the natural sciences. At this scale, verification approaches that rely on "reading" the evidence break down, forcing a timely shift toward
reasoning in executable programs. We conduct extensive experiments with 30 state-of-the-art proprietary and open-source (below 70B) LLMs and find that more than half score below 55% accuracy. Our analysis also reveals that both closed- and open-source models struggle with abstention – the ability to admit that there is no evidence to decide – raising doubts about their reliability in high-stakes data analysis tasks. We release the
benchmark, code, and the LLM leaderboard at https://claimdb.github.io .
Read the full abstract ▾
All authors Michael Theologitis, Preetam Prabhu Srikar Dammu, Chirag Shah, Dan Suciu
Computation and Language
AdaTIR: Adaptive Tool-Integrated Reasoning via Difficulty-Aware Policy Optimization
Zhaiyu Fang and Ruipeng Sun
Tool-Integrated
Reasoning (TIR) has significantly enhanced the capabilities of Large
Language Models (LLMs), yet current
agents tend to exhibit cognitive offloading, redundantly invoking external tools even for simple tasks. In this paper, we suggest that true agentic intelligence requires not just tool invocation, but the adaptive wisdom to discern when to use them. We propose AdaTIR, a framework that shifts the paradigm from static tool invocation to difficulty-aware
reasoning internalization. By introducing a difficulty-aware efficiency reward, AdaTIR dynamically adjusts tool budgets based on task complexity–internalizing
reasoning for simple tasks while selectively invoking tools for complex tasks. Furthermore, we identify a sign reversal problem where tool penalties outweigh correctness rewards, mistakenly penalizing correct rollouts with negative advantages. To resolve this, we propose Clipped Advantage Shaping (CAS), which ensures that correctness remains the primary objective while using efficiency as a secondary constraint. Empirical results demonstrate that AdaTIR reduces tool calls by up to 97.6% on simple tasks and 28.2% on complex challenges while maintaining or enhancing accuracy. Notably, AdaTIR successfully internalizes
reasoning, outperforming baselines by 4.8% on AIME 2024 even when tool access is strictly disabled.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.AI
Gaming the Judge: Unfaithful Chain-of-Thought Can Undermine Agent Evaluation
Muhammad Khalifa, Lajanugen Logeswaran, Jaekyeom Kim et al. (+6 more)
Large
language models (LLMs) are increasingly used as judges to evaluate
agent performance, particularly in non-verifiable settings where judgments rely on
agent trajectories including chain-of-thought (CoT)
reasoning. This paradigm implicitly assumes that the
agent's CoT faithfully reflects both its internal
reasoning and the underlying environment state. We show this assumption is brittle: LLM judges are highly susceptible to manipulation of
agent reasoning traces. By systematically rewriting
agent CoTs while holding actions and observations fixed, we demonstrate that manipulated
reasoning alone can inflate false positive rates of state-of-the-art VLM judges by up to 90% across 800 trajectories spanning diverse web tasks. We study manipulation strategies spanning style-based approaches that alter only the presentation of
reasoning and content-based approaches that fabricate signals of task progress, and find that content-based manipulations are consistently more effective. We evaluate prompting-based techniques and scaling judge-time compute, which reduce but do not fully eliminate susceptibility to manipulation. Our findings reveal a fundamental vulnerability in LLM-based evaluation and highlight the need for judging mechanisms that verify
reasoning claims against observable evidence.
Read the full abstract ▾
All authors Muhammad Khalifa, Lajanugen Logeswaran, Jaekyeom Kim, Sungryull Sohn, Yunxiang Zhang, Moontae Lee, Hao Peng, Lu Wang, Honglak Lee
Computation and Language · Cross-listed from cs.CR
NeuroFilter: Activation-Based Guardrails for Privacy-Conscious LLM Agents
Saswat Das and Ferdinando Fioretto
Agentic Large
Language Models (LLMs) are models able to reason, plan, and execute tools over unstructured data. These abilities enable transformative applications in domains spanning personal assistants, financial, and legal domains, etc. While these systems can substantially improve productivity and service quality, effective agency typically requires access to sensitive personal or organizational information. However, this access introduces critical inference-time
privacy risks, specifically regarding contextually appropriate information disclosure. While recent studies highlight the inability of agentic LLMs to consistently adhere to
privacy norms, existing defenses often rely on auxiliary LLM-based monitors. However, these defenses are expensive and offer limited protection against attacks that are robust to semantic censorship. To contrast this background, this paper proposes a notion of
privacy filters based on activation probing. We show that these filters are both computationally efficient and effective for both single-turn and multi-turn conversational settings. Furthermore, this work provides the first systematic investigation into probing model internals across a conversation trajectory, moving beyond static, single-
prompt analysis to capture the evolving state of
privacy-sensitive interactions, illustrating their applicability across contexts, attack styles, multiple base models, and challenging settings such as mosaic attacks, while providing nuanced insights into the nature of the discovered concepts and modular construction of probes.
Read the full abstract ▾
Computation and Language
Say Anything but This: When Tokenizer Betrays Reasoning in LLMs
Navid Ayoobi, Marcus I Armstrong and Arjun Mukherjee
Large
language models (LLMs) reason over discrete
token ID sequences, yet modern subword tokenizers routinely produce non-unique encodings: multiple
token ID sequences can detokenize to identical surface strings. This representational mismatch creates an unmeasured fragility wherein
reasoning processes can fail. LLMs may treat two internal representations as distinct "words" even when they are semantically identical at the text level. In this work, we show that tokenization can betray LLM
reasoning through one-to-many
token ID mappings. We introduce a tokenization-consistency probe that requires models to replace designated target words in context while leaving all other content unchanged. The task is intentionally simple at the surface level, enabling us to attribute failures to tokenizer-detokenizer artifacts rather than to knowledge gaps or parameter limitations. Through analysis of over 11000 replacement trials across state-of-the-art open-source LLMs, we find a non-trivial rate of outputs exhibit phantom edits: cases where models operate under the illusion of correct
reasoning, a phenomenon arising from tokenizer-induced representational defects. We further analyze these cases and provide a taxonomy of eight systematic tokenizer artifacts, including whitespace-boundary shifts and intra-word resegmentation. These findings indicate that part of apparent
reasoning deficiency originates in the tokenizer layer, motivating tokenizer-level remedies before incurring the cost of training ever-larger models on ever-larger corpora.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.AI
MAS-Orchestra: Understanding and Improving Multi-Agent Reasoning Through Holistic Orchestration and Controlled Benchmarks
Zixuan Ke, Yifei Ming, Austin Xu et al. (+7 more)
While multi-
agent systems (MAS) promise elevated intelligence through coordination of
agents, current approaches to automatic MAS design under-deliver. Such shortcomings stem from two key factors: (1) methodological complexity -
agent orchestration is performed using sequential, code-level execution that limits global system-level holistic
reasoning and scales poorly with
agent complexity - and (2) efficacy uncertainty - MAS are deployed without understanding if there are tangible benefits compared to single-
agent systems (SAS). We propose MASOrchestra, a training-time framework that formulates MAS orchestration as a function-calling reinforcement learning problem with holistic orchestration, generating an entire MAS at once. In MAS-Orchestra, complex, goal-oriented subagents are abstracted as callable functions, enabling global
reasoning over system structure while hiding internal execution details. To rigorously study when and why MAS are beneficial, we introduce MASBENCH, a controlled
benchmark that characterizes tasks along five axes: Depth, Horizon, Breadth, Parallel, and Robustness. Our analysis reveals that MAS gains depend critically on task structure, verification protocols, and the capabilities of both orchestrator and subagents, rather than holding universally. Guided by these insights, MAS-Orchestra achieves consistent improvements on public
benchmarks including mathematical
reasoning, multi-hop QA, and search-based QA, while achieving more than 10x efficiency over strong baselines. Together, MAS-Orchestra and MASBENCH enable better training and understanding of MAS in the pursuit of multi-
agent intelligence.
Read the full abstract ▾
All authors Zixuan Ke, Yifei Ming, Austin Xu, Ryan Chin, Xuan-Phi Nguyen, Prathyusha Jwalapuram, Jiayu Wang, Semih Yavuz, Caiming Xiong, Shafiq Joty
Computation and Language · Cross-listed from cs.CV
Forest-Chat: Adapting Vision-Language Agents for Interactive Forest Change Analysis
James Brock, Ce Zhang and Nantheera Anantrasirichai
The increasing availability of high-resolution satellite imagery, together with advances in deep learning, creates new opportunities for forest monitoring workflows. Two central challenges in this domain are pixel-level change detection and semantic change interpretation, particularly for complex forest dynamics. While large
language models (LLMs) are increasingly adopted for data exploration, their integration with
vision-
language models (VLMs) for remote sensing image change interpretation (RSICI) remains underexplored, especially beyond urban environments. This paper introduces Forest-Chat, an LLM-driven
agent for forest change analysis, enabling natural
language querying across multiple RSICI tasks, including change detection and captioning, object counting, deforestation characterisation, and change
reasoning. Forest-Chat builds upon a multi-level change interpretation (MCI)
vision-
language backbone with LLM-based orchestration, incorporating zero-shot change detection via AnyChange and
multimodal LLM-based zero-shot change captioning and refinement. To support adaptation and evaluation in forest environments, we introduce the Forest-Change
dataset, comprising bi-temporal satellite imagery, pixel-level change masks, and semantic change captions via human annotation and rule-based methods. Forest-Chat achieves mIoU and BLEU-4 scores of 67.10% and 40.17% on Forest-Change, and 88.13% and 34.41% on LEVIR-MCI-Trees, a tree-focused subset of LEVIR-MCI. In a zero-shot capacity, it achieves 60.15% and 34.00% on Forest-Change, and 47.32% and 18.23% on LEVIR-MCI-Trees. Further experiments demonstrate the value of caption refinement for injecting geographic domain knowledge into supervised captions, and the system's limited label domain transfer onto JL1-CD-Trees. These findings demonstrate that interactive, LLM-driven systems can support accessible and interpretable forest change analysis.
Read the full abstract ▾
Computation and Language
SearchGym: Bootstrapping Real-World Search Agents via Cost-Effective and High-Fidelity Environment Simulation
Xichen Zhang, Ziyi He, Yinghao Zhu et al. (+6 more)
Search
agents have emerged as a pivotal paradigm for solving open-ended, knowledge-intensive
reasoning tasks. However, training these
agents via Reinforcement Learning (RL) faces a critical dilemma: interacting with live commercial Web APIs is prohibitively expensive, while relying on static data snapshots often introduces noise due to data misalignment. This misalignment generates corrupted reward signals that destabilize training by penalizing correct
reasoning or rewarding
hallucination. To address this, we propose SearchGym, a simulation environment designed to bootstrap robust search
agents. SearchGym employs a rigorous generative pipeline to construct a verifiable knowledge graph and an aligned document corpus, ensuring that every
reasoning task is factually grounded and strictly solvable. Building on this controllable environment, we introduce SearchGym-RL, a curriculum learning methodology that progressively optimizes
agent policies through purified feedback, evolving from basic interactions to complex, long-horizon planning. Extensive experiments across the Llama and Qwen families demonstrate strong Sim-to-Real generalization. Notably, our Qwen2.5-7B-Base model trained within SearchGym surpasses the web-enhanced ASearcher baseline across nine diverse
benchmarks by an average relative margin of 10.6%. Our results validate that high-fidelity simulation serves as a scalable and highly cost-effective methodology for developing capable search
agents.
Read the full abstract ▾
All authors Xichen Zhang, Ziyi He, Yinghao Zhu, Sitong Wu, Shaozuo Yu, Meng Chu, Wenhu Zhang, Haoru Tan, Jiaya Jia
Computation and Language · Cross-listed from cs.HC
Designing KRIYA: An AI Companion for Wellbeing Self-Reflection
Shanshan Zhu, Wenxuan Song, Jiayue Melissa Shi et al. (+3 more)
Most personal wellbeing apps present summative dashboards of health and physical activity metrics, yet many users struggle to translate this information into meaningful understanding. These apps commonly support engagement through goals, reminders, and structured targets, which can reinforce comparison, judgment, and performance anxiety. To explore a complementary approach that prioritizes self-reflection, we design KRIYA, an AI wellbeing companion that supports co-interpretive engagement with personal wellbeing data. KRIYA aims to collaborate with users to explore questions, explanations, and future scenarios through features such as Comfort Zone, Detective Mode, and What-If Planning. We conducted semi-structured interviews with 18 college students interacting with a KRIYA prototype using hypothetical data. Our findings show that through KRIYA interaction, users framed engaging with wellbeing data as interpretation rather than performance, experienced reflection as supportive or pressuring depending on emotional framing, and developed trust through transparency. We discuss design implications for AI companions that support curiosity, self-compassion, and reflective sensemaking of personal health data.
Read the full abstract ▾
All authors Shanshan Zhu, Wenxuan Song, Jiayue Melissa Shi, Dong Whi Yoo, Karthik S. Bhat, Koustuv Saha
Computation and Language
Social Caption: Evaluating Social Understanding in Multimodal Models
Leena Mathur, Bhaavanaa Thumu, Youssouf Kebe et al. (+1 more)
Social understanding abilities are crucial for
multimodal large
language models (MLLMs) to interpret human social interactions. We introduce SOCIAL CAPTION, a framework grounded in interaction theory to evaluate social understanding abilities of MLLMs along three dimensions: Social Inference (SI), the ability to make accurate inferences about interactions; Holistic Social Analysis (HSA), the ability to generate comprehensive descriptions of interactions; Directed Social Analysis (DSA), the ability to generate relevant information from interactions. We analyze factors influencing model performance in social understanding, such as scale, architectural design, and spoken context. Experiments with MLLM judges demonstrate a path towards scaling automated evaluation of
multimodal social understanding.
Read the full abstract ▾
All authors Leena Mathur, Bhaavanaa Thumu, Youssouf Kebe, Louis-Philippe Morency
Computation and Language
Rewarding How Models Think Pedagogically: Integrating Pedagogical Reasoning and Thinking Rewards for LLMs in Education
Unggi Lee, Jiyeong Bae, Jaehyeon Park et al. (+7 more)
Large
language models (LLMs) are increasingly deployed as intelligent tutoring systems, yet research on optimizing LLMs specifically for educational contexts remains limited. Recent works have proposed reinforcement learning approaches for training LLM tutors, but these methods focus solely on optimizing visible responses while neglecting the model's internal thinking process. We introduce PedagogicalRL-Thinking, a framework that extends pedagogical
alignment to
reasoning LLMs in education through two novel approaches: (1) Pedagogical
Reasoning Prompting, which guides internal
reasoning using domain-specific educational theory rather than generic instructions; and (2) Thinking Reward, which explicitly evaluates and reinforces the pedagogical quality of the model's
reasoning traces. Our experiments reveal that domain-specific, theory-grounded prompting outperforms generic prompting, and that Thinking Reward is most effective when combined with pedagogical prompting. Furthermore, models trained only on mathematics tutoring dialogues show improved performance on educational
benchmarks not seen during training, while preserving the base model's factual knowledge. Our quantitative and qualitative analyses reveal that pedagogical thinking reward produces systematic
reasoning trace changes, with increased pedagogical
reasoning and more structured instructional decision-making in the tutor's thinking process.
Read the full abstract ▾
All authors Unggi Lee, Jiyeong Bae, Jaehyeon Park, Haeun Park, Taejun Park, Younghoon Jeon, Sungmin Cho, Junbo Koh, Yeil Jeong, Gyeonggeon Lee
Computation and Language
Self-Blinding and Counterfactual Self-Simulation Mitigate Biases and Sycophancy in Large Language Models
Brian Christian and Matan Mazor
Fair decisions require ignoring irrelevant, potentially biasing, information. To achieve this, decision-makers need to approximate what decision they would have made had they not known certain facts, such as the gender or race of a job candidate. This counterfactual self-simulation is notoriously hard for humans, leading to biased judgments even by well-meaning actors. Here we show that large
language models (LLMs) suffer from similar limitations in their ability to approximate what decisions they would make under counterfactual knowledge in offsetting gender and race biases and overcoming sycophancy. We show that prompting models to ignore or pretend not to know biasing information fails to offset these biases and occasionally backfires. However, unlike humans, LLMs can be given access to a ground-truth model of their own counterfactual cognition – their own API. We show that this access to the responses of a blinded replica enables fairer decisions, while providing greater transparency to distinguish implicit from intentionally biased behavior.
Read the full abstract ▾