arXiv subjects

Bingbing Wen

Publications and source records attributed to Bingbing Wen.

At least 19 recordsLinked to original sources

arXiv2026

Agentic Abstention: Do Agents Know When to Stop Instead of Act?

LLM agents are expected to act over multiple turns, using search, browsing interfaces, and terminal tools to complete user goals. Yet not every goal is well specified or achievable in the available environment. In such cases, a reliable agent should recognize that further interaction is unlikely to help and abstain from additional tool calls. We define Agentic Abstention, the problem of deciding when an agent should stop acting under uncertainty. Unlike standard LLM abstention, which is usually evaluated as a single-turn answer-or-abstain decision, agentic abstention is a sequential decision problem: an agent can answer, abstain, or gather more information at each turn, and the need to abstain may only become clear after interacting with the environment. We study this problem across web shopping, terminal environments, and question answering, evaluating 13 LLM-as-agent systems and 2 agent scaffolds on more than 28,000 tasks. Our results show that the main challenge is not only whether agents can abstain, but also when they abstain. Some agents never abstain when they should, while others do so only after many unnecessary interactions. This gap is especially large on tasks where the instruction appears feasible until the environment reveals otherwise (e.g., no valid result matches the instruction). We further find that model scale, reasoning, and agent scaffolding affect abstention in different ways, where larger or more capable models sometimes perform worse at timely abstention. Finally, we introduce CONVOLVE, a context engineering method for improving agentic abstention that distills full interaction trajectories into reusable stopping rules. On WebShop, CONVOLVE substantially improves timely abstention without updating model parameters, raising Llama-3.3-70B's timely recall rate from 26.7 to 57.4. Our dataset and code are available at https://lhannnn.github.io/agentic-abstention

cs.AI↗

arXiv2026

Illusions of the Gold Standard: A Large-scale Analysis of Human Evaluation Protocols for Long-form Text Generation

Human evaluation plays a critical role in assessing the quality of generated text. However, the reliability and reproducibility of these evaluations depend on transparent and well-documented protocols -- details that are frequently missing in current practice. In this work, we conduct a large-scale analysis of human evaluation protocols for evaluating long-form generation tasks in *CL conference publications from 2023--2025, including a full manual review of 284 papers and LLM-assisted analysis for another 1.8k+ papers. We define a set of 20 reportable criteria related to reproducibility of human evaluation studies, and apply these criteria to systematically examine reporting norms and practices within the community. We find widespread under-reporting of important aspects of human evaluation study design, leading to ambiguity about what was measured and how, who contributed judgments, and how judgments should be interpreted. Based on these findings, we outline actionable recommendations to support more transparent and reproducible reporting in future research. Our analysis code and annotated dataset can be found at: https://github.com/larchlab/Illusions-of-the-Gold-Standard

cs.CL↗

arXiv2026

STOP: Structured On-Policy Pruning of Long-Form Reasoning in Low-Data Regimes

Long chain-of-thought (Long CoT) reasoning improves performance on multi-step problems, but it also induces overthinking: models often generate low-yield reasoning that increases inference cost and latency. This inefficiency is especially problematic in low-data fine-tuning regimes, where real applications adapt reasoning models with limited supervision and cannot rely on large-scale teacher distillation or heavy test-time control. To address this, we propose STOP (Structured On-policy Pruning), an on-policy algorithm for analyzing and pruning long-form reasoning traces. STOP constructs self-distilled traces from the model. Then it maps each trace into a structured reasoning interface through node segmentation, taxonomy annotation, and reasoning-tree construction. On top of this interface, we introduce ECN (Earliest Correct Node), which retains the shortest prefix ending at the earliest node that both functions as an answering conclusion and yields the correct final answer, removing redundant post-solution reasoning while preserving semantic continuity. Experiments on DeepSeek-R1-Distill-Qwen-7B and DeepSeek-R1-Distill-LLaMA-3-8B across GSM8K, Math 500, and AIME 2024 show that STOP reduces generated tokens by 19.4-42.4% while largely preserving accuracy in low-data fine-tuning. Beyond efficiency, our analyses show that STOP induces much smaller distributional shift than teacher-guided pruning, improves the structural efficiency of generated reasoning, and reallocates reasoning effort away from redundant verification and backtracking toward more productive exploration.

cs.CL↗

arXiv2026

MixAtlas: Uncertainty-aware Data Mixture Optimization for Multimodal LLM Midtraining

Domain reweighting can improve sample efficiency and downstream generalization, but data-mixture optimization for multimodal midtraining remains largely unexplored. Current multimodal training recipes tune mixtures along a single dimension, typically data format or task type. We introduce MixAtlas, a method that produces benchmark-targeted data recipes that can be inspected, adapted, and transferred to new corpora. MixAtlas decomposes the training corpus along two axes: image concepts (10 visual-domain clusters discovered via CLIP embeddings) and task supervision (5 objective types including captioning, OCR, grounding, detection, and VQA). Using small proxy models (Qwen2-0.5B) paired with a Gaussian-process surrogate and GP-UCB acquisition, MixAtlas searches the resulting mixture space with the same proxy budget as regression-based baselines but finds better-performing mixtures. We evaluate on 10 benchmarks spanning visual understanding, document reasoning, and multimodal reasoning. On Qwen2-7B, optimized mixtures improve average performance by 8.5%-17.6% over the strongest baseline; on Qwen2.5-7B, gains are 1.0%-3.3%. Both settings reach baseline-equivalent training loss in up to 2 times fewer steps. Recipes discovered on 0.5B proxies transfer to 7B-scale training across Qwen model families.

cs.LG↗

arXiv2026

Towards Dynamic Dense Retrieval with Routing Strategy

The \textit{de facto} paradigm for applying dense retrieval (DR) to new tasks involves fine-tuning a pre-trained model for a specific task. However, this paradigm has two significant limitations: (1) It is difficult adapt the DR to a new domain if the training dataset is limited. (2) Old DR models are simply replaced by newer models that are trained from scratch when the former are no longer up to date. Especially for scenarios where the model needs to be updated frequently, this paradigm is prohibitively expensive. To address these challenges, we propose a novel dense retrieval approach, termed \textit{dynamic dense retrieval} (DDR). DDR uses \textit{prefix tuning} as a \textit{module} specialized for a specific domain. These modules can then be compositional combined with a dynamic routing strategy, enabling highly flexible domain adaptation in the retrieval part. Extensive evaluation on six zero-shot downstream tasks demonstrates that this approach can surpass DR while utilizing only 2\% of the training parameters, paving the way to achieve more flexible dense retrieval in IR. We see it as a promising future direction for applying dense retrieval to various tasks.

cs.IR↗

arXiv2026

Clarify or Answer: Reinforcement Learning for Agentic VQA with Context Under-specification

Real-world visual question answering (VQA) is often context-dependent: an image-question pair may be under-specified, such that the correct answer depends on external information that is not observable in the image. In such cases, directly answering can lead to confident but incorrect predictions. We propose CoA(Clarify-or-Answer), an ask-or-answer agent that separately models the decision to ask or answer, and what to ask if needed. CoA first determines whether clarification is necessary; if so, it asks a single focused question and then incorporates the response to produce the final answer. We introduce CONTEXTCLARIFY with a set of ambiguous VQA questions and the contrast set that is non-ambiguous. We further introduce GRPO-CR (Clarification Reasoning), a reinforcement learning approach that optimizes clarification question generation with multiple reward signals encouraging well-formed, focused, non-trivial questions that resolve ambiguity. Across three VLLMs and three datasets, CoA achieves consistent improvements at both the module and system levels, improving end-to-end VQA accuracy by an average of +15.3 points (83%) over prompting-based baselines

cs.CL↗

arXiv2025

SusBench: An Online Benchmark for Evaluating Dark Pattern Susceptibility of Computer-Use Agents

As LLM-based computer-use agents (CUAs) begin to autonomously interact with real-world interfaces, understanding their vulnerability to manipulative interface designs becomes increasingly critical. We introduce SusBench, an online benchmark for evaluating the susceptibility of CUAs to UI dark patterns, designs that aim to manipulate or deceive users into taking unintentional actions. Drawing nine common dark pattern types from existing taxonomies, we developed a method for constructing believable dark patterns on real-world consumer websites through code injections, and designed 313 evaluation tasks across 55 websites. Our study with 29 participants showed that humans perceived our dark pattern injections to be highly realistic, with the vast majority of participants not noticing that these had been injected by the research team. We evaluated five state-of-the-art CUAs on the benchmark. We found that both human participants and agents are particularly susceptible to the dark patterns of Preselection, Trick Wording, and Hidden Information, while being resilient to other overt dark patterns. Our findings inform the development of more trustworthy CUAs, their use as potential human proxies in evaluating deceptive designs, and the regulation of an online environment increasingly navigated by autonomous agents.

cs.HC↗

arXiv2025

AutoScale: Scale-Aware Data Mixing for Pre-Training LLMs

Domain reweighting is an emerging research area aimed at adjusting the relative weights of different data sources to improve the effectiveness and efficiency of LLM pre-training. We show that data mixtures that perform well at smaller scales may not retain their advantage at larger scales, challenging the existing practice of determining competitive mixtures in small-scale experiments and directly applying them at much larger scales. To address this, we propose AutoScale, a two-stage, scale-aware data composition framework. First, AutoScale fits a parametric model that predicts the model's loss under different data compositions, then uses it to find an approximate best allocation at smaller, more manageable budgets. Next, leveraging a novel theoretical analysis of how optimal compositions evolve with scale, AutoScale extrapolates that composition to larger budgets without further retraining. Empirically, AutoScale accelerates convergence and improves downstream performance. For instance, when pre-training GPT-2 Large, it achieves a 28% faster perplexity reduction than baselines and up to a 38% speed-up over unweighted training, while yielding best-average results on various downstream tasks. Overall, our findings illustrate how domain importance shifts with training scale, underscoring the need for scale-dependent data curation in LLM training. Our code is open-sourced.

cs.LG↗

arXiv2025

Tensorized Clustered LoRA Merging for Multi-Task Interference

Despite the success of the monolithic dense paradigm of large language models (LLMs), the LoRA adapters offer an efficient solution by fine-tuning small task-specific modules and merging them with the base model. However, in multi-task settings, merging LoRA adapters trained on heterogeneous sources frequently causes \textit{task interference}, degrading downstream performance. To address this, we propose a tensorized clustered LoRA (TC-LoRA) library targeting to address the task interference at the \textit{text-level} and \textit{parameter-level}. At the \textit{text-level}, we cluster the training samples in the embedding space to capture input-format similarities, then train a specialized LoRA adapter for each cluster. At the \textit{parameter-level}, we introduce a joint Canonical Polyadic (CP) decomposition that disentangles task-specific and shared factors across LoRA adapters. This joint factorization preserves essential knowledge while reducing cross-task interference. Extensive experiments on out-of-domain zero-shot and skill-composition tasks-including reasoning, question answering, and coding. Compared to strong SVD-based baselines, TC-LoRA achieves +1.4\% accuracy on Phi-3 and +2.3\% on Mistral-7B (+2.3\%), demonstrating the effectiveness of TC-LoRA in LLM adaptation.

cs.LG↗

arXiv2025

Do Language Models Mirror Human Confidence? Exploring Psychological Insights to Address Overconfidence in LLMs

Psychology research has shown that humans are poor at estimating their performance on tasks, tending towards underconfidence on easy tasks and overconfidence on difficult tasks. We examine three LLMs, Llama-3-70B-instruct, Claude-3-Sonnet, and GPT-4o, on a range of QA tasks of varying difficulty, and show that models exhibit subtle differences from human patterns of overconfidence: less sensitive to task difficulty, and when prompted to answer based on different personas -- e.g., expert vs layman, or different race, gender, and ages -- the models will respond with stereotypically biased confidence estimations even though their underlying answer accuracy remains the same. Based on these observations, we propose Answer-Free Confidence Estimation (AFCE) to improve confidence calibration and LLM interpretability in these settings. AFCE is a self-assessment method that employs two stages of prompting, first eliciting only confidence scores on questions, then asking separately for the answer. Experiments on the MMLU and GPQA datasets spanning subjects and difficulty show that this separation of tasks significantly reduces overconfidence and delivers more human-like sensitivity to task difficulty.

cs.AI↗

arXiv2025

Escaping the SpuriVerse: Can Large Vision-Language Models Generalize Beyond Seen Spurious Correlations?

Finetuning can cause spurious correlations to arise between non-essential features and the target labels, but benchmarks to study these effects involve contrived settings and narrow tasks. In contrast, we consider spurious correlations in multi-modal Large Vision Language Models (LVLMs) pretrained on extensive and diverse datasets without explicit task supervision. We develop a benchmark by sourcing GPT-4o errors on real-world visual-question-answering (VQA) benchmarks, then curating a subset through LVLM-human annotation and synthetic counterfactual evaluation to identify errors caused by spurious correlations. This process yields SpuriVerse, a novel benchmark comprised of 124 distinct types of spurious correlations extracted from real-world datasets, each containing 1 realistic and 10 synthetic VQA samples for a total of 1364 multiple choice questions. We evaluate 15 open and closed-source LVLMs on SpuriVerse, finding that even state-of-the-art closed-source models struggle significantly, achieving at best only 37.1% accuracy. Fine-tuning on synthetic examples that emphasize the spurious correlation improves performance to 78.40%, suggesting that training on diverse spurious patterns generalizes to unseen situations: models appear to learn to avoid "shortcuts" and attend to the overall image context.

cs.CV↗

arXiv2025

MMMG: a Comprehensive and Reliable Evaluation Suite for Multitask Multimodal Generation

Automatically evaluating multimodal generation presents a significant challenge, as automated metrics often struggle to align reliably with human evaluation, especially for complex tasks that involve multiple modalities. To address this, we present MMMG, a comprehensive and human-aligned benchmark for multimodal generation across 4 modality combinations (image, audio, interleaved text and image, interleaved text and audio), with a focus on tasks that present significant challenges for generation models, while still enabling reliable automatic evaluation through a combination of models and programs. MMMG encompasses 49 tasks (including 29 newly developed ones), each with a carefully designed evaluation pipeline, and 937 instructions to systematically assess reasoning, controllability, and other key capabilities of multimodal generation models. Extensive validation demonstrates that MMMG is highly aligned with human evaluation, achieving an average agreement of 94.3%. Benchmarking results on 24 multimodal generation models reveal that even though the state-of-the-art model, GPT Image, achieves 78.3% accuracy for image generation, it falls short on multimodal reasoning and interleaved generation. Furthermore, results suggest considerable headroom for improvement in audio generation, highlighting an important direction for future research.

cs.AI↗

arXiv2025

Know Your Limits: A Survey of Abstention in Large Language Models

Abstention, the refusal of large language models (LLMs) to provide an answer, is increasingly recognized for its potential to mitigate hallucinations and enhance safety in LLM systems. In this survey, we introduce a framework to examine abstention from three perspectives: the query, the model, and human values. We organize the literature on abstention methods, benchmarks, and evaluation metrics using this framework, and discuss merits and limitations of prior work. We further identify and motivate areas for future research, such as whether abstention can be achieved as a meta-capability that transcends specific tasks or domains, and opportunities to optimize abstention abilities in specific contexts. In doing so, we aim to broaden the scope and impact of abstention methodologies in AI systems.

cs.CL↗

arXiv2024

Characterizing LLM Abstention Behavior in Science QA with Context Perturbations

The correct model response in the face of uncertainty is to abstain from answering a question so as not to mislead the user. In this work, we study the ability of LLMs to abstain from answering context-dependent science questions when provided insufficient or incorrect context. We probe model sensitivity in several settings: removing gold context, replacing gold context with irrelevant context, and providing additional context beyond what is given. In experiments on four QA datasets with six LLMs, we show that performance varies greatly across models, across the type of context provided, and also by question type; in particular, many LLMs seem unable to abstain from answering boolean questions using standard QA prompts. Our analysis also highlights the unexpected impact of abstention performance on QA task accuracy. Counter-intuitively, in some settings, replacing gold context with irrelevant context or adding irrelevant context to gold context can improve abstention performance in a way that results in improvements in task performance. Our results imply that changes are needed in QA dataset design and evaluation to more effectively assess the correctness and downstream impacts of model abstention.

cs.CL↗

arXiv2024

Laboratory-Scale AI: Open-Weight Models are Competitive with ChatGPT Even in Low-Resource Settings

The rapid proliferation of generative AI has raised questions about the competitiveness of lower-parameter, locally tunable, open-weight models relative to high-parameter, API-guarded, closed-weight models in terms of performance, domain adaptation, cost, and generalization. Centering under-resourced yet risk-intolerant settings in government, research, and healthcare, we see for-profit closed-weight models as incompatible with requirements for transparency, privacy, adaptability, and standards of evidence. Yet the performance penalty in using open-weight models, especially in low-data and low-resource settings, is unclear. We assess the feasibility of using smaller, open-weight models to replace GPT-4-Turbo in zero-shot, few-shot, and fine-tuned regimes, assuming access to only a single, low-cost GPU. We assess value-sensitive issues around bias, privacy, and abstention on three additional tasks relevant to those topics. We find that with relatively low effort, very low absolute monetary cost, and relatively little data for fine-tuning, small open-weight models can achieve competitive performance in domain-adapted tasks without sacrificing generality. We then run experiments considering practical issues in bias, privacy, and hallucination risk, finding that open models offer several benefits over closed models. We intend this work as a case study in understanding the opportunity cost of reproducibility and transparency over for-profit state-of-the-art zero shot performance, finding this cost to be marginal under realistic settings.

cs.LG↗

arXiv2023

InfoVisDial: An Informative Visual Dialogue Dataset by Bridging Large Multimodal and Language Models

In this paper, we build a visual dialogue dataset, named InfoVisDial, which provides rich informative answers in each round even with external knowledge related to the visual content. Different from existing datasets where the answer is compact and short, InfoVisDial contains long free-form answers with rich information in each round of dialogue. For effective data collection, the key idea is to bridge the large-scale multimodal model (e.g., GIT) and the language models (e.g., GPT-3). GIT can describe the image content even with scene text, while GPT-3 can generate informative dialogue based on the image description and appropriate prompting techniques. With such automatic pipeline, we can readily generate informative visual dialogue data at scale. Then, we ask human annotators to rate the generated dialogues to filter the low-quality conversations.Human analyses show that InfoVisDial covers informative and diverse dialogue topics: $54.4\%$ of the dialogue rounds are related to image scene texts, and $36.7\%$ require external knowledge. Each round's answer is also long and open-ended: $87.3\%$ of answers are unique with an average length of $8.9$, compared with $27.37\%$ and $2.9$ in VisDial. Last, we propose a strong baseline by adapting the GIT model for the visual dialogue task and fine-tune the model on InfoVisDial. Hopefully, our work can motivate more effort on this direction.

cs.CV↗

arXiv2023

OmniMotionGPT: Animal Motion Generation with Limited Data

Our paper aims to generate diverse and realistic animal motion sequences from textual descriptions, without a large-scale animal text-motion dataset. While the task of text-driven human motion synthesis is already extensively studied and benchmarked, it remains challenging to transfer this success to other skeleton structures with limited data. In this work, we design a model architecture that imitates Generative Pretraining Transformer (GPT), utilizing prior knowledge learned from human data to the animal domain. We jointly train motion autoencoders for both animal and human motions and at the same time optimize through the similarity scores among human motion encoding, animal motion encoding, and text CLIP embedding. Presenting the first solution to this problem, we are able to generate animal motions with high diversity and fidelity, quantitatively and qualitatively outperforming the results of training human motion generation baselines on animal data. Additionally, we introduce AnimalML3D, the first text-animal motion dataset with 1240 animation sequences spanning 36 different animal identities. We hope this dataset would mediate the data scarcity problem in text-driven animal motion generation, providing a new playground for the research community.

cs.CV↗

arXiv2022

EGCR: Explanation Generation for Conversational Recommendation

Growing attention has been paid in Conversational Recommendation System (CRS), which works as a conversation-based and recommendation task-oriented tool to provide items of interest and explore user preference. However, existing work in CRS fails to explicitly show the reasoning logic to users and the whole CRS still remains a black box. Therefore we propose a novel end-to-end framework named Explanation Generation for Conversational Recommendation (EGCR) based on generating explanations for conversational agents to explain why they make the action. EGCR incorporates user reviews to enhance the item representation and increase the informativeness of the whole conversation. To the best of our knowledge, this is the first framework for explainable conversational recommendation on real-world datasets. Moreover, we evaluate EGCR on one benchmark conversational recommendation datasets and achieve better performance on both recommendation accuracy and conversation quality than other state-of-the art models. Finally, extensive experiments demonstrate that generated explanations are not only having high quality and explainability, but also making CRS more trustworthy. We will make our code available to contribute to the CRS community

cs.AI↗