arXiv subjects

Bhuwan Dhingra

Publications and source records attributed to Bhuwan Dhingra.

At least 19 recordsLinked to original sources

arXiv2026

SearchAtlas: Analyzing Agentic Search Strategies via Evidential Query Graphs

LLM search agents are often evaluated on final-answer accuracy, overlooking the process. Analyzing a search strategy requires understanding how credible evidence is retrieved to address question constraints. This valuable information is buried in raw search trajectories that are long and difficult to parse. We introduce SearchAtlas, a framework that converts search trajectories into structured graphs whose edges represent how evidence is propagated across the reasoning trace, from the query that retrieves it to the final answer. Our automated parsing pipeline achieves a mean edge F1 of 86.0% against human-annotated graphs and remains consistent across repeated runs. We analyze five search agents on three benchmarks, revealing systematic differences in search scale and evidence aggregation. SearchAtlas exposes fragmented answer support, question constraints that do not reach the answer, and unverified parametric knowledge entering the response. These process failures are strongly associated with incorrect answers, even more so than an LLM judge given either the raw trajectory or the ordered query list, suggesting that the constructed graphs provide useful interpretability. Moreover, an audit of cases in which process-diagnostic scores disagree with final-answer correctness shows that they capture information not reducible to answer accuracy.

cs.CL↗

arXiv2026

Lazy Grounding: Attacking Search Agents with Factual Evidence

Search agents mitigate hallucination by grounding their answers in retrieved web results. However, retrieval-based approaches also introduce an attack surface: agents may cite misinformation from poisoned search corpora containing false or malicious documents. We demonstrate that, in some cases, search agents' reasoning and responses may be steered by completely factual but distracting information. We refer to this failure as lazy grounding. We expose lazy grounding by injecting nearby evidence from answer-changing rewrites of benchmark questions into the search corpora. Each document contains factual evidence that supports a neighboring rewritten question but is retrieved for the original question. Across 12 model-benchmark pairs, the attack causes the accuracy of search agents' responses to drop by 5.9 points on average and by up to 17.3 points, while inducing nearby-answer adoption in every setting. The effect is even stronger when nearby evidence appears later or is more answer-shaped. Our results show that robust search agents must defend against not only misinformation but also the misapplication of factual evidence. The code is publicly available at https://github.com/frankyzha/lazy-grounding.

cs.CL↗

arXiv2026

PRO-LONG: Programmatic Memory Enables Long-Horizon Reasoning

Long-horizon tasks require sustained perception, reasoning, and exploration, and are a persistent challenge for large language model (LLM) agents. This gap is reflected in their limited performance on continual learning benchmarks such as ARC-AGI-3, especially when models are evaluated out of the box. Various agent harnesses have been proposed to close this gap, and each commits to a strategy for handling long sequences of observations, i.e., what information to save from the environment and how to load it into model context, a choice we argue is particularly consequential. Existing methods for context management face a significant tradeoff, as preserving more information makes retrieving relevant details less tractable. We propose PRO-LONG, a minimal context management framework built around programmatic memory for LLM agents in long-horizon, exploratory settings. PRO-LONG addresses the tradeoff by keeping a complete, structured interaction log and capitalizing on recent progress in coding agents to search this history efficiently. On the full ARC-AGI-3 public game set, PRO-LONG improves over a base coding agent by an average of 18.0 percentage points across frontier models, and matches or exceeds state-of-the-art specialized harnesses (up to 76.1% pass@1) while using 4.2-5.8x fewer tokens. With Fable 5, PRO-LONG achieves 97.4% best@2 at a total cost of \$1,750. Relevant code and logs are available at https://github.com/alexisfox7/PRO-LONG.

cs.AI↗

arXiv2026

Automated Benchmark Auditing for AI Agents and Large Language Models

Modern AI benchmarks operate at a complexity that outpaces traditional verification methods. Tasks authored by domain experts often contain implicit assumptions, incomplete environment specifications, and brittle evaluation logic that human annotation cannot reliably catch. We introduce Auto Benchmark Audit (ABA), an agentic framework that systematically audits individual benchmark tasks, uncovering issues such as hidden environment dependencies, specification gaps, and limited grading logic. We run ABA on a collection of frontier LLM benchmarks and previous NeurIPS publications, totaling 168 benchmarks across nine domains. Across this corpus, ABA identifies critical issues including ambiguous task design, execution environment conflicts, and incorrect ground truths in over 25.7% of the evaluated tasks. The precision of these automated audits is validated by expert review and independent third-party reports such as upstream PRs. Crucially, we demonstrate that these problematic tasks severely distorts capability assessments for agents and LLMs: filtering out these tasks with issues shifts model rankings and increases average performance on SWE-bench Verified and Terminal-Bench 2 by 9.9% and 9.6%, respectively. We release the agentic tool and all task annotations to support the future development of frontier benchmarks.

cs.CL↗

arXiv2026

LDDR: Linear-DPP-Based Dynamic-Resolution Frame Sampling for Video MLLMs

Video understanding in multimodal large language models requires selecting informative frames from long, redundant videos under limited visual-token budgets. Existing methods often rely on uniform sampling, point-wise relevance scoring, chunk-wise selection, or agentic exploration, which either miss global dependencies or introduce substantial overhead. We propose LDDR (Linear DPP-Based Dynamic Resolution), a training-free, plug-and-play, and budget-aware video frame sampling framework. LDDR performs query-aware Determinantal Point Process (DPP) frame selection in a task-conditioned feature space, achieving a 3x runtime speedup over standard DPP baselines. It further introduces a Group DPP importance metric to guide frame retention and dynamic resolution allocation, assigning more tokens to informative, non-redundant frames while downscaling or pruning less useful ones. Across four video benchmarks spanning short-, medium-, and long-range videos, LDDR consistently outperforms the next-best baselines, achieving gains of 2.5 points under budget-constrained settings and 1.6 points in high-budget scenarios. These improvements are consistently observed across multiple MLLM backbones, including both open- and closed-source models. Qualitative analysis confirms that relevant frames are selected and allocated a higher budget, facilitating improved video understanding.

cs.CV↗

arXiv2026

Vision2Code: A Multi-Domain Benchmark for Evaluating Image-to-Code Generation

Image-to-code generation tests whether a vision-language model (VLM) can recover the structure of an image enough to express it as executable code. Existing benchmarks either focus on narrow visual domains, depend on paired executable reference code, or rely on generic rubrics that miss domain-specific reconstruction errors. We introduce Vision2Code, a reference-code-free benchmark and evaluation framework for multi-domain image-to-code generation. Vision2Code contains 2,169 test examples from 15 source datasets that span charts and plots, geometry, graphs, scientific imagery, documents, and 3D spatial scenes. Models generate executable programs, which we render and score against the source image using a VLM rater with dataset-specific rubrics and deterministic guardrails for severe semantic failures. We report render-success diagnostics that separate code execution failures from reconstruction quality. Human validation shows that this evaluation protocol aligns better with human judgments than either a generic visual rubric or embedding-similarity baselines. Across nine open-weight and proprietary models, we find that image-to-code performance is domain-dependent: leading models perform well on regular chart- and graph-like visuals but remain weak on spatial scenes, chemistry, documents, and circuit-style diagrams. Finally, we show that evaluator-filtered model outputs can serve as training data to improve image-to-code capability, with Qwen3.5-9B improving from 1.60 to 1.86 on the benchmark without paired source programs. Vision2Code provides a reproducible testbed for measuring, diagnosing, and improving image-to-code generation. Our code and data are publicly available at https://image2code.github.io/vision2code/.

cs.CV↗

arXiv2026

RVPO: Risk-Sensitive Alignment via Variance Regularization

Current critic-less RLHF methods aggregate multi-objective rewards via an arithmetic mean, leaving them vulnerable to constraint neglect: high-magnitude success in one objective can numerically offset critical failures in others (e.g., safety or formatting), masking low-performing "bottleneck" rewards vital for reliable multi-objective alignment. We propose Reward-Variance Policy Optimization (RVPO), a risk-sensitive framework that penalizes inter-reward variance during advantage aggregation, shifting the objective from "maximize sum" to "maximize consistency." We show via Taylor expansion that a LogSumExp (SoftMin) operator effectively acts as a smooth variance penalty. We evaluate RVPO on rubric-based medical and scientific reasoning with up to 17 concurrent LLM-judged reward signals (Qwen2.5-3B/7B/14B) and on tool-calling with rule-based constraints (Qwen2.5-1.5B/3B). By preventing the model from neglecting difficult constraints to exploit easier objectives, RVPO improves overall scores on HealthBench (0.261 vs. 0.215 for GDPO at 14B, $p < 0.001$) and maintains competitive accuracy on GPQA-Diamond without the late-stage degradation observed in other multi-reward methods, demonstrating that variance regularization mitigates constraint neglect across model scales without sacrificing general capabilities.

cs.LG↗

arXiv2026

LensVLM: Selective Context Expansion for Compressed Visual Representation of Text

Vision Language Models (VLMs) offer the exciting possibility of processing text as rendered images, bypassing the need for tokenizing the text into long token sequences. Since VLM image encoders map fixed-size images to a fixed number of visual tokens, varying rendering resolution provides a fine-grained compression knob. However, accuracy deteriorates quickly as compression increases: characters shrink below the vision encoder's effective resolution, making them indistinguishable. To address this, we propose LensVLM, an inference framework and post-training recipe that enables VLMs to scan compressed images, then selectively expand only the relevant images to their uncompressed form via learned tools. Building on Qwen3.5-9B-Base, LensVLM maintains accuracy comparable to the full-text upper bound at 4.3x effective compression and outperforms retrieval-based, text- and visual-compression baselines up to 10.1x effective compression across seven text QA benchmarks. LensVLM also generalizes to multimodal document and code understanding tasks, with the accuracy gain over baselines growing as compression increases. Our analysis validates this approach: training makes visual compression robust to rendering choices, and as compression grows the model increasingly relies on expanded content rather than unreliable visual reading. The analysis also yields practical tool-choice guidance: text expansion is preferable for rendered text, while high-resolution image expansion suits native documents whose layout cues carry task-relevant information.

cs.CV↗

arXiv2026

Document-as-Image Representations Fall Short for Scientific Retrieval

Many recent document embedding models are trained on document-as-image representations, embedding rendered pages as images rather than the underlying source. Meanwhile, existing benchmarks for scientific document retrieval, such as ArXivQA and ViDoRe, treat documents as images of pages, implicitly favoring such representations. In this work, we argue that this paradigm is not well-suited for text-rich multimodal scientific documents, where critical evidence is distributed across structured sources, including text, tables, and figures. To study this setting, we introduce ArXivDoc, a new benchmark constructed from the underlying LaTeX sources of scientific papers. Unlike PDF or image-based representations, LaTeX provides direct access to structured elements (e.g., sections, tables, figures, equations), enabling controlled query construction grounded in specific evidence types. We systematically compare text-only, image-based, and multimodal representations across both single-vector and multi-vector retrieval models. Our results show that: (1) document-as-image representations are consistently suboptimal, especially as document length increases; (2) text-based representations are most effective, even for figure-based queries, by leveraging captions and surrounding context; and (3) interleaved text+image representations outperform document-as-image approaches without requiring specialized training.

cs.IR↗

arXiv2026

Coding Agents are Effective Long-Context Processors

Large Language Models (LLMs) have demonstrated remarkable progress in scaling to access massive contexts. However, the access is via the latent and uninterpretable attention mechanisms, and LLMs fail to effective process long context, exhibiting significant performance degradation as context length increases. In this work, we study whether long-context processing can be externalized from latent attention into explicit, executable interactions, by allowing coding agents to organize text in file systems and manipulate it using its native tools. We evaluate off-the-shelf frontier coding agents as the general interface for tasks that require processing long contexts, including long-context reasoning, retrieval-augmented generation, and open-domain question answering with large-scale corpus contains up to three trillion tokens. Across multiple benchmarks, these agents outperform published state-of-the-art by 17.3% on average. We attribute this efficacy to two key factors: native tool proficiency, which enables agents to leverage executable code and terminal commands rather than passive semantic queries, and file system familiarity, which allows them to navigate massive text corpora as directory structures. These findings suggest that delegating long-context processing to coding agents offers an effective alternative to semantic search or context window scaling, opening new directions for long-context processing in LLMs.

cs.CL↗

arXiv2026

DeepFact: Co-Evolving Benchmarks and Agents for Deep Research Factuality

Search-augmented LLM agents can produce deep research reports (DRRs), but verifying claim-level factuality remains challenging. Existing fact-checkers are primarily designed for general-domain, factoid-style atomic claims, and there is no benchmark to test whether such verifiers transfer to DRRs. Yet building such a benchmark is itself difficult. We first show that static expert-labeled benchmarks are brittle in this setting: in a controlled study with PhD-level specialists, unassisted experts achieve only 60.8% accuracy on a hidden micro-gold set of verifiable claims. We propose Evolving Benchmarking via Audit-then-Score (AtS), where benchmark labels and rationales are explicitly revisable: when a verifier disagrees with the current benchmark, it must submit evidence; an auditor adjudicates the dispute; and accepted revisions update the benchmark before models are scored. Across four AtS rounds, expert micro-gold accuracy rises to 90.9%, indicating experts are substantially more reliable as auditors than as one-shot labelers. We instantiate AtS as DeepFact-Bench, a versioned DRR factuality benchmark with auditable rationales, and DeepFact-Eval, a document-level verification agent (with a grouped lite variant) that outperforms existing verifiers on DeepFact-Bench and transfers well to external factuality datasets.

cs.AI↗

arXiv2026

Over-Searching in Search-Augmented Large Language Models

Search-augmented large language models (LLMs) excel at knowledge-intensive tasks by integrating external retrieval. However, they often over-search -- unnecessarily invoking search tool even when it does not improve response quality, which leads to computational inefficiency and hallucinations by incorporating irrelevant context. In this work, we conduct a systematic evaluation of over-searching across multiple dimensions, including query types, model categories, retrieval conditions, and multi-turn conversations. Our finding shows: (i) search generally improves answer accuracy on answerable queries but harms abstention on unanswerable ones; (ii) over-searching is more pronounced in complex reasoning models and deep research systems, is exacerbated by noisy retrieval, and compounds across turns in multi-turn conversations; and (iii) the composition of retrieved evidence is crucial, as the presence of negative evidence improves abstention. To quantify over-searching, we introduce Tokens Per Correctness (TPC), an evaluation metric that captures the performance-cost trade-off for search-augmented LLMs. Lastly, we investigate mitigation approaches at both the query and retrieval levels and release the OverSearchQA to foster continued research into efficient search-augmented LLMs.

cs.LG↗

arXiv2025

Text-Guided Semantic Image Encoder

Image encoders, a fundamental component of vision-language models (VLMs), are typically pretrained independently before being aligned with a language model. This standard paradigm results in encoders that process images agnostically, without regard to the specific downstream task or text query. To address this limitation, we propose the Text-Guided Semantic Image Encoder (TIE), which generates image representations conditioned on the input text query. VLMs equipped with TIE outperform their conventional counterparts by +1.5 and +1.3 points on average across nine image-to-text benchmarks at the 1B and 3B scales, respectively, with gains reaching up to 6 points on tasks such as DocVQA and InfoVQA. Moreover, TIE-based VLMs attain superior performance while utilizing only half as many image tiles (tokens), resulting in notably improved inference efficiency. TIE also generalizes well with generic queries, indicating that text-conditioned training effectively optimizes the encoder to capture key visual features. Qualitative analysis confirms that TIE consistently attends to query-relevant regions, enhancing both interpretability and query-specific grounding.

cs.CV↗

arXiv2025

InData: Towards Secure Multi-Step, Tool-Based Data Analysis

Large language model agents for data analysis typically generate and execute code directly on databases. However, when applied to sensitive data, this approach poses significant security risks. To address this issue, we propose a security-motivated alternative: restrict LLMs from direct code generation and data access, and require them to interact with data exclusively through a predefined set of secure, verified tools. Although recent tool-use benchmarks exist, they primarily target tool selection and simple execution rather than the compositional, multi-step reasoning needed for complex data analysis. To reduce this gap, we introduce Indirect Data Engagement (InData), a dataset designed to assess LLMs' multi-step tool-based reasoning ability. InData includes data analysis questions at three difficulty levels--Easy, Medium, and Hard--capturing increasing reasoning complexity. We benchmark 15 open-source LLMs on InData and find that while large models (e.g., gpt-oss-120b) achieve high accuracy on Easy tasks (97.3%), performance drops sharply on Hard tasks (69.6%). These results show that current LLMs still lack robust multi-step tool-based reasoning ability. With InData, we take a step toward enabling the development and evaluation of LLMs with stronger multi-step tool-use capabilities. We will publicly release the dataset and code.

cs.CL↗

arXiv2025

Staircase Streaming for Low-Latency Multi-Agent Inference

Recent advances in large language models (LLMs) opened up new directions for leveraging the collective expertise of multiple LLMs. These methods, such as Mixture-of-Agents, typically employ additional inference steps to generate intermediate outputs, which are then used to produce the final response. While multi-agent inference can enhance response quality, it can significantly increase the time to first token (TTFT), posing a challenge for latency-sensitive applications and hurting user experience. To address this issue, we propose staircase streaming for low-latency multi-agent inference. Instead of waiting for the complete intermediate outputs from previous steps, we begin generating the final response as soon as we receive partial outputs from these steps. Experimental results demonstrate that staircase streaming reduces TTFT by up to 93% while maintaining response quality.

cs.AI↗

arXiv2025

When Greedy Wins: Emergent Exploitation Bias in Meta-Bandit LLM Training

While Large Language Models (LLMs) hold promise to become autonomous agents, they often explore suboptimally in sequential decision-making. Recent work has sought to enhance this capability via supervised fine-tuning (SFT) or reinforcement learning (RL), improving regret on the classic multi-armed bandit task. However, it remains unclear how these learning methods shape exploration strategies and how well they generalize. We investigate both paradigms by training LLMs with SFT on expert trajectories and RL with a range of tailored reward signals including a strategic, regret-shaped reward to reduce variance, and an algorithmic reward that enables oracle imitation. The resulting agents outperform pre-trained models and achieve performance comparable to Upper Confidence Bound (UCB) and Thompson Sampling, with robust generalization to 6x longer horizons and across bandit families. Behavioral analysis reveals that gains often stem from more sophisticated but greedier exploitation: RL/SFT agents are more prone to early catastrophic failure than pre-trained models, prematurely abandoning exploration. Furthermore, agents trained to imitate UCB learn to outperform their teacher by adopting more exploitative variants. Our findings clarify when each training paradigm is preferable and advocate tailored reward design and evaluation beyond average regret to promote robust exploratory behavior.

cs.LG↗

arXiv2025

Generalizability of Large Language Model-Based Agents: A Comprehensive Survey

Large Language Model (LLM)-based agents have emerged as a new paradigm that extends LLMs' capabilities beyond text generation to dynamic interaction with external environments. By integrating reasoning with perception, memory, and tool use, agents are increasingly deployed in diverse domains like web navigation and household robotics. A critical challenge, however, lies in ensuring agent generalizability - the ability to maintain consistent performance across varied instructions, tasks, environments, and domains, especially those beyond agents' fine-tuning data. Despite growing interest, the concept of generalizability in LLM-based agents remains underdefined, and systematic approaches to measure and improve it are lacking. In this survey, we provide the first comprehensive review of generalizability in LLM-based agents. We begin by emphasizing agent generalizability's importance by appealing to stakeholders and clarifying the boundaries of agent generalizability by situating it within a hierarchical domain-task ontology. We then review datasets, evaluation dimensions, and metrics, highlighting their limitations. Next, we categorize methods for improving generalizability into three groups: methods for the backbone LLM, for agent components, and for their interactions. Moreover, we introduce the distinction between generalizable frameworks and generalizable agents and outline how generalizable frameworks can be translated into agent-level generalizability. Finally, we identify critical challenges and future directions, including developing standardized frameworks, variance- and cost-based metrics, and approaches that integrate methodological innovations with architecture-level designs. By synthesizing progress and highlighting opportunities, this survey aims to establish a foundation for principled research on building LLM-based agents that generalize reliably across diverse applications.

cs.AI↗

arXiv2025

Cite Pretrain: Retrieval-Free Knowledge Attribution for Large Language Models

Trustworthy language models should provide both correct and verifiable answers. However, citations generated directly by standalone LLMs are often unreliable. As a result, current systems insert citations by querying an external retriever at inference time, introducing latency, infrastructure dependence, and vulnerability to retrieval noise. We explore whether LLMs can be made to reliably attribute to the documents seen during continual pretraining without test-time retrieval, by revising the training process. To study this, we construct CitePretrainBench, a benchmark that mixes real-world corpora (Wikipedia, Common Crawl, arXiv) with novel documents and probes both short-form (single-fact) and long-form (multi-fact) citation tasks. Our approach follows a two-stage process: (1) continual pretraining to index factual knowledge by binding it to persistent document identifiers; and (2) instruction tuning to elicit citation behavior. We introduce Active Indexing for the first stage, which creates generalizable, source-anchored bindings by augmenting training with synthetic data that (i) restate each fact in diverse, compositional forms and (ii) enforce bidirectional training (source-to-fact and fact-to-source). This equips the model to both generate content from a cited source and attribute its own answers, improving robustness to paraphrase and composition. Experiments with Qwen-2.5-7B&3B show that Active Indexing consistently outperforms a Passive Indexing baseline, which simply appends an identifier to each document, achieving citation precision gains of up to 30.2% across all tasks and models. Our ablation studies reveal that performance continues to improve as we scale the amount of augmented data, showing a clear upward trend even at 16x the original token count. Finally, we show that internal citations complement external ones by making the model more robust to retrieval noise.

cs.AI↗