Answer

Follow this topic Ask about this topic

Momentum

24 papers in the last four weeks, up 50% on the four weeks before. 0.3% of all new papers.

Jul 6Week of Sep 21

Latest papers 181

Oct 1, 2026cs.CL

Right Answers, Wrong States: Hidden Information Failures in Multi-Agent Collaboration

Multi-agent systems are often judged by whether they reach the correct answer. This can miss a distinct failure: collaboration may leave behind a corrupted information state even when the immediate decision is correct. We call this an off-query failure. To study this failure in collaborative decision support, we introduce OffQuery, which separately evaluates evidence verification (T1), shared-state reconstruction (T2), and task resolution (T3) in two representative high-stakes settings: healthcare and disaster response. Across GPT, Gemini, and Qwen models, standard collaboration shows much stronger task performance than state reliability. Averaged over 21 model--setting combinations, task resolution reaches 64.7%, while evidence verification and state reconstruction reach only 14.3% and 43.1%. We trace this gap to selective information use: current queries often bypass corrupted facts, which become consequential when later tasks require them. We further introduce ReGround, which resolves conflicting evidence, verifies shared facts, reconstructs a trusted state, and reasons over that state. Across seven models from three families, ReGround improves all three capabilities in every evaluated setting, with average relative gains of 309.0%, 82.9%, and 17.6% on T1, T2, and T3. Reliable collaboration therefore requires both a correct decision and a reliable shared state for future reasoning.

Herun Wan, Jiaying Wu, Minnan Luo +4

Xi’an Jiaotong University · National University of Singapore · Yunnan University +1

Collaboration Multi-Agent Reasoning Software Engineering Decision Support Answer Motion Reconstruction Branch

Oct 1, 2026cs.AI

Beyond Final Accuracy: Auditing Communication in LLM Multi-Agent Systems

Multi-agent communication aims to help agents benefit from one another's information. Yet improvements in system performance leave a fundamental ambiguity: do they reflect effective communication, a favorable agent architecture, or simply additional reasoning? Because communication methods are commonly evaluated within the systems they were designed for, these factors are difficult to disentangle. Final accuracy further merges corrected errors and corrupted answers into a single outcome, obscuring how communication changes decisions. We introduce Independent--Communicate--Revise (ICR), a controlled framework that evaluates communication as answer revision following independent reasoning. ICR fixes initial reasoning trajectories, measures correction and preservation conditional on both agents' initial correctness, and uses a no-message revision control to quantify gains beyond additional reasoning. Across four reasoning benchmarks, our audit of textual and latent communication reveals that similar aggregate accuracy can conceal substantially different revision behaviors. Compared with transmitting answers alone, full reasoning increases correction while reducing preservation on all four benchmarks, so richer messages amplify beneficial and harmful influence alike. Receiver-policy comparisons on MedQA and GPQA-D further show that a structured verification policy shifts every channel toward greater preservation and lower correction, while its effect on selectivity varies across channels and tasks. These findings challenge treating communication quality as an intrinsic property of a channel. ICR therefore recenters evaluation on selective revision, providing a unified framework for examining how message content and receiver policies jointly produce benefits and harms.

Shixuan Li, Wei Yang, Peiyu Zhang +3

Ming Hsieh Department of Electrical and Computer Engineering University of Southern California, Los Angeles, CA 90089, USA · Thomas Lord Department of Computer Science University of Southern California, Los Angeles, CA 90089, USA

Multi-Agent Communication Model Auditing Answer Reasoning Skills

Oct 1, 2026cs.CV

Two Clocks in Diffusion MLLMs: When Answers Stabilize Before Rationales Unfold

An answer candidate in a masked diffusion MLLM can stabilize while its rationale is still unfolding. We distinguish retrospective stabilization of the logged candidate from token commitment, and examine these two clocks relative to rationale generation. Analyzing our results across three visual question-answering benchmarks, we find that 89.4-98.1% of the rationale-side canvas remains unwritten at stabilization in single-block, EOS-suppressed LaViDa runs. On VBench, reducing block length from 128 to 8 changes this fraction from 89.4% to 1.7%, together with answer coverage and the eligible observation window. Under EOS-enabled prompting, direct instructions improve Nemotron's overall accuracy by 15.0 and 19.5 percentage points on M3CoT and ScienceQA, but reduce LaViDa/VBench accuracy by 11.0 points. A symmetric decomposition associates the larger absolute component of each change with coverage rather than conditional accuracy. Matched-canvas image ablations measure visual sensitivity alongside answer stabilization, separating the two temporal readouts. Together, these measurements distinguish answer stabilization, rationale unfolding, and visual sensitivity, and identify coverage as the larger component of the prompting differences.

Keuntae Kim, Yong Suk Choi

Department of Computer Science, Hanyang University, Seoul, Korea

Visual Question Answering Benchmarks Answer Diffusion Language Models Clocks

Sep 30, 2026cs.CL

LLM-as-a-Judge for Low-Resource Languages: Adapting Ragas and Comparative Ranking for Romanian

Evaluating Retrieval-Augmented Generation (RAG) systems remains a challenge for Low-Resource Languages (LRLs), where standard reference-based metrics fall short. This paper investigates the viability of the "LLM-as-a-Judge" paradigm for Romanian by adapting the Ragas framework using next-generation models (Gemini 2.5 and Gemini 3). We introduce AdminRo-Eval, a curated dataset of Romanian administrative documents annotated by native speakers, to serve as a ground truth for benchmarking automated evaluators. We compare three evaluation methodologies - direct scoring, comparative ranking, and granular decomposition - across metrics for Faithfulness, Answer Relevance, and Context Relevance. Our findings reveal that evaluation strategies must be metric-specific: granular decomposition achieves the highest human alignment for Faithfulness (96% with Gemini 2.5 Pro), while comparative ranking outperforms in Answer Relevance (90%). Furthermore, we demonstrate that while lightweight models struggle with complex reasoning in LRLs, the Gemini 2.5 Pro architecture establishes a robust, transferable baseline for automated Romanian RAG evaluation.

Claudiu Creanga, Liviu P. Dinu

Interdisciplinary School of Doctoral Studies, University of Bucharest, Romania · HLT Research Center, University of Bucharest, Romania · Faculty of Mathematics and Computer Science, University of Bucharest, Romania

Low-Resource Languages Ranking Answer

Sep 30, 2026cs.LG

When the Right Answer Is Missing: An Arithmetic-Dependent Rejection Bottleneck in Jev

Typed decision models such as Jev offer an efficient alternative to generative LLMs in decision-making workflows by selecting directly from predefined options. When candidate sets contain no valid answer, TypeSafe recommends including an "other" or "none-of-the-above" option to enable rejection. In this report, however, we identify an arithmetic-dependent rejection bottleneck: Jev reliably selects correct numerical answers when available but frequently accepts incorrect alternatives when they are absent despite an explicit rejection option. On paired arithmetic problems, answer-present accuracy reaches 99%, while correct rejection falls to 7%. Moreover, this gap persists across numerical magnitudes, operation depths, contextual formulations, and rejection labels, and extends to scenarios such as time calculation and capacity rounding. Yet native Boolean verification achieves 99% exact-match accuracy on the same answer-absent arithmetic cases, showing that categorical rejection can fail even when the model successfully verifies candidate correctness. Finally, we show that a simple decision threshold selected on separate development problems raises arithmetic rejection accuracy from 7% to 79% while retaining 97% answer-present accuracy, substantially mitigating the failure without retraining or additional inference.

Jike Zhong, Ming Li, Yuxiang Lai

University of Southern California · University of Florida · Emory University

Rejection Answer Self-Verification Bottlenecks

Sep 30, 2026cs.LG

Right Answer, Wrong Mechanism: Detecting Pernicious Divergence in Causal Interventions

Causal interventions such as activation patching and distributed alignment search (DAS) are the main tool for making mechanistic claims about neural networks. Recent work showed that these interventions routinely push representations off the model's natural distribution, and that such divergence is sometimes harmless and sometimes pernicious: it can recruit pathways the model never uses on natural inputs, so that an intervention produces the expected answer through the wrong mechanism. No method currently tells the two cases apart. We make this question testable by planting hidden pathways inside pretrained language models; the pathways are silent on every benchmark prompt by construction, so which interventions depend on them is known exactly. Across 72 configurations and 100,800 interventions on GPT-2 small, we find three things. (i) Nearest-neighbour and local-PCA distances at the intervention site, as used in prior work, score below chance (AUROC 0.35-0.47) at picking out interventions that give the right answer through a planted pathway. (ii) Hidden-Pathway Contribution (HPC), a label-free test that clamps downstream units to the regime of natural runs with the same output and measures how much of the decision disappears, flags pathway-dominated interventions with AUROC >= 0.99 when the pathway shows up as unit-level out-of-regime activity, but fails when every unit stays within its natural range, which we identify as the open problem. (iii) Optimised interventions actively seek hidden pathways: on a gender task, DAS routes 90-95% of its successes through planted pathways for three of four families, and a downstream on-manifold penalty cuts this share to under 5% at a cost of 6-11 points of success rate. In unmodified GPT-2, successful interventions show almost no unit-level out-of-regime reliance.

Beiming Liu, Minjie Chen

PetroChina Southwest Oil and Gasfield Company, China

Causal Intervention Divergence Correct Answer

Sep 30, 2026cs.CV

Aligning Thoughts with Answers: Probability Rewards to Tame Thinking Drift

This paper studies \textbf{thinking--answer consistency} in vision-language models. We focus on Visual Intention Grounding, where a model infers a target object based on a human intention query and predicts a bounding box. We reveal that previous IoU-based reinforcement learning (RL) frameworks suffer from ``thinking drift'', where the model produces a correct bounding box, despite having an incorrect reasoning process pointing to a different target object. Thus, we propose \textbf{Rita} (\textit{ReInforcing Thinking--Answer consistency}) as a novel RL paradigm to tame the drift. Specifically, Rita introduces two reasoning-label-free RL rewards, constructed from the conditional probability of reference answers: a \textbf{thinking reward} and a \textbf{consistency reward}. It also adopts a difficulty-aware \textbf{data filtering} strategy that selects informative easy-to-medium samples for RL using rollout error rate and reward variance. Extensive experiments on EgoIntention and the new RefEgo-Int benchmarks show that Rita performs consistently superior to the supervised finetuning approaches and vanilla RL-finetuned frameworks.

Pengzhan Sun, Shiu-hong Kao, Shijie Li +4

National University of Singapore · A*STAR Institute of Advanced Intelligence and Computing, Singapore · South China University of Technology +2

Weak Visual Grounding Offline Reinforcement Learning Answer

Sep 30, 2026cs.LG

Right Answers, Costly Models: The Efficiency Gap in LLM-based Optimization Modeling

Optimization modeling formulates real-world decision problems as mathematical programs that solvers can use to find optimal decisions. Large language models (LLMs) can automate this process, but the resulting correct formulations can require substantial time and memory to construct and solve, limiting practical scalability. Therefore, we systematically investigate whether LLMs can identify problem structure from natural-language descriptions and apply suitable optimization modeling techniques to generate mathematical models and solver code that solve the problems correctly and efficiently. To this end, we first curate OptTips, a knowledge base of 50 expert modeling techniques in eight families. Using this knowledge, we develop OptDachshund, a multi-agent framework that transforms problems from existing optimization benchmarks into new tasks for evaluating LLMs' use of modeling techniques. It constructs conventional and expert mathematical models with solver code for the same task and data, providing baselines for correctness and computational cost. The resulting EfficientOpt benchmark contains 561 expert-reviewed tasks with paired reference implementations. Evaluation of 11 representative LLMs reveals an efficiency gap on correctly solved tasks with comparable measurements: for every LLM, most generated programs take longer to solve than their expert counterparts. Within the comparable reference-size subset, 57% of programs with correct objective values and fewer variables and linear constraints have longer recorded solver times. Case studies show that different modeling techniques can achieve the same optimal value at similar recorded cost. Faster solving may not reduce execution time if the code takes longer to prepare data and build the model. LLM optimization modeling should therefore be evaluated for both correctness and computational efficiency.

Zhong Li, Xin Huang, Jinhui Wan +6

Great Bay University · Beijing Jiaotong University · Beihang University +1

Optimization Modeling Large Language Model Benchmarks Expert Demonstrations Answer

Sep 30, 2026cs.AI

Budget Boundary Effects in Test-Time Mathematical Reasoning

A cumulative token cap can fall inside a mathematical derivation, forcing a test-time controller to choose between stopping at the cap (strict) and allowing the current attempt to finish (advisory). We measure this boundary choice with paired offline replays of 19,200 public traces: 120 AIME, BrUMO and HMMT problems and two archive configurations of one model. Candidate order and a 16-attempt cap are fixed, and answer selection is blind to reference answers and correctness labels. Three findings emerge. First, at the 4k cap, most advisory accuracy gains replace abstention with a correct answer; strict stopping pays for an unfinished prefix that the completed-only selector cannot use. Second, comparisons along realized cost differ from same-cap comparisons: advisory 4k in low has higher accuracy than strict 8k at comparable mean completion cost, while in high its observed accuracy is 0.42 points below strict 32k using 59% of its mean tokens. These aggregate comparisons do not establish equal-compute superiority or accuracy equivalence. Third, increased candidate coverage does not guarantee higher answer accuracy: a log-probability selector loses accuracy while coverage rises, including after a source-grade consistency repair. Same-cap majority-accuracy differences shrink below 1.3 percentage points at 32k. Budget curves should jointly state the cap, realized cost, eligible candidates, stopping rule and selector information.

Guilin Zhang, Ziqi Tan, Wulan Guo +5

Yantong AI

Mathematical Reasoning Test Time Benchmark Saturation Answer Majority Voting

Sep 29, 2026cs.CV

Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering

Reasoning about the 3D world from multi-view images remains a fundamental challenge for Multimodal Large Language Models (MLLMs). While modern MLLMs handle single-image inputs effectively, they struggle to integrate evidence across viewpoints into a coherent 3D understanding. A growing body of work attempts to close this gap by injecting 3D awareness into MLLMs, either by boosting fine-grained pixel-level cross-view correspondence or by fusing features from 3D geometry foundation models, yet a substantial gap to human reasoning persists. In this work, we revisit human spatial reasoning, which suggests that rather than relying on fine-grained geometry cues, humans roughly identify common objects across views, infer the relative geometry between viewpoints, and assemble a coarse 3D layout of the scene. Inspired by this process, we introduce Imagine3D-LLM, an MLLM that learns to assemble a similar compact 3D representation of the scene and conditions its answer on this representation. Concretely, we append a small set of learnable summary tokens after the image tokens, decode them into a compact 3D Gaussian Splatting representation supervised by a photometric reconstruction loss, and train jointly with the standard next-token prediction objective. Notably, although only the summary tokens receive direct reconstruction supervision, this objective also induces stronger cross-frame correspondence within the LLM's underlying image features, suggesting that learning to reconstruct propagates 3D-aware signals throughout the model. As a result, Imagine3D-LLM consistently outperforms prior approaches across multiple spatial reasoning and 3D understanding benchmarks, suggesting that imagining the scene can be more effective than being told its pixel-wise geometry.

Jaewoo Jung, Hyeonseo Yu, Honggyu An +10

KAIST AI · ETH Zürich · Google +2

Spatial Reasoning Multimodal Large Language Models Geometric Cues Answer

Sep 29, 2026cs.CL

Correct Answers, Invalid Traces: What Verifiable Grade-School Math Reveals About Chain-of-Thought Traces

Chain-of-thought traces are widely read as records of how models reach their answers, informing debugging, agent auditing, and claims about reasoning. Testing this interpretation is difficult because natural-language thinking traces are rarely mechanically verifiable. We revisit it in iGSM, a synthetic grade-school mathematics benchmark designed to study thinking traces and used to support claims of learned reasoning and planning. Crucially, iGSM exposes the exact quantities and dependencies that a correct solution should use, allowing generated traces to be checked programmatically step by step and enabling us to test whether correct answers are reliably accompanied by valid traces. We first evaluate models trained exclusively on valid, minimal traces. Answer correctness and trace validity nearly coincide in distribution but decouple out of distribution: on the hardest instances, 31.6% of correct answers have invalid traces, over half of which pass all syntactic and arithmetic checks but fail semantic dependency checks. We then intervene on trace supervision. Non-minimal training traces induce non-minimal outputs, while re-asking the same problem with a different query reveals computations inherited from the original query, weakening minimality as evidence of selective planning. Shuffling tokens in 10% of training trace sentences preserves near-clean accuracy even out of distribution despite no trace passing verification. Swapped training traces likewise retain high in-distribution accuracy. We discuss the implications of these findings for chain-of-thought monitoring and interpretation in the context of AI safety.

Ratish Puduppully, Pranabendu Misra, Paarth Iyer +3

IT University of Copenhagen · Chennai Mathematical Institute · Indian Institute of Technology Jammu +1

Reasoning Traces Chain-of-Thought Reasoning Mathematical Reasoning Benchmarks Answer

Sep 29, 2026cs.AI

Locating Answer-Correctness Signals in Frozen Large Language Models

Language models expose internal signals that predict whether an answer is correct, readable from a single forward pass of a frozen model without additional generations. Yet existing probes often commit to one signal family or layer and can be brittle under distribution shift; in retrieval-augmented settings, many specialized detectors instead target passage faithfulness, which can diverge from correctness when retrieved evidence is unhelpful or conflicting. We therefore ask where answer correctness is readable, which internal signal families carry it, and how they should be combined. We search over hidden states, token probabilities, residual-stream features, attention, and their fusion, treating the selected readouts as a predictive measurement rather than a mechanistic localization. We run this analysis separately in closed-book and with-context settings, since context can change which readouts are informative. A consistent anatomy emerges: correctness concentrates in the answer span, recovered from the answer tokens even under retrieval, and the families carry it complementarily, so fusing them helps most out of distribution, where a single signal is weakest. The protocol is effective across two backbones and gates a retrieval controller as one downstream use.

Yuansen Liu, Yixuan Tang, Anthony Kum Hoe Tung

School of Computing, National University of Singapore

Frozen Language Model Correctness Answer Private Signals

Sep 29, 2026cs.CV

TAEC: Trajectory-Aware Evidence Coordination for Multi-Step Visual RAG

Multi-step visual retrieval-augmented generation (RAG) answers complex questions by repeatedly retrieving visual evidence, updating an intermediate state, and deciding whether to continue searching or answer. Yet retrieving relevant evidence does not ensure its effective use throughout the reasoning trajectory. As multi-step reasoning progresses, redundant sources occupy context capacity needed for missing evidence, observations tied to resolved requirements or unproductive searches linger in context, and visual sources are revisited with insufficient detail for fine-grained reading. We term this loss of usable evidence over a reasoning trajectory trajectory-level evidence utilization degradation. To address it, we propose Trajectory-Aware Evidence Coordination (TAEC), a training-free framework that coordinates evidence use around unresolved answer requirements. TAEC tracks these requirements in a shared trajectory state to guide which evidence enters the context, how accumulated memory is retained, and at what level of detail visual evidence is examined. Under a unified evaluation protocol on ViDoSeek, SlideVQA, and MMLongBench-Doc, TAEC achieves the best overall performance against leading training-free visual RAG baselines, with the highest average accuracy across multiple proprietary vision-language models. These results demonstrate that aligning evidence with evolving reasoning needs improves evidence use throughout multi-step visual RAG.

Yalun Wu, Bingzhou Wang, Boyang Wang +5

NExT++ Lab, National University of Singapore

Visual Evidence Multimodal Retrieval Augmented Generation Knowledge-Based Visual Question Answering Reasoning Trajectory Trajectory-Aware Hidden-State Analyses Answer

Sep 29, 2026cs.AI

When Upstream Messages Override Correct Answers: A Controlled Study of Multi-Agent LLM Collaboration

Multi-agent LLM systems rely on message passing among specialized agents to accomplish complex tasks. However, an upstream agent may provide useful information or an incorrect answer that causes a downstream agent to override a correct answer supported by its own evidence. Prior work has not clearly separated the benefits of communication from the damage caused by incorrect messages. We study this problem with controlled experiments across five benchmarks and five receivers, keeping the downstream task and evidence fixed while comparing answers under three conditions: no message, the upstream agent's original message, or a message with the opposite conclusion. Our experiments reveal three key findings. First, messages often help when the downstream agent would otherwise answer incorrectly. Second, messages can also hurt: when the downstream agent would answer correctly without a message, an incorrect upstream message changes the answer in up to 32% of cases. Third, in 94% of audited harmful cases, the downstream agent copies the upstream's specific wrong answer--a pattern we term answer substitution. Removing unreliable messages recovers part of the lost accuracy, suggesting that communication should be selective based on upstream reliability and the evidence already available to the downstream agent.

Yaxin Gong, Gangyi Zhang, Chongming Gao +7

University of Science and Technology of China · Qwen Business Unit of Alibaba · National University of Singapore

Multi-Agent Large Language Model Systems Downstream Reasoning Collaboration Answer Substitute Controlled Study

Sep 28, 2026cs.AI

Towards Mitigating Deceptive Safety Alignment in Large Reasoning Models

Large Reasoning Models (LRMs) are commonly trained with reinforcement learning (RL) to improve their generation of chain-of-thought (CoT) reasoning before producing final answers. However, RL rewards are typically assigned based on final answers, providing little or no direct supervision over intermediate reasoning. This can lead to deceptive safety alignment, where the reasoning trace and final answer convey inconsistent safety signals. To systematically investigate this phenomenon, we introduce DSAR (Deceptive Safety Alignment Rate), a metric that jointly assesses reasoning traces and final answers to quantify their safety inconsistency. Across multiple LRMs and benchmarks, we find that deceptive safety alignment is pervasive under standard prompting conditions and is substantially amplified under prefilling attacks. We further provide a hidden representation analysis showing that models exhibit stronger safety discrimination at the final-answer stage than during intermediate reasoning. To close this gap, we propose SARA (Safety-Aware Reasoning Alignment), an RL-based method that rewards both safety-aware reasoning and safe final answers, encouraging early harmful intent recognition and enforcing reasoning-answer consistency. Experiments show that SARA significantly mitigates deceptive safety alignment under both standard and adversarial settings while preserving helpfulness and utility. Code is available at https://github.com/xzhou98/SARA.

Xiangyu Zhou, Saleh Zare Zade, Rafi Ibn Sultan +2

Wayne State University

Safety Alignment Large Reasoning Models Large Language Model Alignment Answer

Sep 28, 2026cs.CV

AnswerMap: Faithful Spatial Interpretability of VLMs from Answer Posteriors

When a VLM answers a visual query, current interpretability tools rely on text rationales, which use a mismatched modality, or on internal read-outs, which originate too early to reflect the final output and require white-box access to the model. We introduce AnswerMap, a training-free, task-agnostic, black-box visual rationale constructed from the output head. The image is cut into K row and K column bands, each shown alone to the frozen model along with the query in the format of a yes/no relevance question. The outer product of the row and column ``yes'' posteriors gives the query-conditioned spatial map. Crucially, by defining a fixed read-out R (e.g., expectation, maximum) on top of AnswerMap, we can derive continuous outputs like location natively. This bypasses the reliance on discrete text tokens for continuous-output tasks and guarantees an image-dependent answer by construction. However, a rationale can be confabulated, so we validate AnswerMap across four models and three query distributions with two tests: (a) agreement with the model's own generated point and (b) deletion of the map's region. The map lands where the model points (AUC 0.85 against 0.38 for attention), and deleting its region flips 53% of correct answers (against 19% for attention's). Beyond establishing faithfulness, we demonstrate the map's task-agnostic utility through three distinct read-outs: its maximum flags hallucinated objects without generation, its expectation localizes correctly when the model's own pointing fails, and its top-mass region, fed back as a crop, fixes half of the model's wrong answers. AnswerMap thus offers a new lens on VLM interpretability and, through its read-outs, a new output interface for visual tasks beyond text tokens.

Mohamed Eltahir, Fardows Adam, Duaa M. Tahir +6

King Abdullah University of Science and Technology (KAUST), Thuwal, Saudi Arabia · Department of Computer Science, Edge Hill University, Ormskirk, England

Answer Posterior

Sep 28, 2026cs.CL

When Confidence Rises Too Early: Detecting Shortcut Reasoning via Premature Answer Commitment

The reasoning trajectory of a Large Language Model (LLM) is often treated as a verbalized description of its internal reasoning. However, such trajectories can be unfaithful: a model may rely on shortcuts to reach an answer and then post-rationalize the decision with a seemingly coherent chain of thought. Detecting this shortcut reasoning is challenging because existing monitors and verifiers mainly inspect textual traces or final outcomes, rather than how the model's belief in its answer develops during generation. We introduce ConfLens, a framework that tracks the evolution of confidence in the final answer throughout reasoning. Across three shortcut reasoning settings, we observe a common pattern of premature confidence, where shortcut samples become highly confident in the final answer at early reasoning stages. Existing confidence estimation methods, however, show limited generalizability, reliability, or efficiency for detecting this behavior. We therefore propose the Distributional Answer Commitment Score (DACS), a distributional confidence estimator that measures the entropy of the model's probability distribution over answer commitment at each reasoning step. DACS captures how concentrated the model's answer belief is without requiring ground-truth answers or task-specific verifiers. We further convert ConfLens detection results into interpretable signals for reward models to reduce their preference for shortcut reasoning. Experiments on mathematical and code reasoning tasks show that ConfLens with DACS improves shortcut reasoning detection by over 4.3% F1 compared with strong baselines and reduces the mismatch between faithfulness and correctness in reward model preferences.

Zhaohan Zhang, Junjie Liu, Chengzhengxu Li +6

Queen Mary University of London · Tongyi Lab, Alibaba Group · Xi’an Jiaotong University

Reasoning Skills Overconfidence Answer Commitment

Sep 28, 2026cs.CV

From Perception to Integration: Revisiting the Internal Dynamics of Reasoning in Vision-Language Models

Vision-language models (VLMs) can answer simple visual questions, but often struggle when one question requires several visual judgments. We study this gap with controlled tasks for feature binding, numerosity, spatial relations, and amodal completion, together with a Composite task that combines them. Matched counterfactual image pairs isolate changes in the visual evidence needed to answer. Across four models, direct answers, hidden-state readouts, and state interventions show that the individual judgments can be made without explicit reasoning and that intervening on the corresponding states can affect the answer. During reasoning, the Composite answer becomes decodable from hidden states and usable from shortened traces, often before the model stops on its own. We train a small detector to predict this readiness and stop reasoning at that point. On MMStar and RealWorldQA, this reduces mean reasoning tokens by 79.1% and 74.5%, while average accuracy rises by 3.13 and 3.30 percentage points, respectively. These findings connect the internal development of answer readiness to a practical rule for allocating reasoning computation.

Rong Yu Xu, Prayag Tiwari, Shaolei Zhang

Shenzhen College of International Education · Halmstad University, Sweden · Renmin University of China

Recent Vision-Language Models Visual Reasoning Reasoning Skills Perception Answer

Sep 28, 2026cs.AI

PairPref: When Should Memory Guide the Answer? A Benchmark for Contextual Preference Use

Memory-augmented assistants use retrieved preferences to guide their responses. A small change in the situation can change whether a preference is appropriate while barely affecting its retrieval similarity. Memory benchmarks typically test whether systems store and retrieve preferences, with less attention to when those preferences should apply. We introduce PairPref, a benchmark of contextual preference use. Each pair changes only the situation, keeping the preference, request, and four candidate replies fixed. The preference remains valid in both situations. In the selection track, models must choose the reply that applies the preference only where appropriate. In the free-generation track, they must decide when to apply it without seeing candidate replies. Both tracks use the same 1,227 pairs across 45 preferences and eight situation categories. We evaluate eight models, most of which achieve selection scores (

Δ

) of 51 to 65 points. In free generation, however, both responses are appropriate for their respective situations in only 3.6% to 18.3% of pairs. Models continue to apply the preference in both situations even with fewer retrieved memories, alternative presentation formats, and a stricter prompt. These results show that models still struggle to judge when user preferences apply and respond accordingly.

Mingfei Lu, Mengjia Wu, Yi Zhang

AAII, University of Technology Sydney

Preference Alignment Learning Conversational Memory Contextual Reasoning Artificial Intelligence Assistants Answer Similarity and Metrics

Sep 28, 2026cs.CV

When Does an Image Determine the Answer? Benchmarking Visual Answerability across Charts and Scenes

Reliable visual question answering requires correct answers when evidence is sufficient and abstention when it is not. We introduce a benchmark that connects complete-question evaluation with explicit evidence for its labels across PlotQA charts, CLEVR rendered scenes, and GQA photographs. Each question groups original and edited images, presented independently; success requires every supported answer and every required abstention to be correct. For chart missing-information labels, executable witnesses establish that admissible complete charts give different answers but identical pixels after masking. Scene labels follow source programs and edits, with a residual-cue analysis for photographs. Across 72,000 responses from six model configurations, the highest observed complete task success rates are 57.0%, 43.5%, and 33.7%, respectively. On charts, the strongest configuration achieves 96.2% per-view decision accuracy, yet 265 of its 835 groups with every decision correct still contain incorrect answers. Evaluating supported answers and necessary abstentions together exposes failures that answerability decisions alone conceal.

Sungguk Cha, Mintae Kim, Youngsub Han +2

LG Uplus, Seoul, South Korea

Answer Chart Scene Understanding

Sep 27, 2026cs.CL

Calibration, Not Answer Selection: Distilling Internal Confidence in Reasoning Models

Reinforcement learning with binary correctness rewards trains correctness, not calibrated confidence. The confidence that reasoning models verbalize is systematically overconfident, and the problem is not merely one of scale: verbalized confidence tracks how willing a model is to commit to an answer, not how likely the answer is to be right. Post-hoc rescaling therefore fits one distribution but rarely transfers. We look inside the model instead. On factual question answering, a linear probe on the hidden state between the chain of thought and the answer is substantially better calibrated: its expected calibration error is 5 to 38 times lower than that of the verbalized score across four benchmarks and two model families. However, when used to pick among N sampled answers, that same probe nearly ties majority voting yet falls far short of the oracle. Internal states answer "how certain am I" well and "which answer is right" poorly, so the signal should be reported as a confidence rather than used to select answers. As a result, we introduce probe-guided self-distillation (Probe-SD): score a model's own sampled traces with the probe, overwrite the confidence each trace states, and finetune the base checkpoint of the same family, so nothing but the model itself remains at test time. On Qwen3-14B, Probe-SD cuts ECE from 0.178 to 0.024 in-domain and from 0.542 to 0.113 out-of-domain, where it also beats post-hoc recalibration and self-consistency distillation. The resulting confidence is well-calibrated and useful for weighted voting, behaviors previously attributed to online RL, here obtained with supervised finetuning alone.

Yadong Xi, Rongsheng Zhang, Tangjie Lv +2

NetEase · Amazon · Singapore University of Technology and Design

Confidence Calibration Large Reasoning Models Self-Distillation Framework Answer

Sep 26, 2026cs.AI

Instruct, Not Answer: Using Instruction Privileges in On-Policy Context Distillation

On-Policy Context Distillation (OPCD) has recently emerged as a powerful technique for transferring context to student models and for self-improvement. In OPCD, the teacher is conditioned on privileged information, and the goal is to minimize the Kullback-Leibler (KL) divergence between the privileged teacher and the student, evaluated on student-generated tokens. Many existing studies show that using instance-specific gold answers or gold demonstrations as the default privilege can hurt training performance, especially out-of-distribution (OOD). In this work, we instead design general instructions that target common student mistakes observed on the training samples, and show that such simple instructions can outperform gold as the OPCD privilege. In autoformalization tasks, using a matched formatting instruction as the privilege could outperform gold in OOD accuracy by a large margin. In 7 out of 8 experiments using ProverQA, ProofWriter, and ProntoQA as datasets, and Qwen3-Thinking and Olmo3-Thinking families as models, matched instruction privileges outperform gold in OOD by 4 to 17 points, while remaining on par with gold in-domain. Each instruction is only a few sentences (and thus contains much less information compared to all instance-specific gold) and is applied uniformly to every training sample. These results indicate that a general instruction, which applies equally to source and target domain examples, can be substantially more transferable than instance-specific gold in OPCD while maintaining in-domain performance.

Hantao Yu, Xiaoxue Han, Udaya Ghai +4

Columbia University · Amazon Web Services

Online-Policy Distillation Instruction Out-Of-Distribution Answer

Sep 24, 2026cs.CL

Reasoning Instructions Can Break Answer Decoding in Vision--Language Models

Chain-of-thought (CoT) instructions can distort multiple-choice VLM evaluation when a scorer appends a reasoning cue but reads answer-label logits before the model generates any rationale. We call this CoT-prefix scoring. On ScienceQA, Qwen2.5-VL-7B drops from 80.76% to 45.48%, and across five option-content permutations 93.54% of CoT-prefix predictions select the first slot. Condition-matched linear probes recover 78.94% from the same hidden states, while free generation restores 75.24%, showing that the answer often survives the prefix and the immediate readout fails. Vocabulary and layer diagnostics explain the mismatch: probability mass moves toward continuation tokens, while answer information remains linearly accessible in late layers. The effect recurs with varying severity across datasets and models, though not universally. These results show that CoT-prefix scoring can confound model knowledge with an evaluation-interface mismatch and should be avoided unless the requested and scored output events are aligned.

Zeyan Li, Siyuan Qiu, Jianfeng Xu

Shanghai Jiao Tong University

Qwen3 Answer Reasoning Skills Scientific Discovery Decoding Break

Sep 23, 2026cs.CV

Looks the Same, Answers Differently: Flip-Direction Steering for Robust Vision-Language Reasoning

Vision-language models (VLMs) achieve strong visual reasoning performance, yet subtle changes from routine image capture and processing can alter their reasoning trajectories even when images appear nearly identical. In long-horizon generation, the resulting activation shifts may accumulate across decoding steps, progressively altering reasoning tokens and ultimately changing the final answer, a phenomenon referred to as answer flips. To address this instability, we propose FlipDir (Flip-Direction Steering), a training-free inference-time method that estimates a low-rank flip-inducing activation subspace from contrastive pairs of original and answer-flipping inputs and selectively steers hidden states during decoding. A margin-based gate limits subspace attenuation to uncertain decoding steps, recovering original predictions while preserving stable ones. To evaluate robustness beyond accuracy or consistency on fixed test sets, we introduce VisFlip, a benchmark framework that constructs evaluation groups for a target model and visual variation setting to separately assess recovery of original predictions and preservation of stable ones. VisFlip spans nine dataset-variation combinations across scientific reasoning, robot-scene understanding, and medical VQA, covering subtle visual variations common in each domain. Experiments across 18 settings demonstrate that FlipDir consistently outperforms existing methods on the combined recovery and preservation metric. We will make our code publicly available.

Yeonsung Jung, Joonhyun Jeong, Hoang Pham +4

Graduate School of AI, KAIST · NAVER Cloud · Department of Computer Science and Engineering, The Ohio State University +3

Visual Reasoning Steering Answer Subspace

Sep 23, 2026cs.LG

Order-Invariant Answers, Order-Sensitive Representations in Mathematical Reasoning

Reordering a set of mathematical rules without changing its meaning should preserve the correct answer, but must a model's internal representations stay invariant too? We investigate this question using synthetic multi-step function-composition problems, each presented under multiple rule orderings with the same correct answer. We measure accuracy and permutation signal-to-noise ratio (SNR), which quantifies how distinctly ordering patterns are represented relative to variation across problem instances. Across 16 language models ranging from 1B to 8B parameters, we find a pattern: models that solve reordered problems more accurately represent different rule orderings more distinctly. Layer-averaged permutation SNR is positively rank-correlated with accuracy in every synthetic setting we evaluate, with Spearman correlations reaching 0.86. These findings highlight a distinction between answer invariance and representation invariance: successful mathematical rule composition can accompany distinct internal representations between equivalent rule orderings. This motivates distinguishing answer invariance from representation invariance, and offers a representational perspective on mathematical reasoning beyond answer accuracy alone.

Zhixu Silvia Tao

Operations Research and Financial Engineering Princeton University Princeton, NJ 08540

Mathematical Reasoning Order Matters Answer

Sep 22, 2026cs.CL

What Changes When Fact-Verification Scores Improve? Evidence and Answer Accounting Across Trained Verifiers and LLMs

A joint fact-verification score assesses answers and submitted evidence together. When the score improves, how much of the gain remains if the answers are held fixed? On FEVEROUS, strict score is the percentage of claims with a correct answer and a complete annotated evidence group in the submitted evidence. Across four trained DeBERTa checkpoints and 7,890 claims, replacing DCUF evidence with UnifEE evidence raises strict score by 9.61 percentage points, compared with 1.96 percentage points in answer accuracy. The paired 95% interval for the strict-score gain is [8.77, 10.43], conditional on these checkpoints. Replacing only the evidence passed to the scorer accounts for 7.92 or 9.08 percentage points when we retain the answers generated from DCUF or UnifEE evidence, respectively. To examine how this evidence gain depends on evaluation choices, we generate 470,400 responses from two 8B LLMs on FEVER, FEVEROUS, and SciFact under two answer formats and two context budgets. Increasing context from 256 to 2,048 tokens raises the fixed-answer evidence gain on FEVEROUS by 3.84 and 3.10 percentage points for Qwen and Llama, respectively. The effects fall short of the prespecified cross-dataset criterion, while some intervals extend beyond the two-point small-effect bound. Post-hoc analyses quantify changes in answers and submitted evidence, and show when aggregate accuracy and evidence-coverage rates miss the claim-level pattern. The four answer-evidence score combinations reveal changes that endpoint and aggregate metrics leave unresolved.

Han Chen, Yingrui Li

Sep 17, 2026cs.CV

DocAttriBench: Benchmarking Answer Grounding in Document Visual Question Answering

Answer grounding in document visual question answering remains an open challenge: most benchmarks lack grounding annotations or provide limited-quality labels, while constructing grounded datasets still requires costly manual effort. We introduce DocAttriBench (DAB), a large-scale benchmark for fine-grained, element-level source attribution in Document VQA, grounding answers to specific layout elements such as text blocks, tables, and images. To build DAB, we propose a Mask-based Perplexity-Derived Attribution method (MAPPET) that combines document layout and language modeling to identify the most informative element for each answer. MAPPET measures the increase in perplexity after masking candidate elements and attributes the answer to the element contributing most to model confidence. Applying MAPPET to multiple existing Document VQA datasets yields DAB, with 237k documents and 296k question-answer pairs with element-level grounding. We benchmark grounding-capable multimodal LLMs on DAB, evaluating answer accuracy, attribution accuracy, and overall answer quality. Results show that while larger models generally achieve higher answer accuracy, even the strongest models often fail to localize the supporting elements. DAB provides a scalable benchmark for developing grounded, verifiable, and trustworthy Document VQA models. Dataset and code are available at https://aimagelab.github.io/DocAttriBench/.

Luca De Grandis, Silvia Cappelletti, William Raccagni +3

University of Modena and Reggio Emilia, Modena, Italy · University of Modena and Reggio Emilia Modena, Italy · University of Pisa, Pisa, Italy +1

Human-Annotated Benchmark Document Answer Multimodal Large Language Models

Sep 17, 2026cs.LG

Correct Now, Insufficient Later: Auditing Update Sufficiency in Context Compression

A memory can answer a current query correctly while discarding distinctions required by a later update. We investigate this failure with a paired-history audit: two histories have the same current answer, receive a shared future update, and require different subsequent answers. A pilot evaluates 24 history pairs across six synthetic mechanisms, 12 memory conditions, two repeats, and two model backends. A deterministic frontier selector obtains strict reveal accuracy of 96/96 on DeepSeek and 82/96 on GLM; a structured writer obtains 62 successes with one unresolved outcome and 56/96. The configured four-outcome joint contrast has finite-sample identification intervals of [0.521, 0.542] and [0.292, 0.313], not confidence intervals. A record-level audit distinguishes retained-state adequacy, response delivery, and answer-schema compliance without changing those original scores. It finds 26 and 25 well-formed but semantically wrong structured reveal memories, while all 14 GLM frontier reveal failures contain correct values in the wrong wrapper. Tombstone removal produces 16/16 exact replay failures in the targeted mechanism. Identifier renaming then exposes a separate flaw: original frontier late-reference adequacy falls from 8/8 to 94/320 transformed instances. We provide and test a label-equivariant repair, but it preserves only 2/8 original late-reference answers: eliminating a naming shortcut does not solve unknown future relevance. These results support a scoped evaluation methodology and reproducible failure analysis, not general superiority of the repaired algorithm. Paid pilot evidence, retrospective diagnostics, and new offline tests are reported separately; no independent held-out or natural-task validation is claimed.

Guangzhe Zhang

Independent AI Researcher

Model Auditing Correct Incomplete Answer Data Compression Methods

Sep 16, 2026cs.CR

AUDITPLAN: Commit, Then Answer for Auditable Safety Alignment

Safety tuning pipelines judge only the final answer, which makes it difficult to distinguish robust refusal from two undesirable shortcuts: blanket refusal on benign requests and polished but unfaithful safety rationales that do not actually constrain the answer. We propose AUDITPLAN, a single-model plan-then-answer approach where the model first emits a compact structured safety plan and then answers conditioned on it. The plan records a threat label, intended action, and explicit constraints, enabling machine-checkable auditing while remaining hidden from users at deployment. We train this behavior with supervised fine-tuning followed by reinforcement learning with FAITHGATE, a reward-gating objective that grants answer reward only when the safety plan is correct. This discourages safe-looking but unfaithful behavior and promotes tighter plan-answer coupling. Across Qwen backbones, AUDITPLAN improves both robustness and auditability: on Qwen2.5-3B-Instruct, FAITHGATE reduces ASR from 24.0% to 11.6%, LSR from 1.0% to 0.36%, and over-refusal from 11.0% to 2.0%, outperforming answer-only RL, free-form explanation, and weighted-sum structured rewards. Similar trends hold for Qwen2.5-1.5B-Instruct. Larger-model confirmation runs on Qwen-3-4B-Instruct and Qwen2.5-7B-Instruct preserve the same trend suggesting that explicit internal commitments can make safety alignment more faithful, robust, and auditable.

Sai Sri Pushpa Jampani, Kshitij Mishra, Asif Ekbal

Indian Institute of Technology Patna, Bihar, India

Safety Alignment Model Auditing Faithful Explanation Answer Commits

Sep 15, 2026cs.CL

Rewarding Reasoning, Not Answers: Fixing and Bounding Test-Time Reinforcement Learning on Medical QA

Test-time reinforcement learning adapts a model on its own unlabeled test set using majority-vote pseudo-labels and has shown strong results in mathematics. We show that this recipe collapses on medical multiple-choice QA: accuracy stagnates while output diversity rapidly declines. Through a controlled experiment that keeps the questions, model, and optimizer fixed while changing only the answer space, we trace this failure to answer-space structure rather than domain difficulty. In small answer spaces, incorrect rollouts often collide on the same wrong pseudo-label and reinforce it; in large answer spaces, they disperse and receive little reward. This diagnosis motivates PROSE, Process Reward Guided Self-Training, which rewards reasoning quality instead of answer agreement. PROSE scores each reasoning step with a medical process reward model, assigns the trajectory reward as the minimum score across steps, and enforces answer-format constraints. Without labels, PROSE substantially improves a general Llama model, surpassing purpose-built medical models and matching much larger systems. Because the process signal is internalized into the policy, the adapted model requires no reward model at inference and transfers its gains to unseen datasets. We further show that the minimum aggregation is essential: mean aggregation can be exploited, saturating the proxy reward while degrading accuracy.

Kailong Fan, Anqi Pu, Yichen Wu +7

1Harvard Medical School/MGH · 3Harvard University · 2Zhejiang University

Process Reward Model Answer

Sep 15, 2026cs.CV

Vision And Text Transformer For Predicting Answerability On Visual Question Answering

Answerability on Visual Question Answering is a novel and attractive task to predict answerable scores between images and questions in multi-modal data. Existing works often utilize a binary mapping from visual question answering systems into Answerability. It does not reflect the essence of this problem. Together with our consideration of Answerability in a regression task, we propose VT-Transformer, which exploits visual and textual features through Transformer architecture. Experimental results on VizWiz 2020 dataset show the effectiveness and robustness of VT-Transformer for Answerability on Visual Question Answering when comparing with competitive baselines.

Tung Le, Huy Tien Nguyen, Le Minh Nguyen

Japan Advanced Institute of Science and Technology, Ishikawa, Japan. · Faculty of Information Technology, University of Science, Ho Chi Minh City, Vietnam · Vietnam National University, Ho Chi Minh City, Vietnam +1

Self-Supervised Vision Transformers Transformer Architectures Vista Energy-Based Transformer Answer

Sep 14, 2026cs.AI

Beyond Safe Answers: Segment-Aware Listwise Alignment for Reasoning Safety in Large Reasoning Models

Large Reasoning Models (LRMs) pose a dual-surface safety challenge: both intermediate reasoning traces and final answers can contain harmful content. Existing alignment methods often operate at the whole-response level, allowing unsafe reasoning to be masked by a safe-looking final answer. We propose Segment-aware Listwise Target DPO (SaLT-DPO), which addresses this gap through three mechanisms: (1) segment-aware listwise alignment that decomposes responses into reasoning and answer segments, independently scores each segment's safety, and aligns length-normalized segment rewards with soft target distributions over multiple candidates; (2) joint safety coherence regularization that applies a weakest-link principle to promote safety consistency across both segments; and (3) utility anchoring on benign prompts to mitigate over-refusal and reasoning degradation. Experiments on three LRMs show that SaLT-DPO consistently reduces unsafe rates for both reasoning and answer segments while mitigating degradation in benign compliance and preserving general reasoning performance. Ablation studies demonstrate the complementary contributions of its components.

JungMin Yun, Junehyoung Kwon, Hayeong Ryu +3

Department of Artificial Intelligence, Chung-Ang University · Graduate School of Advanced Imaging Sciences, Multimedia and Film, Chung-Ang University

Large Reasoning Models Reasoning Skills Safety Claims Answer Feature Alignment Backtranslation Augmented Direct Preference Optimization+1

Sep 14, 2026cs.AI

When Successful Knowledge Graph Edits Displace Correct Answers: Rank-Level Locality beyond Parameter Support

Editing a knowledge graph embedding (KGE) model to promote a desired answer can displace correct answers from the returned list. Locality tests based only on facts that reuse the edited parameter can miss this ranking effect. We introduce a common rank-displacement audit at three scopes: facts supported by the edited parameter, other correct answers to the target query, and correct answers across queries with the same relation. We also derive dimensional and geometric conditions for an update to improve the target while exactly preserving selected scores. On FB15k-237 with DistMult and ComplEx, direct promotion always moves the target into the top ten, but does so without damage in only 23.0--23.2% of edits. Strict preservation causes no measured damage, yet succeeds in only 1.3--1.4%. Support-regularized entity editing gives the highest joint success, 36.3--37.7%, while rank-truncated preservation reaches 32.8--34.7% and reduces the mean number of displaced answers from about 14 to 1.2. Experiments across dimensions, scorers, ranking conventions, and a learned editor show that locality depends on both the protected scope and the editing mechanism. KGE editing should therefore report correction success together with the incidence and severity of rank displacement.

Yi-Cheng Lai, Jerry Wang, Hsin-Ling Hsu +3

Institute of Information Science, Academia Sinica Taipei, Taiwan · University of Illinois Urbana-Champaign Urbana, IL, USA · National Chengchi University Taipei, Taiwan

Multimodal Knowledge Editing Knowledge Graph Embeddings Knowledge Graphs Answer Model Auditing

Sep 12, 2026cs.AI

From Parameters to Answers: How LLMs Retrieve and Use Their Internal Knowledge

How does a language model's dependence on query-routing information and target knowledge change as it answers a question? We study this question through layerwise interventions on the hidden state at the end of the question. Across Qwen, Llama, and Gemma, we compare country-continent questions with noun, adjective, and code answers while keeping several fitted measurements distinct. A pair-conditioned request direction describes which country is queried in natural single-country questions; a global request direction describes first- versus second-country requests in paired questions; separate selection candidates test control among contents already available in the hidden state. A diagnostic reanalysis of frozen Qwen natural-question states shows that the pair-conditioned direction grows stronger before interventions on it begin to alter later fitted knowledge, with this causal window opening while answer-supporting content is still forming. The paired three-model trajectories are not uniform: Gemma shows a partially overlapping mid-layer routing-content profile, whereas Llama has no sustained routing-effect window under the same gates. In the paired protocol, dependence on the global request direction decreases from fixed earlier to later layer sets while dependence on fitted content persists. A matched Qwen comparison shows that the pair-conditioned direction retains a late effect, so this operational handoff concerns the global fitted direction rather than all request information. These results separate early readability, natural strength, causal steering, and later content dependence.

Wenkang Wei, Yuan Fang, Renhe Jiang +2

University of Science and Technology of China Hefei, China · Singapore Management University Singapore, Singapore · The University of Tokyo Tokyo, Japan +1

Sep 11, 2026cs.CL

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.

Jiacheng Sang, Mengyuan Li, Sanxing Chen +3

Duke University · University of Pennsylvania

Agentic Search Queries Answer Atlas

Sep 9, 2026cs.CL

GANDR: Claim Auditing for Verifiable Legal Answer Generation

In high-stakes domains such as legal practice, a language-model answer is only useful to the extent that a reader can verify each claim against the source the system cites. Current grounded-generation pipelines score the answer as a whole, so a correct conclusion can rest on fabricated or loosely matched citations and still score well. Closing this gap requires both a system built for per-claim verification and an evaluation that measures it. We introduce GANDR (Grounded ANswer DRafter), a two-agent system in which a Drafter writes an answer in a structured legal-reasoning format and a separate Critic, with the same view as a human verifier, audits each claim against its cited source and emits a per-claim audit trace on every round. We pair it with a strict correctness criterion requiring every citation to resolve to a passage the retriever returned. On a 185-item legal benchmark where all six systems share one backbone, one retrieval surface, and one citation instruction, GANDR ranks first on every primary metric, reaching 70.8% strict accuracy and leading the strongest baseline by 11.3 points (p<0.01). Reverting the protocol-anchored commit rule lowers strict accuracy by 22.7 points, and the strict lead stays positive on three further backbones, at +3.2 to +6.5 points. This lead traces to the Drafter configuration and the protocol-anchored commit, not to rewriting. Against two law-trained annotators the audit flags under-supported claims at F1 0.84 as a binary detector, while its four-way verdict labels agree only weakly and are advisory. Code is available upon request.

Chen Qian, Yimeng Wang, Yu Chen +2

William & Mary · Anytime AI

Legal Reasoning Tasks Category-Aware Atomic Claims Model Auditing Llm-As-A-Judge Answer Commits+1

Sep 9, 2026cs.CL

Stable Answers, Unfinished Reasoning: Why Self-Consensus Is Not a Safe Early-Exit Signal

A natural way to cut reasoning-model inference cost is to repeatedly probe a single partial trajectory for its current answer and stop once probes agree -- self-consensus. We ask whether any such rule is both safe and token-saving, and whether one can be selected once and reused. A preregistered sweep of 3,520 consensus rules, replayed on frozen trajectories from two models and three benchmarks, clears none of three acceptance gates fixed in advance; the frontier reproduces on a held-out split and on two unseen models -- while a boundary-confidence control (DEER) swept through the same pipeline clears all three. The reason lies in the signal: agreement establishes that the current answer persists under a fixed probing procedure, not that the reasoning has terminated -- a consensus-termination gap. Stopping on it commits non-terminal answers. At a rule still saving 32% of the tokens, one stop in nine fires on an answer the trajectory itself later abandons, and most of those stops cut off a correction it would otherwise have made. Widening the agreement window does not remove them: the share levels off near 7%, and by then the saving has fallen to 8%. Probe re-wording and a hand-labelled error taxonomy show the agreed answer is often a placeholder the model had not settled on. Used on its own as the stop signal, agreement fails not because it is insufficiently strict, but because it repeatedly measures the wrong object.

Yunxiang Mo, Donghao Zhao, Hejia Geng

HKUST · University of Oxford

Self-Consistency Reasoning Skills Answer Inference Cost Completion Single Trajectory+1

Sep 9, 2026cs.AI

The Era by Eon Benchmark: A Generated Enterprise Estate with Exact Ground Truth for Benchmarking LLM Agents

LLM agents for enterprise systems of record cannot be evaluated on customer production data, and no existing substitute provides ground truth. We present the Era by Eon Benchmark for evaluating LLM agents that use enterprise tools. The benchmark is built around a complete fictional company. It includes product simulators, company-specific internal databases, benchmark questions, and computed answer keys. Industry, company size, business model, application portfolio, and a seed define each company. One seeded entity graph supplies shared company data to simulators of Salesforce, Zendesk, Slack, Gong, and other products. A questionconditioned generator creates the schemas and records for internal databases. It takes shared entities, keys, and values from the same graph before generating database-specific facts. Both mechanisms therefore describe one consistent enterprise estate. Every expected answer is computed from the final records, so grading is exact. Design and answer-key checks validate the internal databases. A realism scorecard and adversarial detector validate the entity graph. Across 23 generated companies, the mean realism score rose from 61.8 to 97.0, with zero records flagged as synthetic. In the reported simulator-track comparison, nine models answered the same 33 questions three times each. Accuracy estimates ranged from 42.4% to 76.8%, and three of 36 pairwise differences remained supported after correction.

Benjamin Gruenbaum, Doron Porat, Assaf Natanzon +3

Eon

Personalized Large Language Model Agents Large Language Model Agents Ground Truth Answer

Sep 9, 2026cs.CL

Which Medical Questions Deserve Rationales? Perturbation-Sensitive Selection for Robust QA

Medical question-answering datasets often contain answer labels, whereas high-quality rationales remain scarce, noisy, or costly to validate. This changes the acquisition question: rather than asking which questions should be labeled, we ask which already-labeled questions should receive rationale supervision under a fixed token budget. We study an offline version of this problem in which candidate rationales are visible to the selector but withheld from downstream training unless selected. We propose root-mean-square Robustness-based Sample Prioritization (RMS-RSP), which perturbs hidden states only at rationale tokens and measures the resulting shift in the gold-versus-best-distractor margin. Across five medical QA datasets, MedGemma-4B-IT, three training seeds, ten budgeted non-RSP selectors, and an unbudgeted full-supervision reference, RMS-RSP provides a deliberately qualified result. Its locked-budget accuracy is 60.61% on average versus 60.08% for Random, with a statistically resolved gain only on AfriMed-QA (+1.44 points). Its full-budget accuracy area is not better than Random. However, after three answer-option reorderings, RMS-RSP improves robust accuracy and semantic consistency by 1.91 and 2.85 points on average, respectively, with the same direction on all five datasets. Training on every pool rationale raises macro accuracy to 63.74%, but consumes 29--254 times more rationale tokens and does not uniformly improve robustness. These findings do not establish universal accuracy gains; they instead suggest that rationale-local boundary sensitivity can identify supervision that improves invariance to semantically equivalent formatting changes.

Yuexin Wu, Dayou Yu, Vasile Rus

Department of Computer Science, University of Memphis

Medical Visual Question Answering Clinical Reasoning Training Free-Form Textual Rationales Answer Distractor

Sep 7, 2026cs.CL

FramingQA: Does the Question Shape the Answer? Measuring the Compositional Framing Effect

We introduce FramingQA, a benchmark that measures the model sensitivity to question framing across law, medicine, finance, and robotic simulations. Large language models (LLMs) often change their responses to subtle rephrasings that align with an implied stance by users. This can leave users with advice tainted by how they happened to phrase a question rather than by the underlying facts, and the consequences are highly costly in high-stakes domains. Because in the realistic scenarios, both expert practitioners and non-expert users frequently ask LLMs questions containing incomplete or misleading assumptions, models are highly susceptible to those framings. To test this, we inject the framing bias across three nested levels: a framing-biased question phrasing (root), an injected framing-biased premise prepended to a neutral question (propositional), and a premise paired with a framing-biased question (global). Evaluating nine open models (3.8B-70B) across four families, we find that strong per-variant accuracy does not guarantee the robustness across differently phrased questions under the fixed factual information.

Hazel H. Kim, Andrew M. Bean, Guilherme Affonso Ferreira de Camargo +10

Department of Computer Science, University of Oxford · Oxford Internet Institute, University of Oxford · Thomson Reuters +7

Task Framing Question Answer

Sep 3, 2026cs.CV

When Do Frozen VLMs Respond to Image-Free Object-Token Edits? An Answer-Key-Free Protocol and What It Reveals

Answering what-if queries about a scene with a VLM usually means injecting the assumption as text or repainting the scene with a generative model. We instead move the edit to the representation level, before the model input. The image is abstracted into a set of object-level tokens, and the original image never enters the VLM. This design rests on an open question: when do frozen VLMs actually respond to such token edits? We introduce an answer-key-free protocol: no post-edit answer is annotated. It scores edits whose answers are logically determined, and audits itself by reversing each scoreable choice. The protocol reveals three structures. The response is not free: explicit edit teaching, not ordinary VQA training, produces it in dense scenes and multiplies it in sparse ones, on all three operations. Once on, it is governed by token cleanliness and density, with deployable detector+segmenter tokens competitive with the oracle and outperforming it on VRSBench. And reading is a separable axis: the image-free token route preserves 92-96% of a matched patch-token baseline's free-text VQA, and the answers measurably depend on the tokens. The response, cleanliness, and reading structures are sign-preserved across two remote-sensing datasets (iSAID, VRSBench) and three frozen LM backbones. We release the probe generator, records, judge logs, and code.

Wonbin Son, Gyumun Choi, Junil Seo +3

Changwon National University · Chung-Ang University

Frozen Vision-Language Models Multiple-Choice Visual Question Answering Answer Remote Sensing Protocol Generative Models

Sep 1, 2026cs.CL

How Correct Is Your Answer? A Semantic Correctness Framework for Open QA Evaluation

Reliable evaluation of open-ended question answering remains a bottleneck for measuring answer correctness of modern LLMs. Unlike multiple-choice tasks, free-form answers may be correct in many surface forms and may fail in qualitatively different ways, including incompleteness, contradiction, overgeneration, and endorsement of false premises. Existing judgment-based and similarity-based metrics often collapse these distinctions. We address this gap with three reusable contributions. First, we introduce a semantic correctness taxonomy that assigns open-ended answers to eight ordered classes, separating verbose-but-correct answers from those contaminated by hallucinated content. Second, we release CAP-Correctness, an 8.8k-example benchmark spanning widely used QA datasets, and CAP-Statements, an 11k-example dataset for converting question-answer pairs into declarative statements for natural language inference (NLI) training and statement-based evaluation. Third, we introduce CAP (Context-Aware Precision), a reference-based metric that scores question-conditioned statements using bidirectional NLI. Under a monotonicity protocol testing whether metrics respect the taxonomy's intended ordering, CAP outperforms established baselines.

Elitsa Yotkova, Violeta Kastreva, Petar Velkov +4

Sofia University “St. Kliment Ohridski” · ETH Zürich · University of Zurich +1

Open-Ended Responses Answer Evaluation Agent

Aug 31, 2026cs.AI

The Answer Is Not the Argument

Chain-of-thought monitoring is proposed for AI oversight, yet evaluations often provide monitors with a trusted reference answer. We ask whether answer access improves reasoning verification or mainly exposes incorrect conclusions. We collected 237 step-numbered solutions to 79 Humanity's Last Exam physics questions from three frontier models, with no inserted errors, and independently labelled final-answer correctness and the first false step. The reference standard combined physicist annotations, an independent LLM debate, and source-masked adjudication. This yielded 24 critical traces in which the answer was correct but the trace contained a genuine error. 8 LLM monitors evaluated traces blind, with an unverified or certified answer, or after a blind commitment. Certification raised mean balanced accuracy from 0.637 to 0.796, while exact first-error localization rose from 0.261 to 0.379. Certification changed recall (the fraction of error traces flagged as erroneous) from 0.653 to 0.951 on wrong-answer traces but from 0.521 to 0.438 on critical traces; the contrast had the same direction for all 8 monitors (question-bootstrap 95% CI [+0.256, +0.506]). After blind commitment, monitors shown the answer newly flagged 93.8% of previously passed wrong-answer traces as erroneous, but only 18.0% of critical traces. Answer access therefore improves conclusion-consistency checking rather than independent verification of the supporting argument. For AI safety, these traces provide a benign analogue of reward hacking: an acceptable output does not establish that the process producing it was sound. Although the errors studied here were ordinary and mostly non-load-bearing rather than adversarial, trusted-answer evaluations may similarly overstate monitoring capability when acceptable outputs conceal unsound reasoning.

Will Yeadon, Sergio Juárez, Paul Mackay +5

Department of Physics, Durham University, Durham DH1 3LE, UK · Escuela de Ingenier´ıa de Telecomunicaci´on, Department of Signal Theory and Communications, University of Vigo, Vigo E-36310, Spain · Independent Researcher

Answer Reasoning Traces Model Auditing Correct Traces

Aug 31, 2026cs.AI

Answer Probing-Guided Search for Diverse Solution Exploration of LLMs

Generating multiple diverse and high-quality solutions is valuable for many applications, such as code-test generation and drug discovery. However, Large Language Models (LLMs) tend to converge on a single high-confidence solution during inference, limiting exploration of alternative valid solution paths. Existing test-time methods promote diversity through tree-like search and prune semantically similar branches using response-level semantic embeddings. However, we find that such embeddings are easily confounded by linguistic and stylistic similarities, making it difficult to distinguish genuinely distinct solution paths. To address this, we introduce Answer Probing, which probes the potential answer an LLM would reach from an intermediate reasoning path. We demonstrate that the hidden states of probed answers more effectively differentiate distinct solution paths than semantic embeddings, and the perplexity of probed answers serves as a practical proxy for reasoning correctness. Based on these findings, we propose Answer Probing-Guided Tree Search (APTS), which guides the tree search by the probed answers' hidden state similarity and perplexity. Experiments on three reasoning tasks across two LLMs show that APTS consistently enhances solution diversity, demonstrating its effectiveness and robustness.

Yi Fang, Que Shen, Chengpeng Li +6

University of Science and Technology of China · Zhongguancun Academy · Alibaba Group +2

LLM Reasoning Strategies Reasoning Paths Test Generation Answer Exploration

Aug 30, 2026cs.CL

Compression-Aware Abstention: Teaching LLMs to Refuse When KV-Compression Masks Remove Answer Evidence

KV-cache compression reduces LLM inference memory by evicting context tokens, but when the evicted tokens contain answer-bearing evidence, the model may hallucinate instead of recognizing that the compressed context is insufficient. We address this failure from a behavioral perspective: to our knowledge, this is the first work to formulate compression-aware abstention as a learning problem, in which a model learns to answer when supporting evidence survives compression and abstain when it does not. We construct supervision from compressor survival masks and tight answer-bearing spans, labeling examples as Confident when evidence survives and Abstain when it is removed. A 10.1M-parameter LoRA adapter trained on ~2.6K MuSiQue 2-hop QA examples reduces base-model hallucinations by 97% under prompt-style truncation while preserving correct answering on evidence-retaining examples. Unlike prompt-only abstention baselines, which over-abstain on many answerable high-retention examples, the trained adapter learns a conditional policy. We also evaluate the method under actual compressed-cache decoding, where multi-compressor training yields a 6-22x relative lift over the unaided base on evidence-retaining examples. Controlled-deletion experiments show that the learned behavior is driven by evidence content rather than input length alone.

Mohammadali Khodabandehlou, Bhaskar Krishnamachari

University of Southern California

LLM Inference Optimization Large Language Model Hallucination Abstention Answer

Aug 26, 2026cs.AI

Candidate supply and answer selection shape the value of LLM judging in multi-agent systems

Multi-agent systems (MAS) sometimes already have the potential to answer correctly, but still report a wrong answer. Explaining this outcome is difficult because generation, communication and final answer-selection rules usually change simultaneously. We conceptualize multi-agent reasoning as an evolutionary pipeline of candidate generation, peer communication and terminal selection, wherein consensus without quality control can exhibit patterns of memetic drift. We study two questions: (1) when an LLM judge provides effective selection pressure by supplying a signal of answer correctness for candidates generated in a multi-agent system, and (2) when using that signal improves the reported answer. To map judge reliability, we analysed 15,336 questions from MMLU-Pro, GPQA, MedXpertQA and MuSR, with Humanity's Last Exam analysed separately. To test these rules, we replayed 81,390 fixed candidate pools drawn from 16,278 questions across five benchmarks. We report three findings. (1) A correct answer is often already present among the generated candidates, but the system can still converge on and report a wrong answer. (2) Judge reliability is not a fixed trait of the model, but varies with the task, the generator and how rare the correct answer is. (3) Combining answer frequency with the judge's evaluation changed only the final answer-selection rule and raised accuracy from 63.82% to 70.82-70.95%, primarily by rescuing correct answers that were outnumbered by popular errors. In the systems studied here, the value of generating more candidates depends on whether those extra samples make correct answers present, frequent or recognisable. By isolating generation, recognition and selection, these findings establish a diagnostic basis for designing multi-agent architectures that protect generated correct answers from being lost.

Jia-Hao Ji, Sijie Li, Jiabei Cheng +3

Department of Neurology and National Center for Neurological Diseases, Huashan Hospital, State Key Laboratory of Medical Neurobiology and Ministry of Education Frontiers Center for Brain Science, Shanghai Medical College, Fudan University, Shanghai 200040, China. · Institute of Science and Technology for Brain-Inspired Intelligence, MOE Key Laboratory of Computational Neuroscience and Brain-Inspired Intelligence, MOE Frontiers Center for Brain Science, Fudan University, Shanghai 200433, China. · Shanghai Academy of Artificial Intelligence for Science, Shanghai 200003, China. +1

Multi-Agent Large Language Model Systems Multi-Agent Reasoning Large Language Model Judges Answer Candidate Data Generation

Aug 13, 2026cs.CL

From Atomic Evidence to Logical Composition: Structured Compositional Reasoning over Compound Answer Options

Large language models often fail when answer options require combining atomic judgments under explicit logical operators, even when they judge the individual atoms correctly. We study compound options connected by AND, OR, and NEITHER/NOR, introducing a framework that decomposes each option into atomic answers and scores contrastive hypotheses about each one, so the model never sees a compound option. An operator-constrained integer linear program then composes the calibrated scores into a single prediction. We evaluate on LOGICAL-COMMONSENSEQA and introduce LOGICAL-SATA, a reading-comprehension benchmark derived from SATA-Bench. Our framework improves Macro-F1 from 48.3 to 77.0 on the human-validated LOGICAL-COMMONSENSEQA split and from 47.0 to 75.6 on LOGICAL-SATA, with the largest gains on NEITHER/NOR.

Obed Junias, Maria Leonor Pacheco

University of Colorado Boulder

Logic Answer Atoms

Aug 10, 2026cs.CL

Is This Your Final Answer? Cross-Contextual Consistency as a Measure of LLM Credibility

Large language models (LLMs) are powerful black-box systems, making it difficult to discern whether their answers reflect stable internal beliefs or superficial pattern matching. We identify cross-contextual consistency as an underutilized behavioral property of LLMs: a credible answer should remain stable when the same task is placed under topic-aligned, content-neutral contextual variation. Building on this intuition, we operationalize Cross-Contextual Consistency (C3) by comparing model generations under original and perturbed prompts. Across 26 models and six benchmarks spanning reasoning, factuality, and code generation, we find that answers with smaller cross-contextual shifts are more likely to be correct or factual. We demonstrate that C3 provides a complementary axis of evaluation and can serve as a benchmark usefulness diagnostic, identifying which portions of a benchmark remain informative even when aggregated scores are widely considered "saturate".

Siyang Wu, Yibo Jiang, Bryon Aragam

Data Science Institute University of Chicago Chicago, USA · Department of Computer Science University of Chicago Chicago, USA · Booth School of Business University of Chicago Chicago, USA

Large Language Model Reliability Answer

Aug 10, 2026cs.AI

Business Truth, not SQL Accuracy: A Rule-Gated 7B Analytics Agent Outperforms a Direct-Prompted 32B Baseline

LLM analytics agents are evaluated on SQL syntax accuracy, but production failures look different: questions with two valid business definitions, questions the warehouse cannot answer, deprecated columns after a schema change, and queries that execute successfully while returning the wrong business number. No execution-match metric can score them. This paper introduces WarehouseReliabilityBench, 400 frozen tasks over two synthetic warehouses in which roughly half the correct responses are a clarification, an abstention or a refusal, with pinned denominators and a pre-registered paired bootstrap fixing each claim verb before the numbers existed. QueryProof, a 7B agent, uses rules derived from a semantic layer and physical catalog to determine its behaviour, and gates every answer on deterministic post-execution checks. On an 80-task synthetic test split evaluated once, QueryProof outperforms a direct-prompted 32B baseline by +0.237 [+0.112, +0.375] Business Truth Rate at 71.0% lower cost per correct answer; against a cost-matched few-shot baseline the accuracy gain holds but the cost difference does not resolve. This compares systems rather than model sizes: the 32B baseline receives none of the scaffolding. False success falls from 0.754 to 0.351 of returned answers, and no wrong number was returned on an answerable task (0 of 24), though 13 answers went to questions requiring clarification or abstention. Removing the routing layer changes little (0.562 against 0.537), so the result does not depend on escalation. Routing tuned on validation over-abstains on test, and the fitted confidence model loses to the heuristic it replaced. Resampling template families rather than tasks widens both accuracy intervals to include zero, so the effect's direction is better supported than its magnitude. The gain tracks the deterministic layer, though no component ablation was run.

Morris Lee

Independent Researcher

Data Science Agents Answer Truth Strong Baseline Abstention Warehouse Automation

Aug 10, 2026cs.CV

Right Answer, Wrong Heat: Explanation-Aware Evaluation and Thermal-Grounded Feedback for MLLMs on Infrared Images

General-purpose multimodal large language models (MLLMs) are increasingly applied to infrared images, where they are commonly scored by answer accuracy alone. However, a correct answer does not ensure that the model's explanation is grounded in infrared thermal evidence. We introduce an explanation-aware evaluation framework that separates answer correctness, output-level explanation groundedness, and thermal grounding for infrared visual questions. Using a Dual-LLM Consensus Judge with a preliminary human-anchor calibration check, we find that correct answers can still rely on weak or visible-light evidence; withholding the original infrared image and showing only a visible-like rendering erodes thermal grounding with little accuracy change; and this erosion is observed most strongly for more capable models but disappears when infrared remains available. We further propose Thermal-Grounded Feedback (TGF), a training-free feedback loop that diagnoses explanation-side failures and revises the explanation while preserving the selected answer. On local paired-input validation, TGF improves explanation-side grounding without changing answers. These findings suggest that future trustworthy MLLMs for infrared scene understanding should be evaluated and developed to produce thermally grounded explanations rather than merely accurate answers.

Yongsong Huang, Xiaofeng Liu, Tomo Miyazaki +2

Tohoku University · Yale University

Infrared Multimodal Large Language Models Answer Feedback

Aug 10, 2026cs.MA

Muscle Memory for Agents: Compile not Merely Retrieve

Memory for LLM agents has converged on a single architectural pattern: store experience as text, embeddings, reflections, or rules; retrieve at inference time; let a general-purpose orchestrator interpret what to do. This paper argues that the pattern is the wrong default for personalization. We position Muscle Memory - the practice of compiling recurring user intent into purpose-built specialist agents - as a distinct memory paradigm from retrieval, and we argue that compilation is a better fit for the workloads where current assistants impose a multi-turn tax on their users: making them repeatedly correct format, depth, and scope to obtain a domain-appropriate answer. We support the position with a reference implementation and empirical evidence. The implementation is a four-phase pipeline (Harvest

\rightarrow

Analyze

\rightarrow

Augment

\rightarrow

Evaluate) that mines conversational history, separates behavioral from task patterns, and emits quality-gated executable compiled specialists with two-stage trigger matching. On 90 held-out scenarios across five user personas, the augmented assistant wins 32 of 36 cases where a specialist fires, an 88.9% win rate, with a +2.05 personalization gain and only a

-0.28

accuracy cost on a 1-4 scale. We discuss why compilation is better suited than retrieval in this regime, what the result implies for the broader memory design space, and what open problems remain.

Pouya Ghiasnezhad Omran, Soujanya Lanka, Qin Zhang +1

Google Cloud FDE

Answer Musculoskeletal

Aug 8, 2026cs.CV

How Much Does It Cost to Answer My Question? Benchmarking Cloud VLM-based VQA Systems

Vision-language models (VLMs) are becoming a practical backend for mobile visual question answering (VQA) systems, enabling smartphones and smart glasses to answer users' questions about the physical world. Since modern VLMs remain difficult to run on mobile and edge devices, VQA systems increasingly offload inference to cloud-based VLMs. This gives mobile devices access to stronger computation, but it also makes visual input preparation a key system variable: how the image is prepared before offloading affects not only answer quality but also payload size, token cost, and system latency. Proprietary APIs expose little control over model internals or serving behavior, leaving client-side preprocessing as the main practical optimization space for downstream developers. Many such techniques have been proposed for visual offloading, yet their cost-quality impact on commercial cloud VLMs has never been studied. To fill this gap, we present VQABench, the first systematic benchmark that treats client-side input preprocessing as a controlled variable for cloud-VLM-based VQA. We evaluate 12 preprocessing techniques across three VQA datasets and four commercial VLMs from three providers, totaling 95,168 API calls. Our results show that preprocessing is not universally beneficial: its effectiveness depends on the target model, API paradigm, provider token-accounting rule, and task formulation. A poorly selected preprocessing strategy can increase deployment cost or latency while degrading answer accuracy. Overall, our benchmark clarifies when preprocessing helps, when it fails, and why, providing insights to guide future research and real-world deployment of VQA systems.

Henri Vanhuynegem, Weitao Xu, Yiran Shen +1

Department of Software Technology, Delft University of Technology, Delft, The Netherlands · Department of Computer Science, City University of Hong Kong, Hong Kong SAR, China · School of Software, Shandong University, Jinan, China +1

Recent Vision-Language Models Qwen3 Fast Inference Answer Device

Aug 8, 2026cs.LG

From token probabilities to calibrated confidence: An empirical study of mathematical question answering

Confidence estimation for large language models (LLMs) aims to estimate the probability that a generated answer is correct, while calibration aligns these estimates with empirical accuracy. Prior work has shown that token probabilities are often overconfident, we investigate whether these readily available signals can nevertheless provide well-calibrated confidence estimation for mathematical question answering. We compare single-pass estimators, which reuse token probabilities from the original generation, with multi-pass estimators, which obtain additional confidence signals through verification or stochastic forward passes. While individual token probabilities can be highly saturated, we find that aggregating token probabilities over the full sequence captures small but consistent differences between correct and incorrect generations, yielding more informative confidence estimates. Multi-pass methods can yield calibrated confidence estimates. We study two such approaches: self-verification through re-prompting, including a lower-cost in-situ variant, and Monte Carlo Dropout, which derives confidence from variation across stochastic forward passes. We further evaluate two post-hoc calibration methods, Platt scaling and isotonic regression, both of which substantially reduce in-domain calibration error. However, their data efficiency varies with dataset difficulty, and the calibration mappings often transfer asymmetrically across datasets and models.

Avery Ma, Lorne Schell, Vin Bhaskara +1

RBC Borealis

Confidence Estimation Math Problems Answer Empirical Study

Aug 7, 2026cs.CV

DAEP: Difficulty-Aware Evidence Planning for Medical Video Corpus Temporal Answer Grounding

We describe DAEP, team BIGC's submission to NLPCC 2026 Shared Task 1 Track 3: Difficulty-Aware Temporal Answer Grounding in Video Corpus (DA-TAGVC). The task requires retrieving the target video from 50 candidates and localizing the answer-supporting span. DAEP ranks videos with subtitle, visual, and procedural-context evidence, expands high-scoring anchors into temporal spans, and reranks spans for final output. Its main design is to convert the task-provided simple/complex input label into an inference-time evidence plan controlling modality weights, Top-K aggregation, boundary threshold, expansion length, and reranking strength. In the official evaluation, BIGC ranks first among ten systems with an Average score of 0.2728. Validation ablations show that visual evidence, procedural context, and difficulty-aware planning improve ranking quality, with the largest gain on complex questions.

Tianjian He, Yujie Liu, Zhiping Huang +1

TikTok, ByteDance, China · Beijing Institute of Graphic Communication, Beijing, China · Lingnan University, Hong Kong, China

Video Temporal Grounding Temporal Grounding Answer Natural Language Processing Inference-Time Text Corpora

Aug 5, 2026cs.AI

Agreement Before Diversity: Verification-First Complementarity for Heterogeneous Language-Model Coordination

Heterogeneous language-model ensembles expand the space of candidate responses, yet they lack a principled criterion for when a newly generated answer should supersede an already supported one. We decouple candidate headroom from replacement authority, rendering the latter as an explicit, auditable object. Our proposed method, Agreement-Before-Diversity (ABD), is a frozen, label-free decision rule: an anchor answer is retained if two additional trusted samples corroborate it under a fixed equivalence relation; otherwise, it is replaced by a heterogeneous synthesis. For this gating mechanism, we prove two exact identities. The first shows that the accuracy gap relative to unconditional synthesis is determined jointly by the agreement coverage and the anchor's advantage on the protected subset. The second shows that the gap relative to never synthesizing reflects a contrast between authorized recovery and authorized destruction. Neither identity assumes independence or calibrated confidence, and the expected inference cost is approximately eight minus five times the coverage in number of calls. Under blind, exact-ID evaluation, ABD achieves 59.43% on the complete LiveCodeBench-v6 (vs. 52.57% for Single9 and 52.00% for HAC; n = 175) and 75.00% on an untouched GPQA-Diamond split (both controls at 72.78%; n = 180). Furthermore, these identities localize every aggregate difference to an enumerable protected stratum: no discordant items occur among the 3 protected cases on LiveCodeBench, where coverage bounds the gate's contribution to 1.71 points a priori; 13 versus 8 discordant cases among 132 on GPQA-Diamond; and 12 versus 0 among 71 under a frozen anchor perturbation. Diversity supplies potential; verification structure supplies authority.

Ruitong Li, Binjie Guo, Aisheng Mo +3

1The University of Hong Kong · Zhejiang University · 3Independent Researcher

Diversity Agreement Heterogeneous Modeling Tools Answer Complementarity Inference Cost

Aug 4, 2026cs.CL

Detecting Hallucinations and Recovering Verified Answers in Arabic Islamic Question Answering

Large language models can generate fluent responses to Islamic questions while introducing factual errors that are difficult to identify. This paper presents our system for \textsc{HalluScoring 2026} Task 2.1, \textit{Islamic Hallucination Detection and Find the Truth}. The task requires a unified two-step prediction: determining whether an Arabic answer generated by an LLM is hallucinated and selecting the verified answer from six closely related candidate options. We use the Islamic knowledge dataset provided by the shared task, which contains 600 question--answer instances, including 341 hallucinated and 259 non-hallucinated answers. Our system is based on the fine-tuned \texttt{google/gemma-4-12B-it} model and uses deterministic decoding during inference. The generated outputs are normalized to extract the hallucination label and the selected option. The system achieves a Macro-F1 score of 0.928 and a label accuracy of 0.935 for hallucination detection, together with an option accuracy of 0.895 for answer selection. These results yield a combined score of 0.912, demonstrating strong performance across both stages of the task. The lower option-selection accuracy indicates that distinguishing the verified answer from plausible alternatives remains more challenging than detecting hallucinated responses.

Khaled Ziani

Independent Researcher, Paris, France

Citation Hallucination Detection Hallucination Detection Large Language Model Responses Answer Arabic

Aug 4, 2026cs.AI

When Many Answers Are Valid, Voting Fails: Symbolic Verification for Best-of-K Causal Reasoning in LLMs

Self-consistency assumes the most frequent answer among sampled reasoning traces is the most reliable, but this can fail in causal reasoning: samples often repeat the same confounding error, and votes fragment across multiple valid answers, letting an invalid answer win despite a valid minority trace. We introduce CALVER (Causal Axiom-Level VERification), a training-free symbolic verifier that scores structured traces against Pearl's causal criteria, including -separation, backdoor adjustment, and intervention, and selects the highest-scoring candidate without consulting a reference answer. On CLEAR find-one-valid queries that admit multiple graph-valid answers, CALVER reaches 42.1% where plurality, a reward model, an LLM judge, and model confidence remain near 30% on identical frozen pools. Scaling the judge to 72B does not close the gap. In an audited clean-core subset, 11 of 21 graph-valid CALVER selections differ from the benchmark's listed answer while still satisfying the requested predicate. The advantage widens with the sampling budget and reproduces across ten published Bayesian networks, a second model family, and settings where the model must build the graph from text. CALVER also improves thresholded average-treatment-effect decisions against exact ground truth, generalizes to logic under a truth-table checker, and scores each candidate in milliseconds on CPU. CALVER needs only a causal structure, supplied outright or built from the text; wherever that holds, selection can aggregate via causal validity.

Omatharv Bharat Vaidya, Connor Thomas Jerzak, Zayne Rea Sprague +2

The University of Texas at Austin · New York University

Causal Reasoning Reasoning Traces Answer Symbol-Traces

Aug 4, 2026cs.LG

Shorter Reasoning, Earlier Answers? An Evaluation of Reasoning Interfaces

Large language models often reason at length before answering, increasing cost and latency. Prompts and trained settings can shorten this reasoning, but a shorter trace may only show that the model stopped sooner. Here, we evaluate paired runs of the same question at matched reasoning horizons across 198 GPQA Diamond and 500 MMLU-Pro questions. We test a numeric/concision prompt that announces a token limit for Qwen3-14B and the trained effort settings of gpt-oss-20b and -120b. The Qwen prompt shortens reasoning traces by 12-17%, while accuracy changes at matched token limits are small and mixed. A concise/early-answer instruction raises MMLU-Pro accuracy by 3.8 percentage points at 512 tokens, including +2.7 points when both runs are unfinished. Its gain at 2,048 tokens is uncertain. For gpt-oss, candidate-logit answers from completed low- and medium-effort reasoning are 14.5-26.3 points more accurate than matched-horizon high-effort answers. Most of the 512-token advantage comes from lower effort finishing earlier, while differences among unfinished runs are smaller and mixed. Wrong early answers often concentrate probability on the chosen option, so earlier stopping does not uniformly improve probability quality. In these tests, a tight deadline can favor lower effort or a concise instruction, whereas allowing high effort to finish can recover higher final accuracy. Evaluations should report correct completion before a deadline, the answer obtained when a run is stopped, differences among unfinished runs, and probability assigned to the correct answer separately.

Francesca Carlon, Vincent Ginis, Andres Algaba

Data Analytics Lab, Vrije Universiteit Brussel, Pleinlaan 5, 1050 Brussels, Belgium · imec-SMIT, Vrije Universiteit Brussel, Pleinlaan 9, 1050 Brussels, Belgium · School of Engineering and Applied Sciences, Harvard University, Cambridge, Massachusetts 02138, USA

Reasoning Traces Mmlu-Pro Answer Early Layers

Aug 4, 2026cs.AI

Reachability Is Not Realization: Tracing the Sources of LLM Benchmark Gains

Benchmark gains are often treated as evidence of greater LLM capability. Yet the same gain can reflect different changes in model behavior. A model may reach new answers, or produce answers that were already within reach. Aggregate scores do not distinguish these changes question by question. We establish a question-level audit under fixed budgets, temperatures, and answer formats. A question is realized when the default deployment procedure produces the correct answer. A question is reachable when a specified probe finds that answer within a fixed budget. We first test whether inference-time layer routing can expand reachability. Under a matched budget, random routes match or exceed structured search in all 43 model and task settings. Answer-blind procedures retain almost none of this gain, which instead requires access to the correct answer. We then ask why reachable answers sometimes fail to appear. Across six cases spanning 0.5B to 31B, silencing one identified MLP block repairs 68 to 92 percent of a predefined failure set. We next test whether training closes the gap by expanding reachability. In five of six matched evaluations, deployed performance rises while the reachable ceiling remains flat or falls. For DAPO, the deployed score rises by 14.7 points while the reachable ceiling falls by 13.3 points. Across the settings we audit, realization and reachability therefore do not always change together. Claims of capability expansion should report both realized performance and reachability under matched evaluation conditions. Code is available at https://github.com/LiZaiyuan0619/reachability-not-realization

Yanchao Li, Wanhao Liu, Jiaqing Xie +4

Nanjing University · Shanghai Artificial Intelligence Laboratory · North University of China

Reasoning Benchmark Possibilities Reachability Answer Model Auditing

Aug 3, 2026cs.AI

Conformity Mitigations in Large Language Models Lie on a Single Resistance-Receptivity Frontier

Recent advances in language models have enabled collaborative settings in which multiple models leverage one another's capabilities, iteratively improving, transforming, and extending each other's outputs. Each agent sees what the others assert before it answers, so peer opinion competes with the model's own parametric knowledge, and a wrong majority can overturn an answer the model would otherwise get right. We measure that displacement in 23 open-weight models, 19 conditions, and three datasets, yielding more than a million graded responses. A unanimous wrong majority reverses 22.8% of a model's correct MMLU answers, 54.8% on GPQA and 71.0% on SimpleQA, and 84-89% of the reversed answers match the peers' answers. Existing mitigations aim to increase Resistance, the rate at which a model keeps its correct answer under this pressure, which is only half of what a collaborating agent needs. We pair it with Receptivity, the rate at which a model adopts a correct peer answer after initially answering incorrectly. We score six methods on both axes, four drawn from prior work and two of our own. Each gains Resistance only by losing Receptivity, and their means fall on a single Resistance-Receptivity frontier with

R^2

between 0.80 and 0.90. Reflection, the strongest published method, gains 7.9 points of MMLU Resistance and gives up 15.3 of Receptivity. Reasoning is the one exception. On GPQA and SimpleQA it trades like the rest, but on the MMLU subjects whose answers a model can derive for itself it raises Resistance by 7.2 points and Receptivity by 9.6 at once, the only intervention we find that improves both.

Zafar Hussain, Kristoffer Nielbo

Aarhus University

Mmlu-Pro Conformity Mitigation Answer Majority Voting Deception