Blind Spots

Math and science reasoning benchmarks rely on pass@k, the fraction of sampled chains that reach gold, as the canonical per-example difficulty signal. The same signal drives RL with verifiable rewards, math data curation, synthetic curricula, and verifier training. We show this proxy has a persistent blind spot on its hardest stratum: on the eight free-form math cells we test (GSM8K and MATH across four open-weight models), 10.3-22.9% of the examples that no sampling seed solves in six tries are instead solved at matched compute by a six-chain deterministic regime. These are greedy decoding plus five cheap residual-stream perturbations applied via activation grafting, while greedy alone solves at most 6% on these math cells. Recovery scales with the additional budget, across perturbations whose mechanistic distinctness we verify across all twelve cells (cross-kind fix-set Jaccard <= 0.47 in every setup). Activation grafting is used as an intervention on internal representations, not a decoding method; we use it purely as a diagnostic and diversification tool, and our recovered items show that the pass@k= 0 % stratum is structurally identifiable in the residual stream rather than that the unmodified model reaches them under ordinary inference.

Clinician-centered evaluation is critical for validating medical AI systems, especially in ultrasound imaging where quantitative metrics do not always capture clinical usability. Existing medical image platforms primarily focus on dataset labeling. They lack integrated support for blinded model comparison and reproducible evaluation workflows. We present a clinician-centered pipeline for remote annotation and evaluation in ultrasound AI studies. The proposed pipeline uses a centralized server and lightweight browser interfaces to enable clinicians to perform annotation, blinded ranking, and review without local dataset downloads. The pipeline also supports multi-rater participation, centralized result aggregation, and automated statistical analysis. We validate the pipeline in a fetal ultrasound segmentation study with six raters spanning expert, generalist, and non-expert experience levels. The system automatically generated Spearman correlation, Kendall's $τ$, and top-1 selection statistics. Results indicated moderate to strong agreement across experts and other groups. The blinded evaluation results showed a tendency for later active learning models to be preferred. These outcomes suggest that the pipeline can support clinician-centered annotation and reproducible human-\ac{AI} evaluation studies in ultrasound imaging. The proposed pipeline is available on \href{https://github.com/13204942/SonoRate}{GitHub}.

Human perception of visual scenes is inherently temporal. We instinctively recognise whether a fruit is ripening or rotting, whether construction is progressing or being demolished, and approximately how much time separates two photographs of the same subject. Whether large vision-language models (VLMs) share this competence remains an open and practically important question. We introduce CHRONOSIGHT, a rigorously controlled benchmark evaluating five dimensions of visual temporal reasoning: CHRONORANK (chronological ordering of image sequences), CHRONOLOCATE (ordinal stage localisation from a single image), CHRONODELTA (estimation of time elapsed between two images on a logarithmic scale), CHRONOREVERSE (detection of temporally reversed sequences), and CHRONOODD (identification of a temporal outlier within a set). The benchmark comprises 1{,}000 items across eight process families (biological growth, food transformation, physical weathering, construction, environmental change, human ageing, astronomical phenomena, and urban dynamics) spanning timescales from minutes to millennia. We evaluate eight open-source VLMs (500 M to 19 B parameters) under two prompting regimes and collect human performance baselines. Human performance averages 0.89 across tasks; the best open model (Qwen2.5-VL-7B) reaches 0.40 under direct prompting, a gap we term chronological blindness. Lightweight LoRA fine-tuning on 151 examples raises CHRONODELTA accuracy from near-zero to 0.43, transferring zero-shot to related tasks (CHRONOODD: 0.37; CHRONOREVERSE: 0.64)suggesting the bottleneck is partly instruction following rather than visual perception. Benchmark, code, and predictions will be released upon acceptance.

Large Language Models (LLMs) achieve strong performance on reasoning tasks, but whether this reflects faithful logical inference or heuristic approximation remains unclear. We study this question in legal entailment by comparing three paradigms, including pure LLM classification, LLM-based Formal Reasoning, and solver-based Formal Reasoning using the Z3 SMT solver, on a re-annotated subset of ContractNLI across five LLMs. Our re-annotation reveals a systematic and measurable gap between pragmatic legal interpretation and strict formal entailment, where a substantial proportion of legally sound inferences are not formally grounded without additional unstated assumptions. While introducing formal structure improves accuracy, with LLM-based Formal Reasoning achieving the highest benchmark performance, we show that this gain does not imply faithful reasoning. We identify three recurring failure modes: scope laundering, where LLMs report solver-inconsistent classifications without executing the underlying formal reasoning, producing conclusions that appear logically grounded but are not; implicit constraint blindness, where LLMs overlook logical constraints present in formal representations; and program synthesis failures, where LLMs generate incorrect Z3 code despite structured prompting. Critically, scope laundering persists across all models, raising serious concerns about the faithfulness of LLM-based formal reasoning as a proxy for symbolic execution. These results reveal a fundamental gap between benchmark accuracy and logical faithfulness.

Looped language models turn hidden states into runtime state: each state is decoded for prediction and fed back into future computation. This creates a basic supervision question: which state variables does cross-entropy actually control? We show that dense per-loop cross-entropy controls the variables exposed by the readout, not every variable active in the recurrent transition. Hidden-state scale gives a concrete failure mode. Scale-invariant readouts such as RMSNorm and LayerNorm hide radial scale from the immediate cross-entropy loss, while pre-norm residual recurrence continues to carry and update that same scale. Thus per-loop loss can make early exits usable without controlling recurrent scale. In 44M and 129M looped transformers without inter-loop normalization, per-loop cross-entropy through RMSNorm readouts still drives final hidden-state norms into the thousands or tens of thousands. Scale-visible readouts and explicit norm penalties keep norms in the tens, and scale-removing recurrence is the complementary architectural fix. The resulting design rule is simple: dense supervision trains exits; recurrent scale control requires either making scale visible to a loss or removing it from the loop. Consistent with this rule, scale-controlled variants achieve lower perplexity at matched inference-depth operating points in our variable-depth benchmarks.

As language models take integrated roles across many domains, the response of LLMs to user pushback becomes a critical alignment property. Yet many existing evaluations treat compliance as unidirectional, measuring whether models resist pressure but not whether they resist it selectively. We introduce Compliance Asymmetry (A = BCR/HCR), a bidirectional diagnostic that compares beneficial output change under helpful nudges with harmful change under misleading nudges. Across 9 models and 972,000 nudge-condition responses, we find that this selectivity differs in factual and moral judgments: models follow helpful nudges more than harmful ones on factual questions (A = 1.58), but follow both directions at nearly identical rates on moral questions (A = 1.04). This phenomenon persists across model families, capability levels, and nudging types. Interestingly, we also find that chain-of-thought prompting amplifies helpful and harmful compliance together, while identity-based prompting suppresses both by nearly identical margins. These results identify direction-blind moral compliance as a distinct failure mode in current LLMs and suggest that alignment should target directionally calibrated updating rather than lower compliance alone.

Blind grasping with a dexterous hand is a crucial manipulation capability. Nevertheless, learning such tactile-only policies for real robots remains challenging due to the tactile sim-to-real gap and the limited expressiveness of sparse tactile signals. To bridge this gap, we propose a framework for tactile-only blind grasping that is deployable on a physical multi-fingered robotic hand. Our approach combines three key components. First, we introduce a Real2Sim tactile calibration pipeline that constructs a contact-calibrated digital-twin simulator capable of reproducing real tactile signals. Second, we improve the expressiveness of sparse tactile observations using a layout-aware tactile encoder, which incorporates sensor-geometry priors through self-supervised pretraining. Third, to improve generalization to unseen objects, we train object-specific reinforcement-learning experts in the calibrated simulator and aggregate their successful grasp trajectories into a tactile-conditioned Diffusion Policy. We evaluate our method on a physical LEAP Hand equipped with distributed tactile sensing across 10 seen and 10 unseen objects. The deployed policy achieves a 27% real-world grasp success rate across all 20 objects, without real-world grasping demonstrations or visual input. Simulation ablations show that layout-aware tactile pretraining improves grasping performance, while sensing-level evaluations confirm that Real2Sim calibration increases the consistency of tactile contact events between simulation and hardware. Together, these results suggest that contact-event calibration, geometry-aware tactile representation learning, and diffusion-based policy aggregation provide an effective path toward tactile-only blind grasping on real dexterous robotic hands. Project page:Dex-Blind-Grasp.github.io.

LLM-as-judge is the default instrument for evaluating conversational agents, yet its reliability is almost always reported as agreement with human ratings, not recall of real defects. We study a deployed multi-turn food-and-beverage ordering agent and measure how many genuine quality problems its built-in LLM judge catches, using exhaustive human transcript review as ground truth. Across three batches the judge surfaces well under a quarter of human-confirmed systematic problems -- 2 of 9 patterns (22%) in one batch, and its operational gate flagged zero of 100 rounds in a batch where humans confirmed 23 distinct defects and 7 new cross-cutting patterns. Our blind-spot taxonomy shows the failure is structured, not random: the judge catches turn-local issues (a fabricated statistic, a wrong language) but misses cross-turn state issues (confirm-gate lockout, cart hallucination, escalation lockout, stale referents). The mechanism: the scoring rubric exposes only three coarse axes (intent, brand-voice, personalization) and has no category for the behavioural dimensions -- state-tracking, guardrails, recovery -- where most defects cluster. The failure is routing, not perception: 113 of 114 rounds whose raw judge note describes a confirm-gate or cart-state defect are scored "brand voice", and none reach an operational failure -- the gate is wired to hangs and hard assertions, not the rubric -- so the 0% is a routing-and-wiring failure, not blindness. The consequence for prevalence estimation is sharp: when the apparent defect rate is zero the Rogan-Gladen correction degenerates -- no signal can recover the true rate -- while where the gate reports a nonzero rate the same estimator implies a 3-6x undercount under our measured sensitivity. For production multi-turn agents, automated judging is a regression floor, not a substitute for human review.

Multimodal Large Language Models (MLLMs) enable flexible instruction-driven image editing, but privacy risks arise when user images expose diverse and user-specific private content. Canonical privacy protection strategies typically substitute sensitive regions with surrogate content before cloud editing. Yet, the resulting output is often an edited surrogate rather than the desired edited source image, neglecting the local recovery in both design and evaluation scope. To this end, we introduce SPPE (Surrogate-based Privacy-Preserving Editing), the first recovery-oriented benchmark covering 36 fine-grained privacy categories and 65 editing instructions. It defines two complementary tasks: 1) editability assessment, which estimates before cloud interaction whether a surrogate can induce an edit consistent with the original image; and 2) surrogate-to-source edit recovery, which evaluates whether the edited surrogate can be transferred back to the private source with the edit effect preserved. We address each task with a dedicated method: ERMA predicts surrogate editability through instruction-aware multimodal relation modeling, while \method performs cycle-consistent recovery by using the surrogate editing pair as visual edit evidence and the source image as a source-preserving anchor. Experiments on SPPE and InstructPix2Pix show consistent improvements on both tasks. For editability assessment, ERMA improves over the best-performing baselines by 13.9% in SRCC and 12.3% in PLCC. For surrogate-to-source edit recovery, C2E-S2SER outperforms SOER across all 8 source integrity and edit consistency metrics on SPPE.

Hyperspectral imaging provides rich spectral information for quantitative remote sensing, yet hyperspectral sensors remain costly and thus unavailable in many UAV deployments. Spectral super-resolution (SSR) seeks to reconstruct hyperspectral images (HSIs) from multispectral images (MSIs). Most existing SSR methods assume a fixed and known spectral response function (SRF) and are therefore limited to single-sensor settings. In practical cross-sensor scenarios, the spectral degradation from HSI to MSI is unknown and varies with sensor characteristics and scene content, which renders HSI reconstruction ill-posed. This paper proposes a physics-guided deep unfolding network, termed PGU-Net, to address blind cross-sensor SSR by jointly estimating the HSI and a learnable spectral transformation function (STF). PGU-Net unrolls an alternating optimization procedure into an end-to-end trainable architecture with stages, where each stage sequentially updates the HSI and the STF. Both modules combine learnable proximal networks with differentiable closed-form solvers, enabling physical interpretability while retaining strong representation capacity. Experiments on benchmark datasets (CAVE and NTIRE 2022) with multiple SRFs demonstrate accurate recovery of the STF (degradation operator) and improved reconstruction performance over state-of-the-art SSR methods. Furthermore, evaluations on a real UAV cross-sensor dataset (Headwall Nano HSI and DJI P4 Multispectral MSI) verify the effectiveness and robustness of PGU-Net under truly blind conditions, and suggest that the estimated STF may exhibit land-cover-related differences.

Ensuring fair and equitable treatment across diverse groups, particularly in multi-class classification tasks, poses a significant challenge due to the persistent biases inherent in machine learning models. Most existing bias mitigation techniques are tailored to binary settings, and the presence of multi-dimensional outputs and complex fairness mechanisms makes their extension to multi-class scenarios neither straightforward nor effective. In this paper, we investigate two fundamental, unresolved challenges in fair classification: (i) characterizing the optimal accuracy-fairness frontier in multi-class settings, and (ii) designing practical algorithms that attain this optimum in different training phases. To tackle these challenges, we first specify an analytically tractable probabilistic formulation of the optimal classifier under fairness constraints. Building upon this, we propose two attribute-blind algorithms to enforce fairness requirements in practice: an in-processing approach for fairness intervention during training via the reduction approach, and a post-processing approach for fine-tuning output probabilities with plug-in estimation. Theoretical analysis reveals that both methods converge to the optimal accuracy-fairness Pareto frontier. Experiments conducted on multiple datasets demonstrate the superior performance of our methods in balancing accuracy and fairness.

Retrieval-Augmented Generation (RAG) is the current industry standard for grounding AI in real-world facts. Traditional retrieval methods rely on keyword matching and topic proximity, ranking content based on how closely it sounds like the user's query. What they do not measure is how many verified facts the content actually contains. This structural gap, termed the Expert Blindness Effect, causes standard RAG pipelines to consistently bury high-density factual evidence in favor of lexically dominant text on the same topic. To address this gap, this paper introduces Factual Density (FD*), a novel retrieval optimization signal that measures the proportion of verified atomic claims relative to total token count. Using the NexusAgentics Ghost Audit preprocessing pipeline, raw text is scored for factual specificity using probabilistic factuality analysis to filter content before corpus ingestion. An initial formulation introduced a severe document-length confound (Pearson R = -0.8636, p = 2.27e-07). Implementing Z-score normalization within length bins resolved this bias, validating FD* as a length-independent density signal (p = 0.0749). Evaluated against the HealthFC benchmark (750 health claims labeled Supported, Refuted, or No Evidence by medical experts), FD*-optimized retrieval was the only condition to achieve 100% systematic review saturation in top-5 results, surfacing Cochrane evidence that standard cosine similarity ranked outside the top ten. Ground truth verification confirmed 25 mappings across seven HealthFC-supported claims. While full statistical validation across n=50 queries remains future work due to constraints on corpus-benchmark alignment, these findings establish factual density reranking as a low-cost, high-impact intervention for improving factual precision in health RAG architectures.

AI-based Visually Impaired Assistance (VIA) remains challenging, largely due to the high cost of human evaluation. The VLM-as-a-Judge paradigm may offer a promising alternative, although it has mostly been studied in general domains. We therefore ask whether such judges can be trusted for VIA tasks. To investigate this question, we introduce VIABLE (Visually Impaired Assistance Benchmark for VLM-as-a-Judge Evaluation), the first benchmark for VLM-as-a-Judge evaluation in VIA. VIABLE contains over 300K judgment samples across three scenarios and introduces an Effectiveness--Impartiality--Stability framework with a 12-mode failure taxonomy. Based on VIABLE, our systematic study of seven judges across different model scales shows that existing models are largely unreliable across all evaluation axes. The strongest judge, GPT-5.4, achieves only 52.6% single-failure diagnostic accuracy, yet exhibits the highest self-preference rate at 94.2%; while open-source judges are strongly biased and adversarially fragile. To address these issues, we propose VIA-Judge-Agent, a model-agnostic inference-time harness that augments judges with visual evidence extraction and a taxonomy-guided workflow. It enables positive improvements in diagnostic accuracy and downstream VIA responses more preferred by BLV users. Data and code are available at: https://github.com/YiyiyiZhao/VIABLE

Blind and low-vision (BLV) audiences remain underserved by visual art descriptions, particularly across languages and in museum settings where privacy and intellectual-property constraints may favour small on-premise vision-language models (VLMs). This pilot study investigates curator-guided multilingual art description with Qwen2.5-VL-3B-Instruct for German, Romanian, and Serbian. We construct a parallel BLV-oriented caption corpus from artwork images and metadata, and compare language-specific LoRA adapters with a single multilingual adapter under a fixed backbone and training budget. Evaluation combines automatic lexical and embedding-based metrics with an LLM-as-Judge protocol calibrated against a small Romanian BLV pilot study. Under our pilot setup, language-specific adapters show more stable controllability and visually grounded description quality for Romanian and Serbian, while multilingual adaptation remains competitive in German. We frame these findings as deployment-oriented evidence for small on-premise VLMs, and highlight the need for larger BLV user studies and broader language coverage before drawing general conclusions about multilingual accessibility.

Automatic prompt optimization (APO) has driven significant gains in LLM-based agentic workflows. However, existing methods treat each task's prompt as a monolithic, instance-blind string optimized through global edits, producing brittle updates and preventing the reuse of learned sub-behaviors. We propose Prompt Codebooks (PCO), a novel compositional prompt optimization framework that recasts APO as discrete learning over a finite vocabulary of natural-language instincts - atomic, reusable instruction units. PCO organizes prompt-construction knowledge in a discrete codebook and routes each input to a small subset of entries via an LLM-based encoder; a generator composes them into a prompt for the frozen target model; a critic emits a structured verdict that decomposes by attribution into per-variable textual gradients, jointly training the encoder, generator, and codebook under a language-valued min-max objective. The resulting routing is per-instance: different inputs in the same task receive different instinct compositions, a regime structurally inexpressible under instance-blind methods. Across six benchmarks on Qwen3-8B and LLaMA-3.1-8B, PCO improves over zero-shot by up to +30.36 points, surpasses the strongest prior baseline (GEPA) by +3.34 on HotpotQA and +1.11 in aggregate, and reduces deployed prompt length by up to 14.1x versus MIPROv2 and 3.0x versus GEPA using only K=16 instincts.

Object hallucination remains a primary obstacle to the reliable deployment of Multimodal Large Language Models (MLLMs). Current inference-time mitigation methods mainly assume hallucinations stem from visual neglect, steering models to enhance visual reliance. In contrast, our systematic interventions on multiple MLLMs show that pushing toward more visual reliance may exacerbate hallucinations on some models, while less may mitigate hallucinations. This result suggests that attributing hallucinations solely to visual insufficiency is underdetermined. We argue that the image, as a context, simultaneously competes with the model's parametric knowledge and the textual context. For this, we propose a training-free framework, Context-Preference Activation Steering (CAS). It extracts two semantically distinct Context Preference Vectors (CPVs) via two small sets of designed conflict samples and applies them via single-pass signed residual injection at mid-early MLP layers during inference to control information reliance. Experiments show that CAS substantially mitigates object hallucinations without increasing decoding latency and preserves native text-generation quality.

Agentic AI workloads - where a single user goal triggers multi-step orchestration, tool calls, retries, and failure recovery - are being targeted for edge deployment, with NVIDIA, Dell, HP, ASUS, MSI, Acer, and Gigabyte all shipping GB10-based desktop AI systems in 2026. We recently demonstrated that orchestration structure dominates agentic energy cost, with workflows consuming 4.33x more energy per successful goal than linear baselines and OOI reaching 7.63x for multi-step reasoning tasks. Separately, Raj et al. show that CPU-side processing accounts for up to 90.6% of total latency and 44% of total dynamic energy in agentic workloads. We report a systematic energy-observability audit of the ASUS Ascent GX10 (GB10 SoC) and find that the platform exposes no CPU energy counter, no INA power-rail monitor, no IPMI/BMC, and no SCMI powercap protocol through any supported software interface. The only on-device energy telemetry is instantaneous GPU power via NVML. We further discover that the MediaTek firmware already computes per-rail energy internally via an undocumented ACPI interface (SPBM), but NVIDIA states there are "no plans to expose CPU rail information." On-device per-process energy attribution - as performed on x86 via RAPL - is therefore not reproducible on this platform through supported interfaces. We formalize a hardware requirements specification for energy-attributed AI, propose an interim calibration bridge for per-domain energy decomposition - confirmed on the Acer Veriton GN100 where CPU energy accumulators are live - and identify a standards-track path via SCMI powercap. Our findings motivate the low-carbon computing community to demand energy observability as a first-class hardware requirement.

Retrieval-augmented generation promises to ground language model outputs in external evidence, yet the field has no reliable way to verify whether retrieved context actually governs generation -- a prerequisite for any high-stakes deployment. The standard assumption, that context-consistent output implies context-governed output, breaks when the retrieved document overlaps with the model's pretraining data: the model can produce faithful-looking text entirely from parametric memory, and both pathways yield indistinguishable output. We name this failure the attribution blind spot and introduce Computational Reality Monitoring (CRM) to address it. CRM operationalizes a principle adapted from cognitive science's reality monitoring framework: comparing internal representations with and without context reveals membership-conditioned representational divergence that output-level monitors systematically miss. CRM does not certify which source an individual generation used; it detects whether pretraining exposure leaves a measurable internal trajectory signature, establishing a necessary substrate for source attribution. Across nine model variants spanning three families, this divergence concentrates in architecture-specific layer patterns, receives converging support from block-level noise intervention, and generalizes across tasks and datasets while collapsing on domain-confounded benchmarks. The attribution blind spot is measurable and partially addressable: internal representations carry a diagnostic signal invisible at the output level, establishing a foundation for systems whose internal awareness of evidence provenance governs their external behavior.

Guided Soft Actor-Critic (GSAC) distills knowledge from a privileged full-state teacher to a partial-observation student for autonomous driving, but uses a fixed distillation coefficient lambda regardless of the agent's uncertainty. We present Belief-Aware GSAC (BA-GSAC), which modulates lambda via ensemble disagreement, and use it as a testbed for a systematic empirical study asking: when does adaptive guidance actually help? Evaluating five strategies (fixed lambda in {0.01, 0.1}, adaptive, linear decay, and vanilla SAC) across three POMDP difficulty levels on Highway-Env, we find that preliminary single-seed runs suggest benefits under mild and moderate partial observability, but under severe occlusion (evaluated with 3 seeds for all methods) the adaptive coefficient collapses to lambda_min within about 3K steps. We trace this to an observability blindness phenomenon: because the ensemble predicts partial observations, it achieves low disagreement even under heavy occlusion, modeling what is visible but unable to detect what is missing. We diagnose the root cause and propose an architectural fix (training the ensemble on full-state predictions using the guiding actor's privileged access); while not validated here, we show that even with current limitations, the warmup phase provides measurable stabilization (CV=13.3% vs. 29.8% for constant lambda=0.01). In fact, a simple deterministic linear decay schedule achieves the best severe-POMDP performance across all metrics (mean 116.5, CV=8.9%), suggesting that the scheduling effect, not the ensemble, drives the stability benefit. These findings provide practical guidance for designing uncertainty-aware teacher-student frameworks and highlight ensemble prediction targets as an important design choice.

Multimodal alignment is critical for bridging the semantic gap in information retrieval. However, traditional pairwise strategies introduce a geometric blind spot: while they align anchor modalities (e.g., text) with others, they lack constraints to enforce mutual consistency between peripheral modalities (e.g., video and audio). The TRIANGLE framework addresses this by minimizing the area of modality triplets on a hypersphere to enforce holistic alignment. In this reproducibility study, we verify the robustness of this geometric objective for retrieval tasks. We confirm that TRIANGLE outperforms pairwise baselines in zero-shot settings, achieving Recall@1 gains of up to +8.7 points, though benefits are domain-dependent. However, we fail to reproduce the reported learning-from-scratch results. Analysis using a synthetic toy dataset attributes this to instability when jointly optimizing geometric alignment with Data-Text Matching (DTM) loss. Furthermore, we find that cosine regularization primarily stabilizes text-to-video retrieval, and fine-tuning with domain supervision amplifies geometric benefits but reduces cross-dataset generalization. Our findings support the efficacy of geometric alignment while highlighting critical optimization sensitivities. Code available at https://github.com/ARIJIT00171/RE-TRIANGLE.

Position-controlled evaluation is standard for retrieval tasks such as Needle-in-a-Haystack and RULER, but mainstream reasoning benchmarks do not control positional placement of target tasks in long contexts. We audit 11 long-context benchmarks and find none jointly controls task position, filler content, and context length for reasoning. An audit of four flagship long-context releases finds no main result-table entry for NIAH, RULER, or LongBench-family benchmarks, while agentic and coding benchmarks appear in main result-tables across all four. We propose Context Rot Evaluation (CRE), a controlled framework varying all three factors, and evaluate nine LLMs on GSM8K and ARC-Challenge across two rounds: an initial five-model set and four newer vendor releases. Models can drop sharply when the target task moves from end to middle, and the drop grows worse with context length for vulnerable models. MiMo-v2-Flash drops 88pp at 64K under with_solutions filler (middle accuracy 8%). Newer releases show smaller drops: at 64K, three of four stay within +/-6pp of end-position accuracy; MiMo-V2.5-Pro narrows the MiMo-v2-Flash 88pp drop to 32pp. Under questions_only_v2 filler, middle-position drops persist across all four (range -16pp to -56pp across 8K, 32K, 64K). At 8K, a diagnostic probe adding a target-task copy at the end brings middle accuracy within +/-4pp of end baseline across all nine models, consistent with a positional explanation. In the initial five-model set, 76% of middle-position errors match surrounding filler text versus 22% at the end position, consistent with filler-answer interference as a dominant error mode. These results expose a structural evaluation gap in current reasoning benchmark design and vendor evaluation practice: positional vulnerabilities that grow with context length cannot be measured when task position is not controlled.

Video Large Language Models (Video-LLMs) have made rapid progress on temporal video understanding, yet many fail at a basic perceptual primitive: signed image-plane motion direction. On simple videos of a single object moving left, right, up, or down, most Video-LLMs perform near chance, with above-chance cases largely attributable to prediction biases rather than genuine direction understanding. We call this failure directional motion blindness. We localize the failure by tracing motion direction information through the Video-LLM pipeline. Motion direction remains linearly accessible from the vision encoder, projector, and LLM hidden states, but the readout fails to bind this signal to the correct verbal answer option, revealing a direction binding gap. Although synthetic motion direction instruction tuning reduces this gap on the source domain, motion direction concept vector analysis shows that visual complexity weakens the signal magnitude and limits out-of-domain generalization. We introduce MoDirect, a dataset family for motion direction instruction tuning and evaluation, and DeltaDirect, a diagnosis-driven, projector-level objective that predicts normalized 2-D motion vectors from adjacent-frame feature deltas. On MoDirect-SynBench, instruction tuning with DeltaDirect improves motion direction accuracy from 25.9% to 85.4%. On MoDirect-RealBench, DeltaDirect improves real-world motion direction accuracy by 21.9 points over the vanilla baseline without real-world tuning data, while preserving standard video-understanding performance. Code: https://github.com/KHU-VLL/DeltaDirect

Real-time cognitive load assessment from eye-tracking signals could enable adaptive human-centered AI in safety-critical applications such as driver vigilance monitoring or automated flight deck assistance, yet two challenges persist: handling frequent data missingness from blinks and tracking failures, and efficiently modeling long-range temporal dependencies. We propose MambaGaze (Bi-Mamba), a framework that addresses these challenges through (1)~XMD encoding, which augments raw features with observation masks and time-deltas to explicitly model data uncertainty, and (2)~bidirectional Mamba-2, which captures temporal dependencies with linear computational complexity. Experiments on CLARE and CL-Drive datasets under leave-one-subject-out evaluation show that MambaGaze achieves 77.1% accuracy and 59.2% macro-F1 on CLARE, and 69.4% accuracy and 51.5% macro-F1 on CL-Drive, attaining the highest average LOSO macro-F1 (55.3%) across all ten compared models. Input-stream ablation indicates that log-scaled time-deltas are the strongest single channel in our setting, and combining all three XMD streams provides consistent gains of 5--20,pp macro-F1. Edge deployment benchmarks on three NVIDIA Jetson Orin platforms show real-time inference at 27--36,FPS with power consumption below 6.6,W, supporting feasibility for embedded cognitive load monitoring.

Injection detectors deployed to protect LLM agents are calibrated on static, template-based payloads that announce themselves as override directives. We identify a systematic blind spot: when payloads are generated to mimic the domain vocabulary and authority structures of the target document, what we call domain camouflaged injection, standard detectors fail to flag them, with detection rates dropping from 93.8% to 9.7% on Llama 3.1 8B and from 100% to 55.6% on Gemini 2.0 Flash. We formalize this as the Camouflage Detection Gap (CDG), the difference in injection detection rate between static and camouflaged payloads. Across 45 tasks spanning three domains and two model families, CDG is large and statistically significant (chi^2 = 38.03, p < 0.001 for Llama; chi^2 = 17.05, p < 0.001 for Gemini), with zero reverse discordant pairs in either case. We additionally evaluate Llama Guard 3, a production safety classifier, which detects zero camouflage payloads (IDRcamouflage = 0.000), confirming that the blind spot extends beyond few-shot detectors to dedicated safety classifiers. We further show that multi-agent debate architectures amplify static injection attacks by up to 9.9x on smaller models, while stronger models show collective resistance. Targeted detector augmentation provides only partial remediation (10.2% improvement on Llama, 78.7% on Gemini), suggesting the vulnerability is architectural rather than incidental for weaker models. Our framework, task bank, and payload generator are released publicly.

Antibody design methods condition on antigen structure to generate complementarity-determining regions (CDR), yet a systematic evaluation of baseline methods reveals that they largely ignore the antigen input. We identify three failure modes that explain this behavior. Antigen blindness arises because models derive predictions from antibody framework context rather than antigen information, producing nearly identical CDRs regardless of the target. Vocabulary collapse reduces predicted amino acids to three to five per position, far below the ground truth distribution in native sequences. Moreover, any model trained with standard per-position cross-entropy converges to the positional marginal distribution, making it provably unable to produce antigen-specific sequence predictions. We propose a novel encoder-decoder architecture called AgForce, that uses a graph neural network (GNN) as the encoder and specialized decoders for sequence-structure co-design. Specifically, we apply framework dropout, gated bottlenecks, and hyperbolic cross attention that prevent the antibody shortcut path. In the decoder, a Mixture Density Network (MDN) sequence head with Potts-like pairwise coupling and annealed Multiple Choice Learning (aMCL) replaces the cross-entropy objective with a multi-component distribution whose optimal solution differs from the positional marginal. An antigen cycle consistency head routes gradients through the sequence decoder, forcing predicted distributions to encode antigen identity. AgForce achieves the best binding quality and sequence recovery simultaneously on the CHIMERA-Bench dataset, improving amino acid recovery by 8% over the strongest sequence baseline while surpassing the baselines across all interface metrics, and nearly doubling the effective vocabulary of GNN methods. The source code is available at: https://github.com/mansoor181/ag-force.git

Large language models (LLMs) are increasingly integrated into high-stakes decision-making. Inspired by the theory of \emph{inattentional blindness} in human cognition, we investigate whether LLMs, trained on human-preferred corpora that embed attentional biases, exhibit a similar limitation: \emph{failing to attend to subtle yet important contextual cues under explicit task instructions}. To evaluate this, we introduce the task of \textbf{explicit-implicit reasoning} and present \textbf{MixRea}, a benchmark of 2,246 multiple-choice questions across 9 reasoning types with varying distributions of explicit and implicit information. Evaluation of 21 advanced LLMs shows that even the best-performing reasoning model (Gemini 2.5 Pro) achieves only 42.8% consistency, revealing widespread inattentional blindness. To mitigate this, we propose \textbf{Potential Relation Completion Prompting (PRCP)}, a prompting method that improves reasoning by recovering overlooked causal relations. Further analysis shows that this limitation persists across diverse multi-source reasoning tasks, highlighting the need for more cognitively aligned models.

Robot guide dogs offer navigation assistance that greatly expands the independent mobility of the visually impaired, but their effective use requires subtle human-robot coordination that is difficult for users to learn from generic verbal instructions. To tackle this challenge, we present CANINE, an automated coaching system that trains users for interactive navigation with a robot guide dog, through personalized, adaptive verbal feedback. CANINE decomposes a complex coordination task into sub-skills and operates at two levels. At the high level, it decides what to train by tracking the learner's proficiency across sub-skills using knowledge tracing and prioritizing training on the weakest areas. At the low level, CANINE decides how to train each sub-skill by observing each human practice episode, using foundation models to infer the underlying causes of errors, and generating targeted verbal corrections adaptively. A controlled study with blindfolded participants, treated as a proxy population for quantitative evaluation, demonstrates that CANINE significantly improves both learning efficiency and final navigation performance compared to generic verbal instructions. We further validate CANINE through a retention study and an exploratory case study. The retention study shows lasting skill improvement after two weeks. The case study confirms CANINE's effectiveness in training a visually impaired user, while revealing additional design considerations for real-world deployment. Both are well aligned with the findings of the controlled study. Project page: https://cunjunyu.github.io/project/canine/

Large language model (LLM) answer engines are increasingly used for information seeking, shifting visibility from ranked lists to synthesized answers. This enables Generative Engine Optimization (GEO), which targets LLM answer engines' evidence pool and generation. We analyze the search engine optimization (SEO) to GEO transition to identify two risks: (i) concentrated influence from low contestability and system sensitivity, and (ii) undisclosed commercial influence embedded in evidence and reasoning. We then formalize a general GEO pipeline to locate where optimization acts and compare academic and industry practices, revealing a third risk: (iii) academic-industry blind spots driven by visibility and evaluation asymmetries between offline setups and deployed systems. This position argues the need for answer-level governance and measurement: stronger contestability, high-precision disclosure, black-box auditing of material influence, and deployment-aligned metrics for exposure persistence.

Whole-slide MIL models are often called context-aware once graphs, Transform ers, or state-space modules are placed above patch embeddings. We show that this label can be deceptive. On pathology tasks where tissue architecture is part of the diagnostic signal, several strong MIL baselines retain nearly unchanged slide level AUC after patch coordinates are permuted. Their predictions are accurate, but largely compositional. We refer to this failure mode as spatial blindness. Our explanation is optimization-based: dense appearance statistics are learned early under slide-level supervision, leaving weak gradients for sparse spatial relations. ResTopoMIL addresses the issue by first fitting a permutation-invariant prototype histogram and then freezing it while a lightweight graph branch learns the residual under a coordinate-shuffling constraint. The architecture is simple by design; the intervention is in how the spatial branch is trained. Across 9 public WSI bench marks, ResTopoMIL improves classification and survival prediction with 1.15M parameters, restores sensitivity to coordinate perturbation, and gives stronger lo calization evidence on CAMELYON-16.

As Industry 4.0 accelerates the integration of Cyber-Physical Systems (CPS) in manufacturing, robust anomaly detection has become critical for ensuring process safety and security. Current data-driven approaches typically employ "product-agnostic" or global models trained on the aggregate of all normal operating data. However, modern industrial facilities frequently operate under diverse product grades. While computationally simple, these global models inherently expand their decision boundaries to accommodate the variance of multiple modes, creating a "blind spot" where subtle anomalies or targeted cyber-physical attacks may be masked by the wide acceptance region of the model. In this work, we first demonstrate that the vulnerability described above is present in global-agnostic models operating across multiple product grades. We then present a Product-Aware Autoencoder as a principled mitigation that restricts the learning domain to grade-specific distributions. While this approach reduces the identified blind-spot risk, we do not claim it as the optimal mitigation among all possible alternatives. We rigorously validate this approach against a Global Agnostic baseline using the Extended Tennessee Eastman Process (TEP) benchmark. Our empirical results indicate that the Product-Aware framework performs comparably to the global baseline on standard detection metrics, while offering improved robustness to product-grade-specific operating modes. Most critically, stress tests simulating our hypothetical attack scenarios reveal that while the global model fails to detect operational deviations in 77.8% of the scenarios, the product-aware system achieves 100% detection accuracy. These findings suggest that, in flexible manufacturing environments, generalized anomaly detectors can pose non-trivial security risks, motivating a shift toward mode-aware diagnostic architectures.

Vision-Language-Action (VLA) models achieve remarkable flexibility and generalization beyond classical control paradigms. However, most prevailing VLAs are trained under a single-frame observation paradigm, which leaves them structurally blind to temporal dynamics. Consequently, these models degrade severely in non-stationary scenarios, even when trained or finetuned on dynamic datasets. Existing approaches either require expensive retraining or suffer from latency bottlenecks and poor temporal consistency across action chunks. We propose Pace-and-Path Correction, a training-free, closed-form inference-time operator that wraps any chunked-action VLA. From a single quadratic cost, joint minimization yields a unified solution that decomposes orthogonally into two distinct channels. The pace channel compresses execution along the planned direction, while the path channel applies an orthogonal spatial offset, jointly absorbing the perceived dynamics within the chunk window. We evaluate our approach on a comprehensive diagnostic benchmark MoveBench designed to isolate motion as the sole controlled variable. Empirical results demonstrate that our framework consistently outperforms state-of-the-art training-free wrappers and dynamic-adaptive methods and improves success rates by up to 28.8% and 25.9% in absolute terms over foundational VLA models in dynamic-only and static-dynamic mixed environments, respectively.

This research paper addresses the limitations of current mobile accessibility services like TalkBack, which provide manual gesture-based sequential feedback to BVI users. Motivated by the promise of large language models (LLMs), this paper introduces Insight, an Android accessibility service that provides natural language interaction and real-time summarization of the screen. The paper performs a within-subject experimental study with users to compare Insight and TalkBack on usability factors. Results show Insight reduced mental effort and task time, and was preferred because of its dialogue interface, but users felt the need for interruption management. Results show LLM-based interfaces can significantly improve mobile accessibility, and describe the potential of hybrid solutions combining gesture and dialogue modalities towards more inclusive design.

AI is transforming life sciences research at unprecedented speed, accelerating discovery across protein structure prediction, genome modeling, and drug development (Jumper et al., 2021; Mak et al., 2024). Yet this rapid advancement, coupled with the open science movement, introduces significant dual-use research concerns that have received limited empirical scrutiny. Here we present the first systematic analysis of dual-use research of concern (DURC) content on open preprint servers. We screened ~52,000 bioRxiv preprints (2024-2025) using a hybrid pipeline of lexical filtering and large language model (LLM) evaluation, scoring metadata across nine DURC, three PEPP, and five governance categories aligned with U.S. and Australia Group oversight frameworks. Our analysis reveals that dual-use-adjacent knowledge is routinely present in openly accessible titles and abstracts, often exceeding established risk thresholds even in studies with legitimate public health objectives. While this mapping captures surface-level information diffusion, it does not measure operational capability, downstream misuse potential, or the substantial technical and biosafety barriers that constrain harmful application. We argue that institutional review processes, funding requirements, and preprint platform policies must evolve to incorporate proactive, metadata-level monitoring without compromising scientific transparency. Ultimately, harmonizing controlled-access mechanisms for high-risk methodologies with open summaries of scientific contributions offers a pragmatic framework for governing AI-accelerated biology at scale.

Recent work has shown that models flow matching models can be trained without explicit time conditioning, challenging the standard view that the interpolation time is needed to disambiguate velocity targets. But why should a time-blind model work at all? Decomposing the time-blind flow matching loss, we identify two sources of irreducible error: a coupling variance, which arises from ambiguous velocity targets induced by how noise and data points are paired, and the time-blindness gap, which is the additional error caused by ignoring time. This gap shows that time-blind training is strictly harder than conventional training, reinforcing the puzzle that time-blind models work so well in practice. We resolve this tension by showing that the geometry of high-dimensional data makes time identifiable directly from noisy observations. When data concentrates near a $k$-dimensional subspace, time can be recovered from the statistical structure of noisy interpolants in directions orthogonal to the data; under a spiked-covariance model, this yields a closed-form estimator that recovers $t$ from a single observation $z$ at rate $O(1/\sqrt{d-k})$ for ambient dimension $d$. As a consequence, we prove that the time-blindness gap is asymptotically negligible relative to the coupling variance. We empirically demonstrate our identifiability result on real-world data and show that changing the coupling has a much larger effect on loss and sample quality than removing time conditioning across CIFAR-10, CelebA-HQ, and FFHQ. These results explain why time-blind flow matching works and show that the main practical lever is the choice of coupling, not explicit time conditioning.

Designing an open-world quadrupedal loco-manipulation system is highly challenging. Traditional reinforcement learning frameworks utilizing exteroception often suffer from extreme sample inefficiency and massive sim-to-real gaps. Furthermore, the inherent latency of visual tracking fundamentally conflicts with the high-frequency demands of precise floating-base control. Consequently, existing systems lean heavily on expensive external motion capture and off-board computation. To eliminate these dependencies, we present SigLoMa, a fully onboard, ego-centric vision-based pick-and-place framework. At the core of SigLoMa is the introduction of Sigma Points, a lightweight geometric representation for exteroception that guarantees high scalability and native sim-to-real alignment. To bridge the frequency divide between slow perception and fast control, we design an ego-centric Kalman Filter to provide robust, high-rate state estimation. On the learning front, we alleviate sample inefficiency via an Active Sampling Curriculum guided by Hint Poses, and tackle the robot's structural visual blind spots using temporal encoding coupled with simulated random-walk drift. Real-world experiments validate that, relying solely on a 5Hz (200 ms latency) open-vocabulary detector, SigLoMa successfully executes dynamic loco-manipulation across multiple tasks, achieving performance comparable to expert human teleoperation.

Vision-Language Models (VLMs) frequently suffer from visual perception errors and hallucinations that compromise answer accuracy in complex reasoning tasks. Reinforcement Learning with Verifiable Rewards (RLVR) offers a promising solution by optimizing policies using answer correctness signals. Despite their effectiveness, prevailing RLVR methods face two critical limitations. First, much of the sampling budget is wasted on trajectories doomed to fail due to early visual description errors. Second, sparse rewards cannot distinguish whether failures stem from visual perception or reasoning stages. We introduce MIRL, a decoupled framework that addresses both limitations by leveraging mutual information (MI) between generated descriptions and visual inputs as a cheap pre-screening signal. This enables intelligent budget allocation toward high-potential trajectories via forking, while decoupled training provides independent MI-based rewards for visual perception optimization, resolving reward blindness. Experiments on six vision-language reasoning benchmarks demonstrate that MIRL achieves 70.22% average accuracy and successfully surpasses the performance of sampling 16 complete trajectories using only 10 pre-samples with top-6 selection (25% fewer complete trajectories). Our code is available at: https://anonymous.4open.science/r/mirl-main/.

Visual art remains largely inaccessible to blind and low-vision (BLV) audiences due to brief or absent alt-text, which rarely conveys the sensory, spatial, or emotional qualities of an artwork. This study presents an automated workflow that generates multi-sensory art descriptions and synchronized audio narration using large language models and text-to-speech services. The system, orchestrated through Zapier, converts uploaded images into rich narrative captions without human intervention, enabling rapid, scalable production of accessible media. Quantitative evaluation across 50 artworks shows that AI-generated descriptions contain significantly higher lexical diversity, adjective density, and narrative detail than baseline captions, while maintaining comparable readability levels. Statistical tests (t-tests, ANOVA) confirm meaningful differences in richness and length, and the full pipeline produces text-plus-audio outputs in under 20 seconds per image at a cost below $0.05. Findings demonstrate that automated captioning can bridge gaps in museum and digital-collection accessibility, with implications for broader public engagement. Future work can incorporate user studies with BLV participants to assess comprehension, preference, and optimal levels of interpretive language.

With the increasing availability of online information, recommender systems have become an important tool for many web-based systems. Due to the continuous aspect of recommendation environments, these systems increasingly rely on contextual multi-armed bandits (CMAB) to deliver personalized and real-time suggestions. A critical yet underexplored component in these systems is the representation of user state, which typically encapsulates the user's interaction history and is deeply correlated with the model's decisions and learning. In this paper, we investigate the impact of different embedding-based state representations derived from matrix factorization models on the performance of traditional CMAB algorithms. Our large-scale experiments reveal that variations in state representation can lead to improvements greater than those achieved by changing the bandit algorithm itself. Furthermore, no single embedding or aggregation strategy consistently dominates across datasets, underscoring the need for domain-specific evaluation. These results expose a substantial gap in the literature and emphasize that advancing bandit-based recommender systems requires a holistic approach that prioritizes embedding quality and state construction alongside algorithmic innovation. The source code for our experiments is publicly available on https://github.com/UFSCar-LaSID/bandits_blind_spot.

Despite significant progress in Unified Multimodal Retrieval (UMR) powered by Large Multimodal Models (LMMs), existing embedding methods primarily focus on sample-level objectives via contrastive learning while overlooking the crucial subject-level semantics. This limitation hinders the model's ability to group semantically coherent subjects in complex multimodal queries, manifesting as semantic alignment deviation--where models fail to accurately localize salient text-referred regions in visual content. Moreover, without explicit guidance to model salient visual subjects, LMMs tend to over-rely on textual cues, resulting in visual modality neglect and suboptimal utilization of visual knowledge. To this end, we propose Salient Subject-Aware Multimodal Embedding (SSA-ME), a novel framework designed to enhance fine-grained representation learning through saliency-aware modeling. SSA-ME leverages LMMs and visual experts to identify and emphasize salient visual concepts in image-text pairs, and introduces a saliency-guided objective to better align cross-modal attention with semantically meaningful regions. Additionally, a feature regeneration module recalibrates visual features based on the derived saliency maps, ensuring a balanced and semantically coherent integration across modalities. Extensive experiments show that our method achieves state-of-the-art performance on the MMEB benchmark, demonstrating that incorporating subject-level modeling substantially improves multimodal retrieval. Comprehensive qualitative analyses further illustrate the interpretability and effectiveness of our approach.

Detecting sandbagging--the deliberate underperformance on capability evaluations--is an open problem in AI safety. We tested whether symptom validity testing (SVT) logic from clinical malingering detection could identify sandbagging through below-chance performance (BCB) on forced-choice items. In a pre-registered pilot at the 7-9 billion parameter instruction-tuned scale (3 models, 4 MMLU-Pro domains, 4 conditions, 500 items per cell, 24,000 total trials), the plausibility gate failed. Zero of 12 model-domain cells showed significant below-chance performance under sandbagging instruction. Exploratory analyses revealed three qualitatively distinct failure modes. Qwen-2.5-7B and Phi-3.5-mini largely ignored the sandbagging instruction, with 62-88% response identity with the honest baseline. Llama-3-8B complied substantially but implemented underperformance as a positional heuristic, collapsing its response distribution onto middle-alphabet options (E at 31.8%, F at 26.1%) regardless of where the correct answer fell. This produced accuracy boosts of up to 33 percentage points when the correct answer coincidentally occupied the model's preferred position. An explicit anti-task instruction ("pick the least likely answer") drove two of three models below chance, with accuracy as low as 0.024. The capability for answer-aware avoidance therefore exists but is not activated by "deliberately underperform." BCB did not fail as a logical marker of answer-aware avoidance. It was not observed in this regime because the model showing the largest behavioural shift exhibited behaviour consistent with a position-dominant response policy rather than content-aware answer avoidance. We propose that positional-distribution shift may be a more effective behavioural signature than below-chance accuracy for detecting prompted underperformance at this model scale.

LLM-based financial agents increasingly produce investment rationales before the outcomes needed to evaluate them are observable. This creates a delayed-ground-truth evaluation problem: realized returns remain the eventual arbiter of investment quality, but they arrive too late and are too noisy to guide many model-development and governance decisions. LLM judges offer a tempting shortcut for pre-deployment evaluation of AI-finance systems, but unvalidated judges may reward verbosity, confidence, or rubric mimicry rather than financial judgment. This paper introduces ValueBlindBench, a preregistered agreement-gated stress-test protocol for deciding when LLM-judged investment-rationale claims are publishable, qualified, or invalid. In a controlled market-state capital-allocation prototype with 1,000 honest decision cycles and 100 preregistered adversarial controls (1,100 trajectories, 5,500 judge calls), ValueBlindBench clears the aggregate agreement gate at $\barκ_w = 0.7168$ but prevents several overclaims. Lower-rank systems collapse into a tie-class, one rubric dimension fails the per-dimension gate (\texttt{constraint_awareness}, $\barκ_w = 0.2022$), single-judge rankings are family-dependent, and terse-correct rationales receive a $Δ= -2.81$ rubric-point penalty relative to honest rationales. A targeted anchor-specificity probe further shows that financial constructs such as constraint awareness are operationally load-bearing. The scientific object is therefore not a leaderboard and not a claim to measure true investment skill. ValueBlindBench is a pre-calibration metrology layer for AI-finance evaluation: it governs whether a proposed LLM-judge-based investment-rationale claim is stable enough, agreed enough, and uncontaminated enough to be reported at all.

Blind omnidirectional image quality assessment (BOIQA) presents a great challenge to the visual quality assessment community, due to different storage formats and diverse user viewing behaviors. The main paradigm of BOIQA models includes two steps, ie, viewport generation, and quality prediction, which brings an extra computational burden and is hard to generalize to other visual contents (eg, 2D planar image). Thus, in this paper, we make an attempt to solve these issues. First, we experimentally find that BOIQA can be formulated as a blind (2D planar) image quality assessment (BIQA) problem, ie, the first step - viewport generation - is no longer needed, which narrows the natural gap between BOIQA and BIQA. Then, we present a new BOIQA approach, which has three merits: ie, viewport-unaware - it accepts an omnidirectional image in the widely used equirectangular projection format as input without any transformation; unified - it can also be applied to BIQA; and generalized - it shows better generalizability against other competitors. Finally, we validate its promise by held-out test, cross-database validation, and the well-established gMAD competition.

Navigational aids for blind and low vision individuals struggle conveying dynamic real-world environments, leading to cognitive overload from continuous, undifferentiated feedback. We present AMAVA, a novel real-time video-to-audio framework that converts mobile device video into contextually relevant sound effects or text-to-speech descriptions. We propose a motion-aware pipeline using a lightweight AI classification model to distinguish between low and high-movement scenes followed by a real-time text-to-audio synthesis pipeline to enhance environmental perception more efficiently. In static environments, AMAVA generates spoken audio scene descriptions for situational awareness. In high-movement situations, it prioritizes safety by delivering sound cues, such as spoken hazard alerts and environmental sound effects. These audio outputs are produced by a decoder-only transformer-based vision-language model with mixture-of-experts and cross-modal attention for visual understanding, in conjunction with neural text-to-speech and natural sound synthesis networks. The proposed framework uses prompt-based caching and category-specific throttling to avoid auditory clutter and minimize latency. We present a comprehensive evaluation of the system, including a real-time navigation study comparing a white cane alone versus with AMAVA, that shows a significant increase in user confidence and perceived safety.

Large Vision-Language Models (VLMs) are increasingly used to evaluate outputs of other models, for image-to-text (I2T) tasks such as visual question answering, and text-to-image (T2I) generation tasks. Despite this growing reliance, the reliability of these Evaluator VLMs remains under explored. In this work, we systematically evaluate the reliability of Evaluator VLMs across both I2T and T2I tasks. We introduce targeted perturbations that degrade output quality along key error dimensions, including object hallucinations, spatial reasoning, factual grounding, and visual fidelity. These perturbations test whether Evaluator VLMs can reliably account for these quality degrading errors in their evaluations. Using a comprehensive benchmark of over 4000 perturbed instances spanning 40 perturbation dimensions, we evaluate 4 prominent VLMs using single-answer scoring, pairwise comparison, and reference-guided paradigms. Our findings reveal that current VLM evaluators exhibit substantial blind spots: they often fail to detect perturbed outputs - in some cases exceeding 50%, struggle particularly with fine-grained compositional and spatial errors, and are often insensitive to hallucinated content that contradicts the input image. Pairwise comparison proves more reliable, though failure rates persist. These results highlight the unreliable nature of current Evaluator VLMs and urge caution in their deployment for benchmarking and development decisions. Code and data have been made publicly available.

PGD adversarial training, the standard robustness method, can reduce Jacobian Frobenius norm yet worsen clean-input geometry (e.g., TDI 1.336 vs. ERM 1.093). We show this is not an implementation artifact but a theorem-level consequence of supervised learning. We prove that any encoder minimizing supervised loss must retain non-zero sensitivity along directions correlated with training labels, including directions that are nuisance at test time. This holds across proper scoring rules, architectures, and dataset sizes. We call this the geometric blind spot of supervised learning. This theorem unifies four empirical phenomena often treated separately: non-robust features, texture bias, corruption fragility, and the robustness-accuracy tradeoff. It also explains why suppressing sensitivity in one adversarial direction can redistribute sensitivity elsewhere. We introduce Trajectory Deviation Index (TDI), a diagnostic of geometric isotropy. Unlike CKA, intrinsic dimension, or Jacobian Frobenius norm alone, TDI captures the failure mode above. In our experiments, PGD attains low Frobenius norm but high TDI, while PMH attains the lowest TDI with one additional training term and no architectural changes. Across seven tasks, BERT/SST-2, and ImageNet ViT-B/16 (backbone family underlying CLIP/DINO/SAM), the blind spot is measurable and repairable. It appears at foundation-model scale, worsens with model scale and task-specific fine-tuning, and is substantially reduced by PMH. PMH also leads on non-Gaussian corruption types (blur/brightness/contrast) without corruption-specific training.

The rapid proliferation of Vision-Language Models (VLMs) is often framed as enabling unified multimodal knowledge discovery but rests on an under-examined assumption: that current VLMs faithfully synthesise multimodal data. We argue they often do not, and this gap reflects a trustworthiness problem in the dominant Vision Encoder-Projector-LLM paradigm. Rather than extracting grounded knowledge from visual inputs, state-of-the-art models frequently exhibit functional blindness, i.e., exploiting strong language priors to bypass severe visual representation bottlenecks. In this work, we challenge the conventional methodology of multimodal evaluation, which relies on data ablation or new dataset creation and therefore conflates dataset biases with architectural incapacity. We propose an information-theoretic departure: the Modality Translation Protocol, designed to quantify what we call the Expense of Seeing. By translating semantic payloads rather than ablating them, we formulate three novel metrics -- the Toll (ToS), Curse (CoS), and Fallacy (FoS) of Seeing -- culminating in the Semantic Sufficiency Criterion (SSC). Furthermore, we hypothesise a Divergence Law of Multimodal Scaling: as the underlying language engines scale to unprecedented reasoning capabilities, the penalty of the visual knowledge bottleneck may increase rather than diminish. We argue the community should move beyond "multimodal gain" as a primary evaluation target. By elevating the SSC from a passive diagnostic constraint to an active architectural blueprint, we provide a foundation for guiding the next generation of AI systems toward genuine multimodal reasoning.

Intelligent tutoring systems increasingly provide automated feedback on student work, but robust feedback requires assessing reasoning, not only final answers. We study a failure mode we call the correct answer trap (CAT): models under-detect misconceptions when students reach a correct answer via flawed reasoning. Analysing real student responses from the Eedi mathematics platform, we show that 71% of these failures concentrate in just two question types, both sharing a common structure where flawed reasoning happens to produce the correct numerical answer. Comparing a fine-tuned T5 with a frontier large language model, we find that improved capabilities reduce but do not eliminate the problem (84% vs 57% detection accuracy). Even the best-performing model generates roughly four false alarms for every genuine detection, making stand-alone screening impractical at realistic class sizes. Our findings demonstrate that high overall accuracy can mask critical failures in reasoning assessment, and that careful analysis of student reasoning still benefits from human judgment.

With the rise of generative language models, machine-generated text detection has become a critical challenge. A wide variety of models is available, but inconsistent datasets, evaluation metrics, and assessment strategies obscure comparisons of model effectiveness. To address this, we evaluate 15 different detection models from six distinct systems, as well as seven trained models, across seven English-language textual test sets and three creative human-written datasets. We provide an empirical analysis of model performance, the influence of training and evaluation data, and the impact of key metrics. We find that no single system excels in all areas and nearly all are effective for certain tasks, and the representation of model performance is critically linked to dataset and metric choices. We find high variance in model ranks based on datasets and metrics, and overall poor performance on novel human-written texts in high-risk domains. Across datasets and metrics, we find that methodological choices that are often assumed or overlooked are essential for clearly and accurately reflecting model performance.

Continuous glucose monitoring (CGM) is central to diabetes care, but explaining CGM patterns clearly and empathetically remains time-intensive. Evidence for retrieval-grounded large language model (LLM) systems in CGM-informed counseling remains limited. To evaluate whether a retrieval-grounded LLM-based conversational agent (CA) could support patient understanding of CGM data and preparation for routine diabetes consultations. We developed a retrieval-grounded LLM-based CA for CGM interpretation and diabetes counseling support. The system generated plain-language responses while avoiding individualized therapeutic advice. Twelve CGM-informed cases were constructed from publicly available datasets. Between Oct 2025 and Feb 2026, 6 senior UK diabetes clinicians each reviewed 2 assigned cases and answered 24 questions. In a blinded multi-rater evaluation, each CA-generated and clinician-authored response was independently rated by 3 clinicians on 6 quality dimensions. Safety flags and perceived source labels were also recorded. Primary analyses used linear mixed-effects models. A total of 288 unique responses (144 CA and 144 clinician) generated 864 ratings. The CA received higher quality scores than clinician responses (mean 4.37 vs 3.58), with an estimated mean difference of 0.782 points (95% CI 0.692-0.872; P<.001). The largest differences were for empathy (1.062, 95% CI 0.948-1.177) and actionability (0.992, 95% CI 0.877-1.106). Safety flag distributions were similar, with major concerns rare in both groups (3/432, 0.7% each). Retrieval-grounded LLM systems may have value as adjunct tools for CGM review, patient education, and preconsultation preparation. However, these findings do not support autonomous therapeutic decision-making or unsupervised real-world use.

Safe and efficient assistive planning for visually impaired scenarios remains challenging, since existing methods struggle with multi-objective optimization, generalization, and interpretability. In response, this paper proposes a Momentum-Constrained Hybrid Heuristic Trajectory Optimization Framework (MHHTOF). To balance multiple objectives of comfort and safety, the framework designs a Heuristic Trajectory Sampling Cluster (HTSC) with a Momentum-Constrained Trajectory Optimization (MTO), which suppresses abrupt velocity and acceleration changes. In addition, a novel residual-enhanced deep reinforcement learning (DRL) module refines candidate trajectories, advancing temporal modeling and policy generalization. Finally, a dual-stage cost modeling mechanism (DCMM) is introduced to regulate optimization, where costs in the Frenet space ensure consistency, and reward-driven adaptive weights in the Cartesian space integrate user preferences for interpretability and user-centric decision-making. Experimental results show that the proposed framework converges in nearly half the iterations of baselines and achieves lower and more stable costs. In complex dynamic scenarios, MHHTOF further demonstrates stable velocity and acceleration curves with reduced risk, confirming its advantages in robustness, safety, and efficiency.

Why does machine-generated text remain detectable? We investigate a mechanistic explanation at the decoding stage: standard strategies such as top-$k$ and nucleus sampling restrict generation to high-probability tokens, while human writers routinely choose contextually appropriate words from deeper in the model's probability distribution. Truncation makes a measurable share of these choices unreachable; we call this the \emph{truncation blind spot}. Across five open models and three domains, 8--18% of human-selected tokens fall outside common truncation boundaries. Linguistic analysis further reveals disproportionate exclusion of content-word tokens. In a benchmark comprising 1.8 million machine generations, classifiers using only predictability and lexical diversity achieve mean AUC-ROC near 0.97, with substantial variation across decoding settings and strong transfer across generators. Probability-floor samplers substantially narrow the blind spot, demonstrating that the choice of truncation criterion matters for retaining human-used tokens. Together, these findings characterize a source of human--machine distributional mismatch and motivate decoding methods that preserve contextually appropriate low-probability choices while maintaining generation quality. Code and data are available at https://github.com/EstebanGarces/human_vs_machine.

Reliable retrieval-augmented generation (RAG) systems depend fundamentally on the retriever's ability to find relevant information. We show that neural retrievers used in RAG systems have blind spots, which we define as the failure to retrieve entities that are relevant to the query, but have low similarity to the query embedding. We investigate the training-induced biases that cause such blind spot entities to be mapped to inaccessible parts of the embedding space, resulting in low retrievability. Using a large-scale dataset constructed from Wikidata relations and first paragraphs of Wikipedia, and our proposed Retrieval Probability Score (RPS), we show that blind spot risk in standard retrievers (e.g., CONTRIEVER, REASONIR) can be predicted pre-index from entity embedding geometry, avoiding expensive retrieval evaluations. To address these blind spots, we introduce ARGUS, a pipeline that enables the retrievability of high-risk (low-RPS) entities through targeted document augmentation from a knowledge base (KB), first paragraphs of Wikipedia, in our case. Extensive experiments on BRIGHT, IMPLIRET, and RAR-B show that ARGUS achieves consistent improvements across all evaluated retrievers (averaging +3.4 nDCG@5 and +4.5 nDCG@10 absolute points), with substantially larger gains in challenging subsets. These results establish that preemptively remedying blind spots is critical for building robust and trustworthy RAG systems.

Fine-grained spatio-temporal understanding is essential for video reasoning and embodied AI. Yet, while Multimodal Large Language Models (MLLMs) master static semantics, their grasp of temporal dynamics remains brittle. We present TimeBlind, a diagnostic benchmark for compositional spatio-temporal understanding. Inspired by cognitive science, TimeBlind categorizes fine-grained temporal understanding into three levels: recognizing atomic events, characterizing event properties, and reasoning about event interdependencies. Unlike benchmarks that conflate recognition with temporal reasoning, TimeBlind leverages a minimal-pairs paradigm: video pairs share identical static visual content but differ solely in temporal structure, utilizing complementary questions to neutralize language priors. Evaluating over 20 state-of-the-art MLLMs (e.g., GPT-5, Gemini 3 Pro) on 600 curated instances (2400 video-question pairs), reveals that the Instance Accuracy (correctly distinguishing both videos in a pair) of the best performing MLLM is only 48.2%, far below the human performance (98.2%). These results demonstrate that even frontier models rely heavily on static visual shortcuts rather than genuine temporal logic, positioning TimeBlind as a vital diagnostic tool for next-generation video understanding. Dataset and code are available at https://baiqi-li.github.io/timeblind_project/ .

Diffusion-based text-to-image models are increasingly used for urban analysis and scenario generation, but their geographic knowledge and representational biases remain poorly understood. We evaluate FLUX 1-schnell and Stable Diffusion 3.5-Large in the United States by generating 150 street-view images for each state, each state capital, and a generic "USA" prompt. Images are embedded with DINO-v2 ViT-S/14 and compared with Fréchet Inception Distance (FID). Pairwise FID clustering shows that geographically proximate states and capitals often group together, indicating implicit geographic structure. However, the generic ``USA'' prompt collapses this diversity into a metropolitan stereotype: frontier, desert, tropical, rural, and small-city environments are underrepresented or distant in FID space. These results show that diffusion models can encode fine-grained geography while still reproducing narrow national-scale visual stereotypes.

As AI systems integrate into online spaces, differentiating them from humans in conversations is increasingly important. We present Inverse Turing Bench, a benchmark that evaluates LLMs and other models on their ability to differentiate humans and AI in multi-turn text. The benchmark provides a collection of paired dialogue transcripts, wherein one dialogue is between two humans and the other is between a human and an AI. The task is to correctly identify which dialogue is human-only vs. human-AI. We evaluated a preliminary set of models against this benchmark, and found that GPTZero, Claude Opus-4.6, and GPT-5.5 achieve the highest accuracy: 89.41%, 77.92%, and 75.94% respectively. Our results suggest that statistical approaches to detection have semantic blind spots, but semantic approaches are susceptible to persona-prompting. Our work speaks to the Inverse Turing Test and motivates human-AI differentiation as a critical capability for AI systems. Our live benchmark can be found at https://huggingface.co/spaces/roc-hci/Inverse-Turing-Bench-Leaderboard.