Papers

Physical events are not understood by their names alone, but by the causal state changes that compose them. A clip-level label such as "bounce" can be correct while hiding the process that makes the event physically valid, from support loss and contact onset to rebound and settling. To make this hidden process explicit, we introduce Atomic Physical Transitions (APTs): minimal, temporally localized state changes that bind a visible cue to an active physical mechanism and before/after dynamical regimes. An APT chain represents a video as an ordered causal transition sequence rather than a single aggregate event label: event labels tell what happened; APT chains explain why it happened. To make APTs learnable by VLMs, we construct mixed-source APT data from human annotations and simulator ground truth, covering 14 transition types across contact, gravity, friction, and rotation/stability, with 27,303 timed instances over 1,246 trials. Using this data, we find that current VLMs miss transition-level physics, with zero-shot recall at most 14% and errors dominated by missed transitions. Direct fine-tuning on APT chains improves transition detection but causes event-level forgetting, indicating that the model learns a specialized answer format rather than a reusable physical representation. We therefore propose APT-Tune, a parameter-efficient recipe that teaches VLMs to use causal transitions without forgetting how to answer video questions. It combines image-pad-aware supervision, format-conditional co-training, and mechanism-conditioned domain-to-type decoding to make APT learning format-robust and physically grounded. With only 11 M LoRA parameters on Qwen3-VL-2B, APT-Tune substantially improves APT recall while also improving event-level video transfer. These results show that APTs are not a new answer format, but a human-aligned causal supervision signal for physical video understanding.

Crowd counting is a fundamental task in computer vision. However, crowd counting in low-light environments remains largely underexplored, despite its practical importance in the real world. Existing methods mainly focus on well-lit scenes or rely on single-modality Red-Green-Blue (RGB) representations, which often become unreliable under extreme darkness and complex non-uniform illumination. To handle this problem, we construct three new low-light crowd counting benchmarks, which consist of two synthetic datasets, SHA\_Dark and SHB\_Dark, and a real-world benchmark LC-Crowd (Low-light Crowd Dataset). Inspired by Retinex-based physical modeling, we introduce depth and Canny edge cues as complementary geometric and structural priors to enhance the intrinsic reflectance representation under low-light conditions. We propose a Multi-Modal Hyper-Graph Fusion module, which formulates RGB appearance, depth geometry, and edge structure cues as nodes in a unified hyper-graph and explicitly captures their high-order complementary relationships via dynamic hyperedge construction and message passing. Furthermore, to adaptively allocate computation in dense prediction, we propose a Deformable Rectangular Sparse Attention (DRSA) module, which concentrates computation on informative regions through anchor-aware estimation and adaptive rectangular window modeling. Based on these designs, we develop a unified Low-Light Counting Network (LCNet) for robust low-light crowd counting. Extensive experiments on three benchmarks demonstrate that the proposed method achieves the best overall performance against existing state-of-the-art (SOTA) methods. The code is in the supplementary material. The datasets will be made public upon acceptance.

The quality of recorded data depends on the stability of the sensor system that acquires it. Sensor motion and aging can degrade the performance and stability of downstream data-driven methods. We present a Wasserstein-GAN-inspired approach for unsupervised inference of physically interpretable transformation parameters that map a changed detector response distribution back to a nominal reference distribution. In contrast to standard generative modeling, the generator is used as a learnable calibration transformation whose trainable weights represent the sought parameters, while the critic provides a distributional distance signal via the Wasserstein objective. We validate the approach on a tracking-detector toy model with controlled layer shifts and demonstrate its application on high-granularity Geant4-simulated calorimeter data with cell-wise aging effects. The method recovers aging coefficients for individual cells with correlation to ground truth and improves agreement between calibrated and reference energy-sum distributions, while exhibiting the expected degradation at increasing channel-to-channel noise levels. These results indicate that adversarial distribution matching can serve as a data-driven component of calibration strategies in settings where direct labels for degradation parameters are unavailable.

A theory can accommodate a counterexample by adding an exception. The harder problem is deciding how far the exception should extend and which predictions should survive. We present DeFAb, a dataset and symbolic verification framework built from curated biological knowledge in the Gene Ontology. Models fill gaps in a theory or propose exceptions to its rules. A verifier checks the consequences of each edit, allowing different solutions to receive credit. Simple symbolic repairs solve all 336 tasks in the released evaluation panel. This result exposes a limitation of the revision test. An exception tied to one entity can succeed without identifying a pattern that applies to other cases. We develop a contextual task that requires an exception to reach the relevant cases and preserve predictions elsewhere. Two planned studies compare training with verified and rejected examples, and revision with and without feedback about consequences. DeFAb makes the scope of a proposed explanation testable, providing a controlled step toward studying theory revision in scientific reasoning.

Biomedical NLP pipelines routinely presuppose clean input text, yet large-scale corpora assembled through automated PDF parsing harbour pervasive OCR-like artifacts, token splits and merges, hyphenation remnants, and character-level corruption, that systematically erode lexical evidence and degrade downstream classifiers. We introduce a conservative, fully auditable spell-correction reliability layer conceived as a safety-oriented preprocessing module rather than a maximal-accuracy corrector: under conditions of uncertainty, the system abstains from editing, in accordance with a medical do-no-harm philosophy. The deterministic architecture couples bounded edit-distance candidate generation with corpus-derived n-gram scoring and a suite of biomedical safety gates that protect domain-critical terminology. We evaluate the layer both intrinsically, on a manually curated benchmark of 2,104 token-level cases, and extrinsically, on a tri-class CORD-19 topic classifier (Prevention, Treatment, Epidemiology) spanning 10,000 examples under a principled four-run protocol (Clean, Noisy, Restored, Safety). Intrinsically, the layer attains 94.61% error-fix recall on synthetic errors with zero harmful edits on negative controls. Downstream, it recovers approximately 80.45% of the noise-induced macro-F1 degradation, elevating macro-F1 from 0.7654 (Noisy) to 0.7717 (Restored) while preserving near-clean performance (Safety: 0.7721). A supplementary case study on 103 real-world OCR-extracted abstracts classified with BioBERT confirms that transformer encoders appeared relatively robust to mild noise, motivating a future grey-box architecture that integrates bounded neural signals and UMLS lexicons without compromising auditability. The system is fully deterministic, artifact-driven, and designed with deployment and auditability in mind.

In 2011, IBM's Watson was something like a sealed capsule of its era's queryable knowledge. Its DeepQA system defeated the strongest human Jeopardy! champions, but the knowledge that let it do so lived in a curated billion-document corpus running on a cluster of POWER7 servers, frozen at build time and impossible to move or copy. We show that the same kind of artifact, a snapshot of what a culture can answer, is now portable and essentially free. We evaluate a single 9 GB open-weight model (Qwen2.5-14B, 4-bit) against the complete open Jeopardy! clue dataset, 529,939 clues across all 41 broadcast seasons from 1984 to 2025. To our knowledge this is the first time a model has been run over the full corpus. The 41 years mark only how long the questions were collected. What they test is far older and broader: the accumulated body of human general knowledge a culture considers worth knowing, from ancient history and dead languages to science, literature, and geography, with a verified answer for every item. The model answers 67.0% of all clues under a strict forced-response protocol with exact and fuzzy matching, and exceeds 85% on factoid categories. We treat training-data exposure as something both systems share rather than a flaw unique to language models. Watson's case is in fact the more extreme one. Its corpus was assembled to contain Jeopardy answers and it was tuned on past clues, and it could not answer anything outside that curated distribution. The decisive test is whether a model can answer clues that did not exist when it was built. On clues aired after its training cutoff, the local model holds 65% and Claude Opus 4.8 holds 95%, while Watson by construction scores zero. The capability survives the move from a server room to a file you could seal in a time capsule, and unlike Watson it is not frozen to its own moment.

Computational mental health (CMH) classifiers often degrade under distribution shift because human annotators and distant-supervision pipelines reward different linguistic signals. We introduce TSS (Triple-Stream Stress probe), a multi-channel diagnostic framework that decomposes text into (A) lexical character n-grams, (B) a small, mostly content-free morpho-syntactic channel, and (C) a 154-feature psycholinguistic style channel. Across four English datasets (N=12,906), TSS reveals a lexical interference effect: adding lexical features to the style channel reduces Macro-F1 on human-labeled data (mean drop 0.072, p<10^-4) but not on auto-labeled data. We propose Degree of Divergence (DoD), a difference-in-differences statistic adapted from econometrics for label-source auditing, with instance-level bootstrap inference; the headline estimate is DoD(BC-A) = 0.0374, 95% CI [0.0097, 0.0651], p=0.0032. A platform-stratified Twitter-only DoD (which removes the Reddit vs. Twitter contrast) reproduces the pattern with bootstrap inference: DoD-Tw(BC-A) = +0.096 (p<0.001) and DoD-Tw(AC-A) = -0.089 (p<0.001). Interventional masking (pos_only) retains ~95-99% of Channel C's performance after destroying content words on human datasets, indicating that the style channel does not rely primarily on lexical surface form. TSS is positioned as a diagnostic audit framework, not a clinical screening tool: it flags label-source-specific shortcut learning before generalization claims are made.

Traditional industrial agents rely on modular pipelines, including Router, Retriever, Planner, Executor, Responder, Reviewer, and other components. These systems often fracture into a labyrinth of ad-hoc patches, leading to cascading errors and high latency. We propose OneModel, an applicable paradigm shift from external workflows to internalized knowledge representation. Unlike modular systems that slice fluid user intents into static steps, OneModel consolidates complex business logic and SOPs directly into the model parameters. Through Continual Pre-training (CPT) and logic-compilation SFT, we transform fragmented business rules into intuitive model reasoning within a unified attention space. Deployed in our global financial service system, OneModel effectively breaks the trade-off between latency, accuracy, and complexity. Online A/B testing demonstrates an end-to-end latency reduction of more than 50 percent, from 18.7 seconds to 8.0 seconds, while the Intelligent Resolution Rate (IRR) increases from 64.3 percent to 83.3 percent. The results show that OneModel can replace brittle engineering logic with internalized cognitive intuition, offering a scalable blueprint for transitioning industrial agents from complex, error-prone workflows to unified model architectures.

Large Language Models (LLMs) exhibit extreme sensitivity to surface-level prompt variations, in which minor lexical changes can trigger disproportionate performance fluctuations. Moving beyond black-box optimization and coarse-grained templates, we present the first large-scale, n-gram token-level mechanistic analysis of prompt stability, leveraging a dataset of 132,000 prompt variants. Our investigation reveals a fundamental Scaling Law of Prompt Performance Stability: higher average task performance is strongly associated with lower variance and greater robustness across prompt perturbation. We identify two core linguistic drivers underlying this robustness: (1) Domain-Specific Terminology, which tightly anchors semantic boundaries, and (2) Explicit Action Directives, which formalize reasoning trajectories. Together, these elements constrain the model's interpretative space, effectively “locking in” more deterministic generation behavior. Building on these insights, we introduce an automated Prompt-Refining Agent that systematically restructures input queries by injecting domain anchoring and operational constraints. Empirical evaluation shows that our approach reduces performance variance by 40.7% in code generation task, while preserving or improving mean performance. These findings provide a statistically grounded and mechanistically interpretable framework for achieving robust prompt engineering.

Long-term memory remains a critical bottleneck for conversational AI agents, whose finite context windows cannot support coherent recall across thousands of turns. We present AgentMemBench, a unified, reproducible benchmark evaluating five memory management strategies under identical conditions: in-context windowing (ICW), external key-value store (EKV), graph-based episodic memory (GEM), compression-based summarisation (CBS), and web-augmented memory (WAM). All are assessed across three public datasets covering long-term multi-session dialogue (LoCoMo), task-oriented document grounding (MultiDoc2Dial), and persona-grounded multi-session chat (MSC), using Recall@k, MRR, nDCG@k, Answer F1, an LLM-judge Faithfulness score, Memory Footprint, and Latency over 491 annotated question turns. Generation and judging both use Qwen2.5-7B-Instruct (4-bit), with greedy decoding for determinism. Our results show that (1) EKV dominates on every quality axis (macro Recall@5 0.792, MRR 0.677, F1 0.156, Faithfulness 0.354); (2) long-range recall is decisive: on LoCoMo, where the gold turn lies many sessions back, ICW, WAM, GEM, and CBS retrieve almost nothing (Recall@5 <= 0.005) while EKV alone reaches 0.573, showing that recency windows, summaries, and entity graphs collapse at long horizons and only dense retrieval scales; (3) CBS is the runner-up on retrieval (0.556); (4) WAM equals ICW on in-corpus recall by construction, since external results carry no in-corpus provenance; and (5) EKV's recall advantage carries a footprint cost (~5,100 vs ~300 tokens for ICW/WAM), an explicit accuracy-efficiency trade-off. We additionally evaluate two published memory systems (MemGPT/Letta, HippoRAG) against the same harness, and release all code, environment, and result artefacts for full reproducibility.

Depression treatment often requires switching medications due to inadequate response or adverse effects. Estimating individualized treatment effects in this setting is challenging because treatment assignment is confounded by patient characteristics, switching induces time-varying selection, and counterfactual outcomes are not observed in follow-up data. Using a proprietary longitudinal major depressive disorder (MDD) clinical trial dataset, we formulate a next-visit counterfactual prediction task to estimate Hamilton Depression Rating Scale (HAMD-17) total scores under alternative treatments. We benchmark 8 estimators, including meta-learners, residual-based methods, and tree-based approaches. Causal Forest (CF) demonstrates the most favorable and consistent performance across all criteria. Our analysis shows that symptom benefits concentrate in specific switch directions, with dose intensification being generally beneficial. Notably, we identify a counterintuitive exception where a lower-intensity regimen outperforms a higher-intensity alternative for specific patient subsets. While crude observational comparisons substantially overstate gains, confounding-adjusted estimates yield modest, actionable magnitudes. These findings provide prospectively testable candidates for clinical decision support in depression care.

Background: This work investigates the presence of implicit bias in Large Language Model (LLM)-based chat AI models directed toward people with intellectual disabilities (ID). Objective: The study aims to identify and measure representational differences related to people with ID and examine them to identify implicit biases inherent in AI chat generation technologies. Methods: Utilizing the GPT-4-Turbo model, we requested story-generation based on 10 prompt stems with and without descriptors for ID. This process was repeated using four other LLMs (OpenAI GPT-4o, Meta Llama-3-3-70B-Instruct, Anthropic Claude-3-5-Sonnet, and Mistral-Large-2411). The resulting 25,000 computer-generated stories were analyzed using a separate GPT-4-Turbo model instance to detect differences in how people are represented related to themes of bias described in previous literature. Results: Our findings reveal differences in how people are represented between story datasets with and without ID descriptors. These differences go beyond established characteristics of ID and imply the presence of mostly negative implicit biases. Identified differences related to considering people with ID as younger, with themes of paternalism and infantilization; depicting them as more inspirational and symbolic; as needing help more often, being dependent, and being saved; and having a negative perception of them and more hesitation to include them. Conclusions: These implicit biases are considered within the context of past discrimination towards people with ID and highlight the need for diligence against implicit bias towards people with ID in AI development. This research underscores the importance of assessing and mitigating implicit bias in decision-making technologies to prevent future societal harm.

Large language models (LLMs) augmented with tools are emerging as autonomous agents capable of using Web engine, APIs, and code to solve complex, long-horizon tasks. Current tool-using benchmarks for information seeking on academic graphs rely on synthetic templates, simplified solution spaces, or narrow tasks such as paper-centric tasks, leaving key challenges underexplored - realistic user intent, complex multi-step API planning, rich parameter filling for APIs, grounded answers with references, and comprehensive evaluation of both the process and the outcome. We introduce AISE-Bench, a real-world, full-cycle annotated benchmark for information seeking on academic knowledge graphs. AISE-Bench release contains 1,133 QA pairs, including query taxonomies, full API execution trajectories, validated parameters, and source-grounded answers with reference links. To support high-quality annotation, we design a customized agent workflow to enable annotators to plan, execute, and revise complex API workflows efficiently. We develop a comprehensive evaluation protocol measuring answer quality, reference grounding, API-planning correctness, and execution success. Among the 14 evaluated methods, even the strongest model (PLAY2PROMPT with Gemini-3-Pro) achieves only moderate performance and often struggles with API planning and execution. AISE-Bench establishes a challenging new testbed for quantitatively evaluating and improving the stepwise correctness, grounded summarization, and traceable reasoning of multi-step API-using LLM agents. Our code and data are available at https://aise-bench.github.io/.

Collecting reliable social data from low-literacy populations remains a persistent challenge, particularly when surveys involve sensitive topics and marginalized communities. Traditional paper-based and web-based survey modalities often suffer from high attrition and incomplete responses due to literacy barriers, social pressure, and interactional discomfort. In this paper, we present findings from an initial field evaluation comparing multiple survey modalities paper-based interviews, digital web-based surveys, conversational AI (convAI) surveys, and convAI enhanced with layered value-sensitive design conducted with low-literacy women across India. Using data from 315 participants, we show that convAI significantly improves survey completion rates relative to traditional modalities, with the highest completion and lowest drop-off observed when value-sensitive and culturally aligned conversational design elements are fully integrated. These results demonstrate the importance of human-centered and value-sensitive interaction design in enabling inclusive, ethical, and scalable data collection; motivating more `AI for social good' applications.

Medical AI has shifted from reasoning to agentic AI, a new paradigm that autonomously invokes external tools during reasoning, rendering intermediate reasoning steps and tool outputs transparent to users. Although proven to outperform previous models, physician trust in agentic AI remains largely unexplored. To address this, three physicians evaluated 315 multimodal clinical cases quantifying both process-oriented cognitive trust and outcome-oriented behavioral reliance. Comparing agentic AI against non-agentic baselines, physicians exhibited significantly higher cognitive and behavioral trust for the agentic model (P < 0.001). Specifically, on treatment planning tasks, physicians trusted the agentic reasoning most, preferring it in 89.57% of cases. Furthermore, process-oriented cognitive trust is significantly associated with outcome-oriented behavioral reliance (P < 0.001). However, measurable over-reliance on incorrect agentic outputs still exists, highlighting the inherent limitations of decision-logic transparency alone and underscoring the continuous need for rigorous clinician oversight.

Surgical outcomes depend not only on patient factors and postoperative care but are also strongly influenced by the quality of the operation itself. Yet, for much of mod-ern surgery, intraoperative quality has been assessed indirectly through outcomes and operative reports. The increase in minimally invasive procedures inherently guided by endoscopic video, together with advances in artificial intelligence, creates an unprecedented opportunity to systematically observe, measure, and improve surgi-cal care. This chapter introduces AI-enabled Surgical Quality Assurance as a frame-work for using surgical data to support continuous assessment and improvement in the operating room. We first review existing approaches to surgical safety, from sys-tem-level interventions to procedure-specific standards. We then describe how AI can transform intraoperative video into clinically meaningful information, including recog-nition of anatomy, instruments, workflow, surgical actions, quality criteria, adverse events, and critical moments. Finally, we outline the major challenges that must be addressed before these systems can deliver routine clinical value, including representa-tive data collection, robust validation, workflow integration, regulation, liability, pri-vacy, and equitable access. Rather than replacing surgical judgment, AI for quality assurance should be understood as a set of tools for augmenting the surgical team, scaling expert review, and helping surgery evolve toward a learning system in which intraoperative care is continuously observed, assessed, and improved.

Artificial Intelligence (AI) has the potential to be transformative for development, but Africa is currently facing a fragmented and challenging "AI divide". This paper provides an empirical analysis of the current state of the AI landscape and how it compares with Africa's technological preparedness for the future. In our analysis, we approach the "AI Divide" from three angles: infrastructure, accessibility, and human capacity. First, we look at the physical constraints that prevent Africa from integrating digitally. We then evaluate the human-centred factors that limit the development of AI technology on the continent. Finally, we examine the human capacity to develop AI systems on the continent and provide three focused case studies. Our investigation shows that the physical infrastructure needed to build an AI economy on the continent is lagging, with only 38% internet penetration, poor broadband coverage and less than 1% of all data centres globally. Other constraints include high data costs relative to income, gender-based digital divides, and the need to build more representative NLP models that can understand Africa's native languages. However, there are positive trends towards the emergence of local initiatives and grassroots movements, such as startups and universities, contributing to AI development on the continent. Based on these findings, we provide concrete recommendations to policymakers to help develop a more comprehensive and equitable AI ecosystem on the African continent.

AI-native course assessments in senior computer science courses and related fields should grade students by AI-resilient skill: the ability to achieve outcomes beyond a strong AI baseline. Such assessments should allow students to use AI freely, while reducing the extent to which greater private AI budget or more intensive AI use, by itself, becomes a grading advantage. This paper proposes a minimal formal framework for this goal. The framework specifies a real task, an executable evaluator, a declared AI-native Pareto frontier, and a grading rule based on Pareto surplus. The central claim is simple: Pareto surplus provides a measurable, protocol-relative certificate that a submitted artifact achieves a tradeoff not already supplied by the declared AI baseline, and grading by this surplus is AI-resilient with respect to that baseline. Interpreting surplus as evidence of student skill requires the surrounding assessment protocol–for example, design reports, ablations, prompt traces, oral checks, or reproducibility explanations–but the grading certificate itself is behavioral and executable. The framework is then extended to practical complications, including self-improving AI loops, budget neutrality, server-mediated feedback, and prompt-based red teaming. As a concrete instantiation, we describe an AI-resilient approximate-membership assignment centered on Bloom filters for COMP 480/580 at Rice University, designed to test whether students can improve beyond AI-generated implementations.

Large language models are increasingly deployed in agentic pipelines that depend on the model evaluating its own outputs without external verification. The reliability of these pipelines depends on an implicit assumption: that the model applies relevant concepts the same way when it generates an output and later evaluates that output. We propose a new measure, generator-evaluator self-consistency, to test this assumption directly and apply it to 10 frontier models across 491 concepts. We find, first, that there is substantial variation in self-consistency. Second, we find that in a clinical setting with physician-validated mistakes (Proniakin et al., 2025), across models, those with higher self-consistency are linked to greater vulnerability to mistakes. Thus, even when models consistently apply concepts they may not be safe to deploy. This is evidence of a consistency dilemma in LLMs: self-consistency is operationally useful, but models that are more consistent are also more prone to mistakes.

Transparency is increasingly mandated for public-sector AI systems, with organisations required to publish statements describing their AI use and oversight arrangements. However, the existence of such artefacts is often treated as equivalent to transparency itself, despite limited evidence that they proportionately serve relevant stakeholder groups. From a requirements engineering perspective, this raises a validation concern: compliance with mandated disclosure criteria does not necessarily ensure transparency adequacy for stakeholders with different levels of risk exposure, decision control, and involvement. This paper presents an empirical analysis of 92 publicly available AI transparency statements published by Australian Government agencies under the national AI governance mandate. We introduce the stakeholder Risk–Control–Involvement–Need (RCIN) framework to differentiate stakeholder classes according to their structural position and transparency needs. Using a structured rubric derived from the mandated criteria, we evaluate how both the mandate and published statements are calibrated to each stakeholder class. The findings show that while structural compliance is widespread, transparency calibration is uneven. Criteria serving high-control stakeholders are consistently realised, whereas criteria most critical for high-risk, low-control stakeholders are fewer and less substantively addressed. We conceptualise this as the Transparency Illusion: a condition in which transparency appears satisfied through compliant artefacts yet remains unevenly calibrated to stakeholders bearing the greatest exposure to AI-supported decisions. The study frames transparency as a stakeholder-calibrated validation problem, demonstrating that artefact-level compliance does not constitute requirements validation in this context.

Delirium is common in hospitalized patients and is often missed in routine care. We present a user-centered interactive machine learning (UC-iML) framework for delirium detection support that combines physician-guided feature refinement with interpretable modeling. Using 3,862 labeled admissions from six Toronto hospitals in the General Medicine Inpatient Initiative (GEMINI), we integrate administrative variables, laboratory results, medications, and a radiology-derived text indicator. Physicians guide feature refinement and model evaluation, and Shapley Additive exPlanations (SHAP) are used to summarize feature attribution. We evaluate standard supervised classifiers with temporally separated holdout testing and a later-phase validation cohort. Compared with automated and baseline variants, the proposed framework shows better overall discrimination and stronger temporal robustness, while the explanations highlight clinically meaningful signals. These results support UC-iML as a practical human-in-the-loop framework for clinically relevant delirium modeling.

Legal AI benchmark research frequently invokes the assumption that large language models can improve access to justice, including for people who cannot access lawyers in order to understand and exercise their legal rights. We argue that current benchmarks are not equipped to support this assumption because they evaluate legal reasoning over inputs that have already been preprocessed by legal experts, which measures the upper bound of model performance. Access to justice depends on a lower bound: how models perform when inputs come from pro se litigants, whose prompts may contain noisy narratives, buried facts, omissions, folk-legal assumptions, and surface-level errors. These degradations are comparable to conditions under which LLMs are known to degrade in the general machine learning literature, including long-context sensitivity, underspecification, hallucination, and typographical perturbations. We connect evidence from pro se literature with this body of machine learning research and present a small perturbation experiment on LEXam, a legal benchmark, to illustrate the gap between these two bounds. If model development continues to focus on benchmarks that measure only the upper bound, this gap may remain hidden or even widen. We conclude by calling for legal benchmarks that directly measure robustness under pro se-like inputs so that access-to-justice claims about legal AI can become empirically testable.

A Markov Logic Network (MLN) is a probabilistic relational model used in Statistical Relational Artificial Intelligence for defining a probability distribution on the set of possible worlds with domain $D$ for an arbitrary finite domain $D$. An MLN consists of soft constraints with associated weights which are nonnegative real numbers. In this study we consider a language speaking about a property $P(x)$ and a relation $R(x, y)$. We consider an MLN for which every Boolean combination of $P(x)$ and $R(x, y)$ is a soft constraint (with associated weight). Let $n$ denote the size (cardinality) of the domain. We show that, for every choice of weights, if the weights are scaled by $1/n$ then, for every first-order sentence $\varphi$, the probability that $\varphi$ holds tends to either 0 or 1 as $n \to \infty$; that is, a 0-1 law for first-order logic holds. Morover, the limit probability does not depend on the weights. If we instead use the standard semantics of MLNs, in the case of which the weights are not scaled, then the limit behaviour is more complicated and depends on the weights. With unscaled weights we get 7 qualitatively different cases which depend on the weights. In some cases we have a 0-1 law for first-order logic, in some cases not, but we may still have a convergence law. The influence of the weights on the asymptotic probability of a first-order sentence may be in the form of a sudden “phase transition” from one of the 7 cases to another. The presence of a convergence law has positive implications for inference on large domains.

Audio-visual speech enhancement (AVSE) exploits visual cues such as lip movements to recover speech in noisy environments. Recent work introduced diffusion-based unsupervised AVSE, where a speech diffusion model conditioned on visual features via cross-attention is trained and used as a data-driven prior for posterior sampling-based speech enhancement. Despite promising performance over its audio-only counterpart, the impact of explicitly enforcing cross-modal alignment in the fusion remains unclear. In this work, we propose to augment the diffusion training objective with a contrastive audio-visual loss to encourage stronger use of visual information while keeping the posterior sampling framework unchanged. Experiments across matched and mismatched test data show consistent improvements in interference suppression, signal reconstruction, and perceptual quality, with the largest gains at low SNRs. Code is available at https://github.com/ cexauce/AV-CA-DiffUSE

Long-horizon web agents often fail in ways hidden by final-answer evaluation: they may visit useful pages, produce a well-formed answer, and terminate confidently while still missing fields, over-including unsupported items, or relying on stale evidence. We study these failures with Parallel WebBench, a parallel web-exploration benchmark containing 1,679 verified records: 350 manually curated parallel tasks and 1,329 reconstructed records with verified URL-based trajectories. We train WebExplorer-style agents with GRPO under human-only, balanced human-synthetic, and synthetic-heavy data mixtures. At 16k context and 16 interaction rounds, the best GRPO model improves completion over WebExplorer-8B from 50.7% to 96.0% and GPT-4.1-mini-judged element-wise F1 from 0.2489 to 0.4529, but binary accuracy remains far below completion. Trace-level analysis identifies three persistent failure modes: context-bound search loops, premature termination on partial answers, and synthesis collapse after relevant evidence has already been retrieved. These results show that synthetic-data GRPO reduces abstention and improves partial correctness, but leaves a completion-correctness gap that requires evidence-grounded coverage and synthesis diagnostics.

We propose a rigorous and systematic mathematical framework for tracking the cognitive trajectories of a user, in the context of algorithmic sycophancy and AI-driven delusional spiraling. Using tools from dynamical systems theory and stochastic differential equations, we explore how individuals perceive, interpret, and update their beliefs as they interact with AI chatbots that possess hidden traits of sycophancy. We treat the evolving conviction as a continuous log-odds state variable, coupled into a stochastic differential equation, navigating a multi-valley potential energy landscape. Our analysis reveals several critical observations governing the stability and rigidity of belief dynamics. We demonstrate that the baseline prior perception of the individual is systematically enhanced by sycophantic feedback beyond a critical threshold. Here, the perceptual potential landscape undergoes a structural phase transition that severely deepens any incremental initial tilt present in the baseline state, transforming the landscape and giving rise to deep, highly resilient attractor basins that trap the individual in unshakeable, self-reinforcing, delusional convictions. Finally, we demonstrate that genuine external information can successfully challenge these rigid states. If this incoming evidence is strong and authentic enough to overcome the internal feedback barrier, it can correct the structural asymmetry caused by sycophancy, inducing a perception reversal that successfully restores the objective belief state.

Multimodal Large Language Model (MLLM)-based web agents provide practical, high-precision solutions for visual browser automation; however, they inherently expand the attack surface, introducing novel vision-based vulnerabilities. Existing adversarial evaluations targeting these agents frequently rely on permissive threat models and visually conspicuous artifacts. In this paper, we investigate a constrained vulnerability detection setting: a trusted web platform where the evaluator acts solely as an unprivileged third party, such as a merchant or advertiser, controlling only a semantically legitimate, spatially constrained region, such as an ad slot, a sponsored card, or a localized widget. Operating under these realistic constraints, we propose MIRAGE, a novel visual indirect prompt injection framework for targeted next-action hijacking. Our approach leverages diffusion models to generate perceptually benign adversarial images strictly confined to the attacker-controlled boundaries permitted by the trusted service provider. To maximize attack efficacy within such a restrictive setting, we introduce a robust optimization technique combining curvature-aware adversarial diffusion guidance with sparse, dark-pixel residual perturbations. Comprehensive evaluations against prominent MLLM web agent frameworks, specifically SeeAct and OpenClaw, empirically demonstrate the potency, realism, and stealth of our proposed MIRAGE.

Generative AI tutors provide real-time, personalized learning support, but also create a new education inequity: students with access to premium AI services may receive clearer explanations, more personalized guidance, and better scaffolding than students limited to free or low-cost services. To address this challenge, we propose FairTutor, an equity-aware model-routing framework that achieves cost-effective AI tutoring via pedagogically motivated multi-agent orchestration. FairTutor combines query analysis, pedagogical planning, low-cost model generation, evaluator-guided critique and revision, and selective escalation to premium AI models. We introduce access-tier AI Education (AIED) Advantage Gap to measure the quality difference between premium-access and budget-constrained tutoring, and TutorAccessEval, a benchmark spanning math, reading, writing, science, and language learning. Empirical evaluations show that FairTutor achieves 97.1% of premium pedagogical quality (in floor-adjusted Likert scale) while reducing serving cost by 71.6%. Sensitivity analysis reveals a tunable cost–quality Pareto frontier, enabling FairTutor to be tailored to the needs of diverse student populations.

UI videos provide a natural input for generating interactive webpages, as they capture both webpage appearance and action-triggered state transitions. However, directly applying video-capable vision-language models to this task remains insufficient. Existing models typically rely on sparse sampling or compressed temporal representations, which may miss short action boundaries and break the state-action-state transitions needed to implement webpage behavior. We formulate UI video-to-code generation as executable state-transition recovery from interaction videos, and identify this failure mode as state-transition misalignment. We introduce Video2Code, an action-aware video-to-code approach for recovering executable UI state transitions. Rather than allocating the visual budget uniformly across the video, Video2Code first performs coarse video understanding to locate action-critical regions, then invokes a temporal clipping tool to revisit these regions at higher temporal resolution before generating HTML/CSS/JavaScript code. We instantiate Video2Code with action-aligned video-code supervision and evaluate it under both visual and functional criteria. Experiments show that Video2Code substantially strengthens the underlying open-source model for UI video-to-code generation, improving functional correctness over direct video observation, especially on dense multi-step interactions.

LLM-as-user-simulation has become core infrastructure for conversational AI: agent benchmarks (tau-bench), training pipelines, and a growing body of fidelity studies all rely on LLMs role-playing the human side of dialogue. Existing frameworks measure communicative fidelity – whether simulators talk like humans – against ground truth from paid participants role-playing assigned goals. We argue this has a structural blind spot: when the goal is assigned, the user's willingness is exogenous, so no framework can test whether simulators make decisions like real users whose motivation is endogenous, latent, and decaying. We introduce decision fidelity – whether a simulated population reproduces the decision-state dynamics of real users facing real, consequential choices – and measure it on a unique testbed: 2,790 production conversations between an LLM sales agent and real customers, including 793 with verified payment outcomes. Using a teacher-forced probe protocol that holds context and instrument fixed, we find a systematic, outcome-correlated failure we call the disengagement deficit: simulators reproduce eventual buyers almost exactly (depth bias +0.09) but inflate eventual non-buyers toward the purchase frame (depth bias +0.40; d=0.38, p<0.001), halving expressed resistance (25.1% to 13.5%) and nearly doubling deliberation (21.9% to 40.1%) while fabricating no purchases. The deficit replicates across model families (DeepSeek: d=0.41, p=0.002) and resists the obvious fix: instructing the simulator that it may disengage cuts marginal bias five-fold but barely moves the outcome-conditioned contrast (d=0.34, p=0.008). Real non-buyers say "not now" and stop; simulated non-buyers ask about price. Evaluating or training sales and persuasion agents against such simulators overstates funnel progress exactly where it matters most – the customers who walk away.