Papers

Autoregressive multimodal large language models (MLLMs) suffer from error snowballing: a single incorrect inference early in a chainof-thought (CoT) trace corrupts all downstream reasoning. We find that in state-of-the-art open-source MLLMs, once the first error occurs, the reasoning cascades into failure across all remaining steps in 65% of such cases (a metric we term the snowball rate). Existing mitigations-sampling multiple chains, post-hoc self-verification, or full program synthesis-either lack symbolic grounding, catch errors too late, or sacrifice the flexibility of natural language reasoning. We propose Constraint-Anchored Reasoning Traces (CART), a neuro-symbolic framework that trains MLLMs to interleave natural language reasoning steps with symbolic constraint assertions: lightweight, machine-checkable statements about visual content (e.g., count(red_objects) = 3). A dual-pronged Constraint Propagation Module-combining a learned neural grounding head with Boolean Constraint Propagation-continuously verifies these anchors against extracted visual features and checks their mutual logical consistency. When a contradiction is detected, a backtrack controller halts generation and reverts to the last consistent checkpoint, preventing error propagation. A variable-frequency emission mechanism allows the model to adaptively control anchor density, avoiding trace bloat. We construct 218K training instances by augmenting GQA, CLEVR-CoGenT, and VCR with ground-truth constraint annotations derived from scene graphs, and fine-tune open-source MLLMs (LLaVA-NeXT, Qwen2-VL) via LoRA. On five benchmarks, CART reduces the snowball rate from 0.65 to 0.14, improves GQA accuracy by +4.6 percentage points over trainingonly baselines, and achieves 89.1 F1 on POPE-all with at most 18% inference overhead.

This paper proposes a causal independence principle for value – the value Causal Markov Condition (v-CMC) – and develops the conceptual and mathematical foundations of a "causal value theory" linking causality and utility. After motivating a local formulation of the v-CMC, we introduce a probability-value duality that translates standard causal-inference results into the value setting. In particular, we formulate local, global, and decomposition versions of the v-CMC and prove their equivalence. We also define v-separation and show that it is sound and complete for conditional value independence. Furthermore, we derive a Bellman-type recursion as a special case of the v-CMC, thereby generalizing standard Bellman recursion from linear chains to causal DAGs. Finally, we show how the v-CMC supports modular transfer and updating of utility information across causal contexts and develop algorithms for causally structured utility elicitation and canonical influence-diagram construction.

Large language models and LLM-based agents are widely used as personal chat assistants, enterprise copilots, and autonomous workflow agents. In all these applications, memory (the ability to retain, access, and reason over information accumulated over long contexts and multiple interactions) plays a crucial role in determining the reliability of any agent. We introduce RECON (Reasoning over Extended Contexts with Obfuscated Narratives), a benchmark for evaluating compositional reasoning over long contexts. RECON spans 24 case files across three domains (criminal, medical, and financial), each ranging from 50k to 100k tokens, and tests agents on six memory intensive tasks: reconstructing multi-hop evidence chains, propagating cascading invalidations, resolving source conflicts, counterfactual reasoning, satisfying temporal constraints, and temporal fact retrieval. Recent memory benchmarks evaluate whether agents can retrieve scattered facts or detect if a fact has changed whereas RECON evaluates what happens after the change, whether agents can trace which downstream conclusions are affected, which survive through independent support, and how alternative timelines would have unfolded. Our evaluation reveals substantial limitations across current architectures: even the strongest non-Oracle system reaches only 22.4% Accuracy, with retrieval and reasoning each surfacing as challenges.

We study Controlled Query Evaluation (CQE), a declarative approach to confidentiality-preserving data access, in the context of Description Logic (DL) ontologies, and for confidentiality policies expressed through Epistemic Dependencies (EDs). We first address the problem of answering queries (specifically, Boolean unions of conjunctive queries) under known semantics for CQE (GA- and IGA-entailment). Our results show that if the TBox is expressed in $\text{DL-Lite}_{\mathcal{R}}$, CQE is computationally intractable in general. Moreover, in the presence of EDs, the IGA semantics has recently been proven not to satisfy an important confidentiality preservation property known as indistinguishability. With the goal of defining computationally easier and confidentiality-preserving forms of CQE, we introduce a new semantics for CQE, based on the notion of minimal policy violation (MPV). We show that the new semantics provides a sound approximation of the previous ones, while satisfying the indistinguishability property. We also prove that, in the case of $\text{DL-Lite}_{\mathcal{R}}$ ontologies, query entailment under the MPV semantics can be decided in polynomial time in data complexity. Finally, we present a software implementation of our framework that we used to evaluate the feasibility of this new approach using an existing benchmark for OWL 2 QL.

We describe DS@GT's submission to the eRisk 2026 Task 1 challenge on conversational depression screening, in which systems interview LLM personas that simulate individuals with varying depression profiles and produce a Beck Depression Inventory II (BDI-II) score plus four key symptoms per persona, without directly asking sensitive mental health questions. Our pipeline evolved through three stages: a monolithic single-model prototype to start off, a baseline multi-agent architecture that separates conversational interviewing from BDI-II scoring under a coordinating orchestration layer, and a final hybrid configuration that replaces the paid GPT-5-nano interviewer with the open-source Gemma 27B. To offset the model's weaker reasoning and instruction-following, the hybrid adds three algorithmic components: a precomputed dialogue tree that standardizes interview openers and follow-ups, a reliability-weighted consensus aggregation inspired by the Weaver framework, and a cluster-based imputation step for unprobed symptoms. We submitted three fully automated runs across all 20 personas, with Run 1 from the paid baseline and Runs 2 and 3 from the hybrid. Hybrid Run 3 achieved an ADODL of 0.9063, ranking 3rd among all complete-submission runs and placing DS@GT 2nd among the 21 teams overall, while outperforming our paid baseline Run 1 (0.8841) at roughly one-quarter of the per-persona API cost. These results support our central hypothesis that with sufficient algorithmic supervision, a weaker open-source model can compete with a stronger proprietary model in the conversational interviewer role. Our source code is available at https://github.com/dsgt-arc/erisk-task1-2026.

Semi-supervised learning (SSL) is an effective solution for medical image segmentation with limited annotations. Existing SSL methods mainly rely on pseudo-labels generated by teacher-student supervision or cross-network consistency. However, these methods lack an explicit structural reference for judging pseudo-label quality. Low-quality pseudo-labels may lead to unreliable training, error accumulation and confirmation bias when processing unlabeled data with substantial appearance variations. To address this issue, we proposed OFD-Net, a teacher-free single-network framework for reliable semi-supervised medical image segmentation. OFD-Net employs an Orthogonal Feature Disentanglement Module (OFDM) to capture OFD features for reliable SSL by disentangling unlabeled data into background and foreground representations with a reliable structural distribution, thereby effectively reducing error accumulation and alleviating confirmation bias among unlabeled data. Specifically, OFD-Net explicitly employs a Disentanglement Guidance Module (DGM) to inject the resulting structural priors of foreground-background into the decoder by deformable convolution processing, and outputs predictions with clearer foreground representations. Based on DGM and the OFDM, we further develop a reliability-aware pseudo-label learning mechanism that evaluates unlabeled supervision according to the structural consistency between the main prediction and the disentangled foreground-background responses, and then down-weights unreliable regions during training. Extensive experiments on four public medical image segmentation benchmarks, namely ISIC-2016, Kvasir-SEG, Synapse, and ACDC, validate the effectiveness of OFD-Net. These results confirm that orthogonal foreground-background disentanglement enables OFD-Net to establish an efficient and reliable training paradigm within a teacher-free single-network framework.

The widespread adoption of high-level deep learning libraries, while accelerating model development, has increasingly abstracted away the internal mechanics of neural networks, creating a gap between practical usage and fundamental understanding. To address this, the paper presents a self-contained neural network framework implemented entirely from scratch without relying on automatic differentiation or pre-built deep learning modules. The implementation encompasses all essential components, including multi-layer architectures, diverse activation functions, regularization techniques, and state-of-the-art optimizers. Beyond serving as a pedagogical instrument that demystifies forward/backward propagation, gradient dynamics, and optimization landscapes, the framework demonstrates robust performance when applied to a multi-class classification task, successfully validating its correctness, numerical stability, and generalization across varied configurations. The extensible design and clean modularity further position it as a reliable baseline for educational purposes and future research exploration.

The increasing adoption of Large Language Models (LLMs) as AI components in modern software systems introduces distinct security risks to the software supply chain. While many considerations and safety mechanisms are in place for components of the traditional software supply chain, the recent rapid adoption of AI components and platforms has overlooked these hard learned lessons. Selecting and integrating AI models without clear guidance on how these choices affect system security may leave applications vulnerable to threats, such as malicious components, data leakage, and unintended behavior. The goal of this study is to understand practitioners' decision making process and security considerations in selecting and integrating AI components through an exploratory semi-structured interview study. Toward this goal, we conducted semistructured interviews with 22 software developers, architects, and AI practitioners across diverse organizations about how they integrate AI components into their software. Our analysis finds that practitioners' model selection is predominantly driven by functional criteria, including performance, accuracy, cost, and specific features, e.g., tool calling or multimodal support, while security is rarely considered as an evaluation criterion. We observe a consistent lack of security concern throughout the AI component integration process, with established software supply chain lessons overlooked or ignored. The industry is repeating the historically costly mistakes of early software dependency management, prioritizing rapid reuse and availability over security and provenance. We distill our findings into actionable recommendations for AI adopters, model providers, and researchers, advocating for a proactive, security-by-design approach that integrates security evaluation into component selection and sustains it throughout the software development lifecycle.

This position paper argues that claims about explanation stability are scientifically invalid without cross method validation. Just as statistical significance requires the test statistic to be specified, stability should either be evaluated across multiple attribution paradigms or explicitly scoped to the computational objective of a single method. In controlled chest X ray experiments, DenseNet201, ResNet50V2, and InceptionV3 achieved AUC values above 99%, yet their stability rankings reversed across attribution methods. LayerCAM ranked InceptionV3 as the most stable model, with an IoU of 0.777, whereas GradCAM++ favored DenseNet201 and reduced InceptionV3 stability score by 17.3%. These findings demonstrate that explanation stability is an emergent property of the model method pair rather than an intrinsic characteristic of the model alone. We therefore argue that explanation based claims should be validated across multiple attribution methods and that regulatory submissions should explicitly specify the attribution operators used to avoid creating illusory safety assurances.

Large language models now translate natural-language descriptions of decision problems into solver-ready optimization models, and they fail silently. A generated model often runs and still encodes the wrong problem, while standard evaluation compares optimal values against labeled answers that deployment does not provide. How to certify such a model without any reference is the question this paper addresses. We develop falsification-based verification. Every numeric quantity in a problem description plays a role that the text itself states, such as a capacity, a requirement, or a unit cost, and any correct model must respond to changes in these quantities as the stated roles dictate. From duality and sensitivity analysis we derive a battery of solver-based tests that are individually sound, so a violation certifies a faulty model and the false-positive rate is zero by design. We characterize the errors that no test of this kind can see, give conditions under which each canonical error class is detected with certainty, and prove that perturbation testers with tuned thresholds cannot be simultaneously sound and nontrivial. Across 326 ground-truth models, a synthetic family, and four public benchmarks with two generators, the battery flags 0.0% of faithful models while a threshold tester flags 54.9%; it detects 56.1% of core formulation errors, 70.0% under certified preconditions, and 40.4% of the errors that value-based scoring provably cannot see, and it reproduces the predicted detectability pattern including its blind spots. Every flag carries a machine-checkable certificate that localizes the defect, and a full audit costs about 25 millisecond-scale solver calls per model. Classical sensitivity analysis and duality thus offer a rigorous, label-free audit that complements existing evaluation of AI-generated optimization models.

Existing hallucination detection methods are typically conducted at the inference stage, without making any modifications to the model itself. In this paper, we are interested in exploring fine-tuning strategies that enhance the detectability of hallucinations in the resulting model. Focusing on semantic-entropy-based detection, we observe that many erroneous outputs remain undetected because the model produces nearly identical incorrect answers across multiple runs. To address this, we propose diversity-oriented fine-tuning to encourage more varied generations. We introduce two specific strategies: one based on Supervised Fine-Tuning (SFT) and the other on Direct Preference Optimization (DPO). Extensive experiments are conducted to evaluate our approach and analyze the behavior of the models before and after fine-tuning. We find that after adopting our fine-tuning methods, the models become less likely to produce low semantic entropy responses for hallucinated answers, thereby improving the effectiveness of hallucination detection, eventually yielding results better than or comparable with state of the art methods. The code will be publicly released.

Full fine-tuning remains a strong way to adapt pretrained LLMs, but it updates all weights and can be expensive. LoRA reduces the number of trainable parameters, but it does not directly answer which pretrained components should be trained and which can be frozen during adaptation. We introduce TopoTuner, a topology-guided fine-tuning framework for selective freezing of attention projection matrices. \method treats each projection matrix as a row cloud and uses Wasserstein distances between persistence diagrams to measure how its topology changes during fine-tuning. TopoTuner learns a reusable freezing profile from a source dataset and transfers it to efficiently fine-tune models on out-of-domain datasets, evaluating whether task-specific topological drift generalizes across question answering and sentiment analysis tasks. Across LLaMA-3.1-8B, Mistral-7B-v0.3, and Qwen3-8B-Base, TopoTuner is competitive with full fine-tuning while training only 1-2% of the model parameters, and outperforms LoRA in 7 out of 9 model-dataset settings, which can change up to 39.57% of the projection parameters. Along with minimized updates, TopoTuner reduces training time by 20.4% relative to full fine-tuning and 5.5% relative to LoRA on average. TopoTuner opens a new direction for reusable freezing profiles, where fine-tuning behavior learned on one dataset can be shared across multiple tasks.

We present TellTale, a text-only approach to ambivalence/hesitancy (A/H) recognition in interview videos, evaluated on the BAH dataset as part of the 3rd A/H Video Recognition Challenge (11th ABAW Workshop, ECCV 2026). Although the dataset provides video, audio, facial crops, and transcripts, TellTale relies on the transcript alone and combines three probability streams. Two text encoders, multilingual-e5-large and mDeBERTa-v3-base, are fine-tuned with parameter-efficient LoRA adapters under a multiple-instance learning (MIL) objective, in which transcript chunks are scored individually and pooled with a smooth maximum so that only the video-level label is needed for supervision. The third stream requires no training: a quantized 14B instruction LLM is prompted, zero-shot, to rate each transcript for A/H. The three probabilities are combined by a weighted average and a single decision threshold, both selected on participant-grouped cross-validated predictions. On the organizer-scored private test set of 152 videos from unseen participants, TellTale achieves a Macro-F1 of 0.7364 and an average precision of 0.7940, compared with 0.2827 Macro-F1 for the official vision-based baseline.

Hardware engineering exposes coding agents to a form of long-horizon work that is difficult to capture with pass-at-k: progress is continuous, tool feedback is delayed and heterogeneous, and a backend failure may require revising RTL rather than tuning another physical-design parameter. Existing benchmarks measure RTL generation, repository repair, verification, PPA evolution, or physical implementation, but their different designs and oracles make it hard to determine where an agent succeeds or fails across abstraction boundaries. We introduce CLOSER-Bench, a controlled evaluation protocol for budgeted cross-stage design closure. For one design and one hidden objective, it pairs spec-to-RTL, RTL-to-GDS, and spec-to-GDS tasks, records every simulator, synthesis, STA, and place-and-route invocation, and measures final quality, anytime progress, tool cost, and cross-stage recovery. The benchmark is built on open-source Verilator, Yosys, OpenROAD, KLayout, Sky130, and the Harbor agent harness. A ten-task pilot spanning RTL repair, mutation-based verification, coverage, PPA optimization, design-space exploration, cross-model debugging, and security establishes the executable harness and exposes a sharp completion–closure gap: three agents solve a localized AXI repair task, while the matched verification-closure task separates a frontier agent from two otherwise successful baselines. We further validate a full RTL-to-GDS flow and construct a macro-based AXI/DMA streaming accelerator for the stage-paired evaluation. These results motivate treating hardware closure as a budgeted sequential decision problem rather than a collection of independent code generation tasks.

Translating unstructured clinical prescriptions into patient-specific foot orthoses (FOs) is hindered by a semantic-physical misalignment: high-level clinical intent is not mapped deterministically onto the 3D geometric parameters of the orthosis, and existing design workflows remain dependent on manual expertise with no instantaneous biomechanical validation. We present TANS-FO, a research prototype-a modular pipeline with closed-loop feedback for computational design automation of customized FOs, not a clinically validated therapeutic device. A Text-Aligned Neural Surrogate (TANS) uses cross-attention to project clinical-text embeddings onto a continuous lattice-density field, while a Graph Neural Network (GNN) surrogate predicts plantar stress in real time as a substitute for Finite Element Analysis (FEA). The framework is anchored on the open-access PicoFoot-5K anthropometric database (5,230 subjects; 30+ anatomical parameters). Under standardized quasi-static loading, the GNN surrogate agrees with an Abaqus reference solver (R^2 = 0.94), and the full pipeline synthesizes manufacturing-ready lattice insoles within minutes. On the Male 18-40 cohort, the proposed system attains a surrogate-predicted peak-pressure reduction of 34.7% over parametric CAD, with a fit error of 0.42 mm. Separately, an exploratory feasibility observation (n = 12; 2-week follow-up; no control group) using VAS pain reporting indicates short-term comfort improvement (VAS 6.4 -> 2.1), but this data is explicitly classified as preliminary observational evidence only-not evidence of clinical efficacy.

Differential privacy (DP) is increasingly deployed to limit membership inference risk in machine-learning systems. Prior work has shown that DP-SGD can widen accuracy disparities across demographic groups, but this framing treats fairness as a purely outcome-side concern. We argue that privacy cost, the information leakage borne by each group, is itself a form of harm, and adopt a compensatory-fairness framework in which a group that involuntarily bears greater privacy exposure is owed proportionally greater benefit from the system. From this principle we derive the Privacy-Cost Equity Ratio (PCER), a group fairness metric defined as a group's positive prediction rate normalized by its per-group overfitting gap. By a standard membership inference bound, this overfitting gap upper-bounds each group's vulnerability to inference attacks, making PCER a conservative measure of benefit relative to exposure. PCER needs only per-group train and test accuracy (no shadow models), making it a practical post-hoc audit tool. We evaluate PCER alongside standard fairness metrics across six benchmark–attribute combinations spanning tabular and NLP domains, under DP-SGD at a range of privacy budgets, and validate the overfitting-gap proxy against a direct threshold membership-inference attack. The results reveal patterns that outcome-based metrics miss. On COMPAS, PCER uncovers a persistent double disadvantage: the protected group bears both greater privacy exposure and worse predictive outcomes, something demographic parity gap masks entirely. Sensitivity analysis shows very strong privacy guarantees collapse both groups' overfitting to a numerical floor, rendering exposure-based audits uninformative in that regime. Together, these findings show that fairness audits of privacy-preserving systems must account for who bears the cost of protection, not only who benefits from its outcomes.

Large language models (LLMs) are increasingly used to automate data-processing workflows, yet coding agents typically produce scripts that are not automatically materialized as persistent, editable platform artifacts. We call this disconnect the NL2Pipeline gap. To bridge it, we introduce DataFlow-Harness, a platform that guides an LLM agent to construct platform-native directed acyclic graphs (DAGs) through typed, incremental mutations rather than free-form scripts. The platform combines DataFlow-Skills for procedural guidance, a Model Context Protocol (MCP) layer that exposes the live operator registry and current pipeline state, and DataFlow-WebUI, which synchronizes conversational authoring with a visual DAG editor. On a 12-task data-engineering benchmark, DataFlow-Harness achieves a 93.3% observed end-to-end pass rate. Relative to Vanilla Claude Code, it reduces measured monetary cost by 72.5% and generation latency by 49.9%; its observed pass rate is within 0.9 percentage points of the Context-Aware Claude Code baseline while its cost is 42.8% lower. Per-task analysis indicates that Skills are most useful when construction depends on implicit procedural knowledge. These results show that live platform grounding can produce persistent, editable workflow artifacts with an observed reliability close to script-generation baselines and with lower measured construction cost and latency.

Long-horizon AI agents are becoming increasingly capable, yet their interaction with users remains surprisingly thin. In most workflows, users give an initial instruction, receive only selective textual updates, and lose a clear sense of what the agent is doing or when to step in. This leaves a missing part in the current agent ecosystem: an always-on Jarvis-style mediator that keeps the agent continuously reachable to the user. Such a mediator should support real-time spoken interaction with the user, answer questions without interrupting the worker, proactively report progress or confusion, and inject user guidance back into the agent's execution when useful. In this work, we introduce JarvisBench, a benchmark for measuring the dual value of mediation in long-horizon agent workflows. JarvisBench contains two complementary tracks: an agent-collaboration track that measures whether mediation improves downstream task completion, and a user-interaction track that measures whether mediation makes ongoing execution more understandable, responsive, and accessible to users. We instantiate the benchmark with a modular reference Jarvis prototype and evaluate it on 34 text-only WildClaw tasks executed in OpenClaw. Preliminary results with GPT-5.5, Claude Opus 4.7, Gemini-based, and GPT-based worker agents suggest that Jarvis-style mediation can provide trace-grounded responses to user questions and improve task performance when sparse user guidance is injected at appropriate moments. The results also show that effectiveness depends strongly on the mediator's LLM brain, highlighting both the promise of this missing middle layer and the need for broader community effort. Demo page https://cchen1436.github.io/jarvis

The rapid digitalisation of elite sport has created new opportunities for integrating artificial intelligence (AI), performance analytics, and decision-support systems into athlete development and competition management. However, existing solutions remain fragmented, typically addressing isolated tasks such as performance analysis, athlete monitoring, or referee support. This paper presents \textbf{FST$\cdot$ai~2.5}, an explainable, uncertainty-aware, and secure AI framework for Olympic and Para-Taekwondo. \textbf{FST$\cdot$ai~2.5} introduces a unified digital ecosystem integrating athlete intelligence, competition analytics, federation-scale data management, AI-assisted decision support, athlete and event digital twins, explainable performance indicators, and adaptive training recommendations. The framework supports World Taekwondo (WT), Member National Associations (MNAs), coaches, referees, analysts, and athletes through transparent, secure, and federation-aware governance. By combining multi-source competition data, athlete-performance information, and contextual evidence, \textbf{FST$\cdot$ai~2.5} provides tactical diagnostics, longitudinal athlete monitoring, performance forecasting, personalised development planning, and federation-wide benchmarking using explainable and uncertainty-aware AI. Prototype deployments demonstrate the feasibility of the proposed framework. Although developed for Olympic and Para-Taekwondo, the methodology is broadly applicable to explainable AI, digital twins, and trustworthy decision support in combat sports and other high-performance sporting environments.

The RLxF programme argues that learning signals should come from world feedback rather than from internal model proxies. We instantiate this position in safe model-based control and distil it into three concrete design principles. Empirically, across four world-model architectures spanning a 2x MSE range, MPC planning is statistically equivalent (TOST, n=200), and dynamics-based uncertainty penalties increase collision rates from 26% to 34%: the standard MBRL safety proxy is anti-correlated with safety in this regime. Replacing the model-internal proxy with three world-feedback signals (a sensor-derived margin via minimum lidar, a temporal signal via time-to-collision, and an outcome-supervised feedback model g_psi trained on prior collision labels, structurally analogous to outcome-trained reward models in RLHF) reduces collisions to 1-14% without retraining the world model or the planner. The mechanism is structural: model uncertainty has support over state-prediction space, whereas task risk has support over constraint boundaries, with empirical correlation r < 0.15. From this we extract three RLxF principles (ground risk in world outcomes, validate proxies before deployment, and substitute outcome-trained feedback models when direct world signals are unavailable) and argue they apply equally to model-based control and to verifier-based or RLHF approaches in LLM alignment.

In high-risk environments such as disaster response, situational awareness depends not only on detecting hazards but also on communicating them clearly to human operators. Vision Language Models (VLMs) have shown strong potential for scene understanding in safety-critical settings, yet their value as part of human-facing robotic systems remains underexplored. We present a VR-based Human Robot Interaction framework for studying how VLM-assisted robots can support situational awareness in simulated hazardous environments. In our system, a robot explores a virtual scene and queries a VLM to identify potential hazards and annotate user-facing points of interest. These annotations are presented to a human operator through an immersive VR interface. This framework enables controlled evaluation of both robotic hazard identification and the communication of safety-critical information to users. Results from our study indicate that the annotated VR interface was preferred over the unannotated baseline and that participants reported high clarity, usefulness, and comfort when interacting with the system. These findings suggest that combining VLM-based robotic perception with immersive visualization is a promising approach for supporting situational awareness in hazardous settings.

Function-preserving network growth techniques such as Net2Net and progressive stacking expand a model's capacity without destroying its learned function, but existing formulations either tolerate numerical perturbations or require a full rebuild of the training program. We formalize Exact Network Surgery: the in-place insertion of a residual block into a live computational graph such that (i) the network function is preserved – bit-exactly under explicit floating-point hypotheses – and (ii) inserted parameters remain trainable immediately after insertion. We prove an identity-morphism theorem for gated residual blocks, a structural-locality theorem showing that a reactive invalidation engine recomputes exactly the downstream cone of the insertion point, leaving every other node's value and optimizer state untouched, and an escape-from-initialization proposition showing that the Gradient Shadowing gate alpha, initialized at zero over a randomly initialized branch, receives a generically non-zero gradient at insertion time. We identify a degenerate configuration – zero-initialized output projections combined with a zero gate – that is an exact saddle point gradient descent cannot escape. Every claim is validated on the reference implementation in NeuroDSL, a reactive graph engine in Julia: grafting is bit-exact on every logit tested (0 mismatches out of 1600); the gate escapes zero at the first optimizer step and unlocks branch gradients at the second, exactly as predicted; the degenerate configuration exhibits gradients identically zero for the entire 600-step run; surgery cost tracks downstream cone size with r = 0.9992 while graft-plus-invalidation bookkeeping is constant (about 0.75 ms) across insertion depths; and training resumes bit-identically across a real process restart. A flagged preliminary appendix reports first single-seed observations on post-insertion gate dynamics.

Background: Evaluating automated Software Requirements Specification (SRS) generation is challenging because few datasets provide fine-grained traceability between source requirements, intermediate elicitation artifacts, and generated specifications. Aims: We aim to study whether legacy SRS documents can be transformed into traceable synthetic pre-SRS artifacts that support fine-grained evaluation of LLM-based SRS generation. Method: We conduct an empirical study using ReqGenX, a controlled pipeline that decomposes SRS sections into source-grounded atomic statements, routes atoms to standards-inspired artifact types through multi-LLM plurality voting, and generates artifacts using constrained prompts with iterative judge-guided refinement. We evaluate ReqGenX on seven PURE SRS documents using grounding, quality, information retention, and downstream reconstruction analyses. Results: ReqGenX produces faithful and usable atoms, with median AlignScore values typically between 0.96 and 0.99 and Prometheus scores ranging from 4.34 to 4.85. Generated artifacts remain strongly grounded in their source atoms, with AlignScore values typically between 0.80–0.94 and judge pass rates near 100%; stricter Prometheus evaluation yields pass rates from 54.8% to 97.1%. In a downstream SRS reconstruction case study, artifact-backed atoms remain recoverable from generated SRSs, with SBERT means between 0.69 and 0.75 and AlignScore medians between 0.76 and 0.84. Conclusions: Traceable synthetic pre-SRS artifacts can support more fine-grained evaluation of LLM-based SRS generation, while exposing tradeoffs among faithfulness, information retention, and artifact completeness.

We propose a language representation for multimodal data in which any observation, whether image, video, or text, is expressed as a bag of atomic propositions, simple statements about the entities, actions, and relations in a scene. A global semantic codebook unifies these into a shared vocabulary of canonical atomic propositions, placing every modality and observation into one interpretable space that spans fine grained facts to high level concepts and composes into richer ones. This brings interpretability with reasoning, cross-modal understanding and retrieval, and compositionality that enables complex multimodal understanding, rich data curation and complex structured retrieval. We demonstrate the framework on autonomous driving and open-world data.

Coding agents are increasingly capable of sustained engineering and empirical research. Greater autonomy makes their research decisions themselves a target for evaluation: what to investigate, which experiments to run, and when to stop. We introduce Gauntlet, a develop-freeze-evaluate protocol for studying these decisions as agents build standalone game-playing programs. Starting from a game description, a raw observation/action interface, and an empty policy file, an off-the-shelf agent develops a controller from bare interaction in one autonomous session. We call the capability under study compiled agency: the shipped program plays with zero model calls. The capability is real and advancing: environment access adds 10 to 78 percentage points of held-out success over construction-only controls, and progress across model generations comes in steps, with tiers that defeat one generation entirely falling to the next. The protocol yields StarCraft II controllers that defeat every fair built-in AI, and in the newest generation the strongest cheating tier as well, and Civilization controllers that win complete games by conquest. Replaying more than 5,000 frozen versions exposes the research behind the programs: gains that plateau early; rigorous local investigation beside sparse validation of what actually ships; and stops that follow a race between the agent's own evidence and a model-specific transcript budget it was never given. When validation panels are refreshed mid-session, so that only the evidence changes, shipped success rises 12.4 points in nine of nine completed pairs of twelve initiated. The experiments an agent designs are part of the capability it delivers; Gauntlet makes them measurable and improvable–a step toward agents whose research practice, not just whose code, can be engineered. We release the benchmark, the corpus, and the replay tooling.

Multimodal LLMs report strong performance on scientific reasoning benchmarks, yet most treat perception and reasoning as a single measurable process. We introduce a five-task diagnostic experiment across physics and geometry benchmarks that isolates failures to perception, reasoning, or both. Incorrect diagram interpretation degrades performance even on problems models solve correctly from text alone, and accuracy generally rises from raw images to human-authored captions. Recovery under corrected captions is high for some models, separating perception-blocked failures from genuine reasoning bottlenecks. Which reasoning error follows a perception failure depends on domain: physics failures resolve into calculation errors, geometry into conceptual misapplication. As a discussion beyond our core experiments, InternS1-mini, despite heavy scientific pretraining and thinking capabilities, falls below the weakest model from experiments on every task, with reasoning traces frequently truncating before completion.

Knowledge-editing benchmarks certify local correctness, whether an edited model produces the new fact on near-edit prompts but not how much of the original fact remains decodable inside the model. We study residual knowledge directly with a linear trace probe: after editing a fact, we ask whether the original object is still recoverable from the model's hidden states. On GPT-2-XL, across three mechanistically distinct editors applied to 50 CounterFact edits, the original object remains linearly decodable well above chance after a successful edit (probe accuracy 0.96 for ROME, 0.86 for constrained fine-tuning, and 0.79 for the memory-based editor GRACE, against a chance level of 0.50; all edits reach 100% generation-based success). The GRACE result is the most informative: GRACE changes zero base-model weights, overriding the fact through an external memory, yet the original object is still decodable from the underlying network, so the residual trace cannot be attributed to an incomplete weight update. We read this as evidence that editing, even when behaviorally successful, suppresses rather than erases the original association in representational space. We also report a relearning-savings instrument that did not behave reliably in our setting and discuss why; we treat it as a negative methodological result rather than evidence. Code and data are released.

Previous AI alignment efforts have focused primarily on first-order social norms – teaching models what is socially acceptable or unacceptable (e.g., `do not steal'). However, social intelligence depends not only on norm recognition, but also on anticipating who will enforce it and how (e.g., public shame or even imprisonment). These second-order expectations, known as metanorms, govern how people respond when social rules are broken. We introduce a novel framework for evaluating metanorm reasoning in Large Language Models (LLMs) along two dimensions: emotional appraisal and behavioral response, and propose new classification tasks, namely, predicting self-regulation in violators, and other-regulation in observers. We release a multi-perspective dataset, NormReact, of 450 norm violation scenarios, hand-annotated for emotions and behavioral responses across norm violators' gender and observers' social closeness. Current LLMs portray a harsher social world: across six models, they overpredict negative sanctions where humans would expect inaction, and alignment with human judgments deteriorates as social distance increases. These findings suggest that AI systems in norm-sensitive domains from conflict mediation to policy simulation, may risk producing a distorted picture of social regulation: one that over-represents punishment and under-represents the tolerance, restraint, and relational calibration that characterize actual norm enforcement in real world.

Social media has become a central arena in which armed conflicts are contested, yet the pro-Israel and pro-Palestine communities on Telegram, whose broadcast architecture yields an unusually direct record of deliberate political communication, have not been systematically compared at scale. This study presents a multi-method computational analysis of 87,617 messages from sixteen Telegram channels, eight pro-Israel and eight pro-Palestine, spanning May 2021 to June 2026 and covering multiple conflict escalations. It combines sentiment analysis, three stance detection methods drawn from distinct paradigms (keyword matching, zero-shot DeBERTa via natural language inference, and a fine-tuned BERTweet model), and a framing analysis, all evaluated against 736 manually annotated messages. The fine-tuned model performed best (72.1% accuracy, 0.721 macro F1 under 5-fold cross-validation), outperforming both label-free baselines by 8 to 11 points; the baselines stalled in the low-to-mid 60s, indicating a hard ceiling for stance detection not adapted to in-domain language. The central finding emerges only when sentiment, stance, and framing are read together: the two communities deploy the same death- and victim-related vocabulary in opposite emotional registers, pro-Israel channels predominantly neutral and report-style, pro-Palestine channels markedly more negative, consistent with writing from the distinct discourse positions of acting party and affected party.

Modern multilingual tokenizers often fragment Ukrainian and other underrepresented Cyrillic-script languages more heavily than English, creating disparities in cost and context capacity. We quantify this overhead across nine production tokenizers and five languages with standardized Cyrillic and Latin representations, covering 8.37 million word forms. On a corpus benchmark, Ukrainian shows 68-121% token overhead on modern tokenizers and 220% on the older cl100k, measured through full-text fertility on the BrUK and Brown corpora. Overhead is negatively associated with Cyrillic vocabulary allocation in the subset with independently verified English baselines, although the association is not statistically significant (Spearman rho = -0.536, p = 0.215, n = 7). We evaluate two mitigation strategies. LLMLingua-2 reduces Ukrainian input length by 47-49% on an e-commerce RAG benchmark of 1,536 products and 145 queries, with no compression-induced value losses among 80 retrievable cases. A balanced byte-level BPE tokenizer trained with a 200K vocabulary cap, converging at 158,184 actual entries, reduces the held-out UK/EN ratio from 2.22x to 1.30x. Romanization increases Ukrainian token counts by 2-19% on most tokenizers. Across the five languages, tokenization efficiency favors the script more prevalent in web data. These findings indicate that training data allocation contributes to Cyrillic tokenization overhead and that mitigation is possible at both inference and tokenizer-design stages.