Search research

Large language model (LLM) agents increasingly take actions (screening applicants, recommending credit, triaging patients), yet fairness for LLMs is still measured by grading answers. We introduce AgentFairBench, a cheap, reproducible, multi-domain benchmark for demographic disparity in the actions of LLM agents. Grounded in a companion framework, the Bias Conduction Framework (BCF, restated here), it spans three regulator-anchored domains: hiring, lending, and medical triage. Synthetic, demographic-neutral profiles are evaluated in counterfactual matched sets that vary only a name-coded race x gender signal (in the Bertrand Mullainathan tradition), under four agent scaffolds of increasing agency (direct, chain-of-thought, multi-agent deliberation, tool-augmented). A NumPy-only harness computes counterfactual flip rate, mean absolute score difference (MASD), action-rate disparity, and tool-invocation disparity, with bootstrap confidence intervals, paired tests, and false-discovery-rate control, for single-digit dollars per model. A live leaderboard with a held-out private split and a contamination canary admits external models by submission. Our pilot (864 decisions plus a test-retest replication) carries a methodological lesson: comparing a six-group score spread against a two-run noise difference overstates disparity by ~ 2.4X through statistic arity alone. Against an arity matched noise floor and an omnibus group test, claude haiku 4 5 shows no demographic effect above sampling noise (0 of 120 pairwise and 0 of 9 omnibus contrasts survive correction); a planted-bias test confirms the instrument detects disparity when present. The contribution is a sound, sensitive, adoption-ready instrument, the arity matched null methodology, and open artifacts to scale it. Code, data, and harness are released under open licenses, with an anonymized review artifact.

Recursive LLM agents can broaden their search by spawning specialists. Some branches later request tools that send data or deploy code. When should a branch receive authority to act? We distinguish sandbox spawning, in which external controls prevent the specified harm, from capability activation, in which a selected branch crosses an irreversible-action boundary. Progressive Risk Vesting (PRV) holds a trajectory-level risk budget in escrow and debits it as branches are activated. We prove an anytime harm bound for adaptively generated trees. Branch outcomes may be dependent, but each local certificate needs to remain valid conditional on the full pre-activation history, including the information used to select the request. When activation gates, branch charges, and compute constraints are held fixed, delayed vesting preserves every policy available under irrevocable spawn charging. Marginal risk estimates can still fail after branch selection. In a stylized branching model, trajectory harm changes as the authority reproduction number $\mathcal{R}_A$ crosses one. As local risk $p$ approaches zero, trajectory harm is proportional to $p$ below criticality, proportional to $\sqrt{p}$ at criticality, and retains a positive floor above it. A finite-type occupancy model yields risk and compute shadow prices. For nested fanout modes with decreasing marginal value per unit risk, these prices produce a threshold rule. Branching calculations and a split-sample experiment illustrate the results. These synthetic studies do not estimate safety in deployed agents. The analysis suggests a design rule: search broadly in the sandbox and grant recursive authority sparingly, with an explicit risk charge.

Procuring supervised fine-tuning (SFT) data forces a buyer to decide, before any downstream training, whether a candidate corpus is worth acquiring. We present \sys{}, a statistics-first gating architecture that treats procurement as a cost-aware routing problem over three intrinsic quality axes -- diversity, utility, and redundancy. Cheap blind measurements are summarised into per-axis estimates with confidence intervals; a gate accepts a decision only when intervals are tight, sample sizes are adequate, and the axes agree, otherwise it escalates the case to an adjudicative debate between a buy-advocate and a reject-advocate judge, resolved by a presiding verdict. On a controlled benchmark of 12 datasets ($2{\times}3{\times}2$ grid over the three axes) with 5 seeds, the gate reaches 0.90 accuracy and 0.83 $F_1$ at $0.017 per unit, sitting between an always-verify baseline (0.75) and an oracle upper bound (0.98) while spending less than always-escalate ($0.020). We further report honest negative diagnostics of the debate path: a con-side win rate of 0.80 ($p\approx3{\times}10^{-6}$) and a 52% position-flip rate under advocate swapping expose negativity and positional biases that a naive LLM-judge would hide. We frame the injected-knob evaluation explicitly as a controlled synthetic benchmark for measurement fidelity and routing calibration, and delimit external validity as future work.

Training a language model on data scattered across bandwidth-limited nodes that cannot be centralized is a setting that arises in clinical networks, enterprise knowledge bases, and scientific consortia. We study the regime in which data must remain distributed across nodes, and ask what statistical guarantees are in principle achievable under explicit bandwidth budgets; we aim to characterize what is provably possible, not to demonstrate a deployment-ready system. Existing theory treats either training-time consistency or inference-time calibration in isolation, and no prior work makes bandwidth a first-class statistical parameter. We analyze two protocols, Federated Probe-Logit Distillation (FPLD) for training and Federated Conformal RAG (FC-RAG) for inference, as the analytical vehicles for our results. Our first main result is an explicit high-probability KL-consistency rate for FPLD with simultaneous dependence on node count $K$, per-node sample size $n$, quantization budget $B$, probe-set size $m$, and vocabulary size $V$; bandwidth enters only through an exponentially vanishing quantization term. Our second main result is a distribution-free marginal-coverage bound for FC-RAG, whose novel retrieval-bandwidth slack $Δ_{\mathrm{RAG}} = f_{\max}\sqrt{K^{-2}\sum_i v(B_i)}$ makes per-node retrieval bandwidth a first-class statistical parameter, with arithmetic aggregation across $K$ nodes shrinking the slack as $K^{-1/2}$ in the per-node-uniform regime. A Pinsker-type corollary composes the two bounds into an end-to-end coverage guarantee. Synthetic experiments verify the predicted scaling along the bounds' parameters; small-scale experiments on a GPT-2 testbed illustrate that the qualitative bandwidth-accuracy tradeoff survives on a real language model. A deployment-scale empirical evaluation is out of scope.

The rapid advancement of generative technologies has made synthetic images nearly indistinguishable from real ones, thereby creating an urgent need for robust detectors to counter misinformation. However, existing methods mainly rely on universal artifact features that are shared across multiple generators. We observe that as the diversity of generators increases, the overlap of these common features gradually decreases. This severely undermines model generalization. In contrast, focusing only on unique artifacts tends to cause overfitting to specific forgery patterns. To address this challenge, we propose LEGO (LoRA-Enabled Generator-Oriented Framework). The core mechanism of LEGO employs an MLP to modulate multiple LoRA (Low-Rank Adaptation) blocks, each pretrained to capture the unique artifacts of a specific generator, followed by attention-based feature fusion. Unlike conventional methods that seek a single universal solution, LEGO delegates unique artifact extraction to specialized LoRA modules by dividing its training procedure into two stages. Each LoRA module is individually trained on a single-generator dataset to learn generator-specific representations, then MLP and attention layers are trained on mixed datasets to dynamically regulate the contribution of each module. Benefiting from its modular yet robust design, LEGO can be naturally extended by incorporating new LoRA modules for adaptation to newly emerging next-generation datasets, while still achieving substantially better performance than prior SOTA methods with fewer than 30,000 training images, less than 10% of their training data, and only 5 epochs in each training stage.

Ovarian cancer is frequently diagnosed at an advanced stage, making preoperative contrast-enhanced computed tomography (CT) central to staging and surgical planning; yet the scarcity of annotated imaging data, compounded by privacy regulations, limits the development of generalizable computational models in this domain. Text-conditioned 3D CT synthesis has shown promise, but existing pipelines depend on paired radiology reports and have been evaluated only on chest CT. We propose OvESyn (Ovarian Evidence-based Synthesis), a framework that constructs standardized Findings and Impression sections directly from CT-derived imaging descriptors and routine clinical metadata, without any original radiology report, and uses them to condition a latent diffusion model adapted to 493 high-grade serous ovarian carcinoma patients. This is the first text-conditioned 3D CT synthesis framework adapted to an abdomino-pelvic oncologic setting. A systematic ablation over two adaptation axes, vision-language encoder alignment and generator fine-tuning, identifies generator domain adaptation as the operative mechanism for crossing the domain gap and establishing the target anatomy: without it, synthesis remains anchored to the thoracic pretraining domain, with Precision and Recall collapsing to zero and FID2.5D exceeding 140, regardless of encoder alignment. Encoder alignment instead refines intensity and fine detail. The full OvESyn attains the best distributional and intensity fidelity (FID2.5D 29.35, Precision 0.671, Wasserstein-1 0.044), while the generator-only variant maximizes coverage (Recall 0.645), reflecting a fidelity/coverage trade-off governed by encoder adaptation. Requiring only automatic segmentations and routine preoperative metadata, OvESyn supports transferability to report-scarce settings and provides a foundation for synthetic cohort generation in abdomino-pelvic oncologic imaging.

Computer-use agents learn from what their actions change, so training one needs applications it can act on, break and reset. The applications that matter most are login-gated and stateful, so synthetic environments stand in for them. Recent pipelines generate such environments in bulk, which moves the bottleneck from how many exist to what is inside each one. The returns, we find, come from three properties: how much behavioural depth an environment carries, whether it targets the interaction an agent actually fails, and whether it improves alongside the model. We present Echoverse, which compiles specifications into stateful applications whose tasks are graded against the application's own database, and a co-evolution loop that reads every graded rollout twice: as repairs to the environment, its tasks and its verifier, and as training signal for the model. Trained on twelve such environments, a 9B model improves from $36.5\%$ to $67.1\%$ across fourteen evaluation splits, within fourteen points of the much larger frontier model that taught it. We examine each property in turn. On the same domains, shallow environments push live-site accuracy below the base model ($80.0 \to 75.0$) while deep ones raise it ($80.0 \to 85.0$ and $48.0 \to 65.0$); drilling one interface control across many renderings transfers to held-out widget families and to the open web; and repairing a single environment lifts the model trained on it from $16.2\%$ to $38.5\%$. The same worlds serve as reinforcement-learning environments, where a reward combining the grounded verifier with a dense per-step judge raises held-out score from $58.8\%$ to $68.0\%$. We release four environments as a benchmark, with their applications, seed data and grounded graders. Code: https://aka.ms/echoverse

Joint audio--video (AV) generators are trained on data in which \emph{what an event looks like} and \emph{what it sounds like} are spuriously correlated. We present a \emph{controlled causal study} of the resulting failure mode. In an AV structural causal model where the audio is, by construction, independent of the video's nuisance appearance, models that let audio read video directly---through cross-attention or a shared latent---learn a \emph{visual shortcut}: they predict sound from appearance rather than the causal event and, when the appearance--event correlation is broken at test time, synthesize the wrong event's sound. Crucially, the popular remedy of routing both modalities through a \emph{shared common-cause latent} does \emph{not} fix this---a bottleneck, an unsupervised shared/private factorization, and a faithful shared-prior model all grab the appearance proxy and fail like the direct model. Blocking the shortcut instead requires an \emph{intervention on the nuisance}: under the stated assumptions we prove that counterfactual invariance is necessary and sufficient to identify the causal predictor, and we verify the mechanism from feature-vector SCMs to procedural pixel video, real images with spectrogram audio, moving real digits, and a conditional generator. On a \emph{real, pretrained} V2A generator (MMAudio), an input-intervention test shows the model is far from invariant to sound-irrelevant edits, though a generic-noise control reveals it is broadly input-brittle rather than specifically colour-shortcutting---clean isolation of the shortcut needs the controlled confounds our synthetic studies provide. We characterize \emph{when} the shortcut occurs, compare the objective against supervised counterfactual augmentation, and isolate the \emph{unknown-nuisance} regime---where the intervention cannot be applied---as the central open problem.

Best-arm identification is a canonical model for data-driven decision-making, but in many applications each reward observation is costly. Motivated by the growing availability of cheap predictions from machine learning and large language models, we study fixed-confidence best-arm identification in which each costly reward pull is paired with a cheap but correlated proxy score. The marginal mean of the proxy can be estimated offline and is treated as known, whereas its correlation $ρ$ with the reward, which governs how much the proxy helps, is unknown and must be learned online in pair with real rewards. We show that a control-variate adjustment turns this model into a heteroscedastic identification problem whose oracle sample complexity improves by residual variance $1-ρ^2$. The central difficulty is that the correlation must be learned from the same costly samples that identification consumes online, and that a plug-in estimate of the residual variance is anti-conservative and can compromise correctness. We propose PROBE (PRoxy OLS for Best-arm Exploration), a phase-elimination algorithm that directly maintains an upper certificate on the residual variance with an ordinary least squares fit, whose exact chi-square law keeps the certificate valid regardless of the unknown correlation. We prove that PROBE is $δ$-PAC and attains the known-correlation oracle sample complexity up to a constant multiplicative factor and a constant additive calibration cost. The guarantee extends to the $(ε,δ)$-PAC setting under minimal changes to the algorithm. Numerical experiments on synthetic instances and on an auto-loan pricing replay with large language model and tabular proxies confirm that the sample savings of PROBE scale with the strength of the reward-proxy correlation, exactly as the theory predicts.

Stereo visual SLAM systems built on local descriptors suffer from semantic ambiguity, instance-level confusion, and independently moving objects, each corrupting data association and accumulating as trajectory drift. Prevailing semantic and dynamic SLAM methods address this through binary feature rejection, sacrificing correspondence density for outlier suppression. We contend that contextual implausibility is better expressed as a graded quantity than an exclusion criterion. We present Know-Your-Scene (KYS)-SLAM, a modular extension of ORB-SLAM3 that supplants feature rejection with continuous correspondence modulation. The contribution is the reframing of contextual evidence as correspondence cost, applied within feature matching and leaving the geometric backend unmodified. Each keypoint is augmented with semantic, panoptic, and motion priors fused through a hierarchical compatibility formulation, in which semantic class and instance identity enforce structural plausibility while a zero-shot motion score down-weights features on independently moving objects. That score comes from a training-free module fitting a depth-aware ego-motion model to background optical flow and classifying panoptic segments via self-calibrating, coverage-aware thresholds, so only segments with sufficient motion evidence are penalized and static structure is left unpenalized. Penalizing correspondences rather than discarding them preserves the geometric support bundle adjustment depends on. Under one fixed configuration, no coefficient retuned per sequence or dataset, KYS-SLAM reduces per-sequence ATE RMSE by 17.4% on outdoor KITTI and 27.7% on indoor EuRoC across 21 stereo sequences with no regressions, and by 6.6% on dynamic subsets of KITTI Tracking and 17.8%, up to 31.2%, on Virtual KITTI 2 -- cross-domain transfer across outdoor driving, indoor flight, and synthetic imagery under one set of constants.

Synthetic PET images are valuable for quantitative imaging workflow development, scalable virtual imaging trials, and deep learning model training, but conventional physics-based simulation approaches are computationally intensive, limited in anatomical variability, and often fail to capture heterogeneous PET uptake. This study developed a pretrained domain-adapted diffusion (PAD) model for anatomy-conditioned PET synthesis from uniform organ activity maps. PAD adopts a natural-image pretrained text-to-image decoder with an upstream conditioning encoder and a downstream PET-domain adapter. A two-phase training strategy was used, with the first phase learning coarse uptake distributions and the second refining local image details. Uniform organ activity maps were generated from CT-based segmentations by assigning each organ its mean uptake from the paired PET image. Evaluation included quantitative accuracy, noise assessment, radiomic analysis, tumor segmentation performance, and a human observer study. PAD-generated images achieved high quantitative accuracy, with concordance correlation coefficients above 0.92 between organ mean SUVs and assigned activity values. The synthesized images showed noise levels and texture characteristics similar to target PET images and produced comparable tumor segmentation performance. In a two-alternative forced-choice observer study, four readers achieved approximately 50% accuracy, indicating visual indistinguishability between synthesized and target images. PAD also generated realistic PET images from XCAT-derived activity maps, demonstrating compatibility with phantom-based anatomical priors. Overall, PAD provides a diffusion-based framework for generating clinically relevant heterogeneous PET images from uniform organ activity maps derived from clinical segmentations or digital phantoms, supporting data augmentation and downstream imaging studies.

Significant progress has been made in detecting synthetic images, however most existing approaches operate on a single image instance and overlook a key characteristic of real-world dissemination: as viral images circulate on the web, multiple near-duplicate versions appear and lose quality due to repeated operations like recompression, resizing and cropping. As a consequence, the same image may yield inconsistent forensic predictions based on which version has been analyzed. In this work, to address this issue we propose QuAD (Quality-Aware calibration with near-Duplicates) a novel framework that makes decisions based on all available near-duplicates of the same image. Given a query, we retrieve its online near-duplicates and feed them to a detector: the resulting scores are then aggregated based on the estimated quality of the corresponding instance. By doing so, we take advantage of all pieces of information while accounting for the reduced reliability of images impaired by multiple processing steps. To support large-scale evaluation, we introduce two datasets: AncesTree, an in-lab dataset of 136k images organized in stochastic degradation trees that simulate online reposting dynamics, and ReWIND, a real-world dataset of nearly 10k near-duplicate images collected from viral web content. Experiments on several state-of-the-art detectors show that our quality-aware fusion improves their performance consistently, with an average gain of around 8% in terms of balanced accuracy compared to plain average. Our results highlight the importance of jointly processing all the images available online to achieve reliable detection of AI-generated content in real-world applications. Code and data are publicly available at https://grip-unina.github.io/QuAD/

In federated language modeling, $K$ nodes each hold $n$ samples but cannot pool data or exchange full-precision gradients or weights. We study the minimax rate at which a conditional distribution over $V$ tokens can be estimated when each node may upload at most $B$ bits per query in a public probe set. In federated probe-logit distillation (FPLD), each node transmits a scalar-quantized logit vector on the probe set, and an aggregator distills a global parametric student. Prior work (Dubey and Huo, 2026) establishes a high-probability KL rate $O(d/(Kn) + ρ\sqrt{V \log V / m} + K^{-1} \cdot 2^{-2B/V})$ plus optimization slack, with the bandwidth term in its trace-sharpened form. Whether this bandwidth-term rate is tight, and how the upper bound generalizes to heterogeneous per-node bandwidths, are left open. We close both gaps. First, the dithered FPLD construction has a matching single-round lower bound $Ω(K^{-1} \cdot 2^{-2B/V})$ under non-degeneracy, pinning the bandwidth-axis rate at $Θ(K^{-1} \cdot 2^{-2B/V})$. $T$-round sequential refinement with nested/scaled residual quantizers achieves $O(K^{-1} \cdot 2^{-2TB/V})$; vanilla FPLD's $T$-independent bandwidth term is suboptimal for every $T > 1$. Second, we establish a heterogeneous-bandwidth upper bound for per-node budgets $B_i$, paired with a closed-form optimal allocation $B_i^* = B_{\mathrm{tot}}/K + (V/2) \log_2(w_i / \bar{w}_g)$, a log-tilted water-filling rule that is the per-node analogue of reverse water-filling for distortion-rate optimization. A plug-in adaptive variant estimates the weights from a short warm-up phase and attains $1 + O(\sqrt{\log(K/δ)/(m T_0)})$ relative suboptimality. Synthetic n-gram simulations confirm that empirical KL is bracketed by the upper and lower bounds and that the optimal allocation strictly dominates uniform and inverse-weighted baselines under heterogeneous clipping.

Enterprise AI agents are typically granted static credential sets at configuration time, holding every tool the role might need for every task they perform. This persistent over-privilege expands the attack surface. We argue that capability scoping must follow a dynamic least-privilege principle and be treated as a prevention mechanism before a detection one. A credential that does not exist in an agent's context cannot be misused regardless of the agent's reasoning or evasion sophistication. We outline a three-source architecture instantiating this principle: role-based ceilings, a task-context classifier, and policy-derived combination prohibitions creating a layered proactive defense against LLM agent misalignment and misuse cases. The architecture supports both enforcing and observe-only deployment; the latter records agent permission requests inconsistent with task context, producing a behavioral signal usable in misalignment research. As a first step toward evaluating this architecture, we contribute a synthetic dataset of 600 enterprise task prompts grounded in a multi-department company policy, labeled with minimum required permissions across a 15-permission tool-based taxonomy that maps directly to deployable credentials or enforceable guardrails. The dataset is constructed via a two-pass pipeline that separates prompt generation from permission labeling to avoid circularity, and is validated against a 60-record/688 decisions human-reviewed sample (Cohen's $κ= 0.917$ pre-review and $κ= 0.967$ post-review). Iterating between dataset and policy reduced ceiling violations from 46 to 3, a 93% reduction. This shows that synthetic prompt generation can drive policy refinement when the two are developed together. The dataset, environment specification, and generation pipeline are released to support evaluation of dynamic scoping mechanisms.

Modern vision models achieve strong performance on standard benchmarks, yet their aggregate accuracy reveals little about which scene properties drive their predictions. Existing robustness benchmarks provide important stress tests, but typically manipulate global 2D image properties, rely on entangled real-world variation, or cover only a limited set of 3D objects and scene parameters. We introduce MAPS (Manifolds of Artificial Parametric Scenes), a scalable instrument for controlled attribution of vision model behavior to scene parameters. MAPS comprises 2,618 curated photorealistic 3D meshes validated for recognizability across 560 ImageNet classes and provides a Blender-based rendering pipeline for on-demand image generation under continuous variation of nine independent scene-factors spanning background, camera, and lighting, extensible to other factors. To showcase its applicability, we use MAPS to evaluate 20 convolutional and transformer-based models by quantifying their reliance on these scene factors through regression-based sensitivity analysis. We find a near-universal failure axis across all tested architectures: camera distance and elevation consistently dominate recognition failure regardless of ImageNet accuracy. However, the full sensitivity structure reveals that modern CNNs and transformers cluster together, distinct from older architectures, suggesting that fine-grained architectural design choices, rather than the coarse CNN-versus-transformer distinction, are the stronger determinant of sensitivity profiles.

Video game engines have been an important source for generating large volumes of visual synthetic datasets for training and evaluating computer vision algorithms that are to be deployed in the real world. While the visual fidelity of modern game engines has been significantly improved with technologies such as ray-tracing, a notable sim2real appearance gap between the synthetic and the real-world images still remains, which limits the utilization of synthetic datasets in real-world applications. In this letter, we investigate the ability of a state-of-the-art image generation and editing diffusion model (FLUX.2-4B Klein) to enhance the photorealism of synthetic datasets and compare its performance against a traditional image-to-image translation model (REGEN). Furthermore, we propose a hybrid approach that combines the strong geometry and material transformations of diffusion-based methods with the distribution-matching capabilities of image-to-image translation techniques. Through experiments, it is demonstrated that REGEN outperforms FLUX.2-4B Klein and that by combining both FLUX.2-4B Klein and REGEN models, better visual realism can be achieved compared to using each model individually, while maintaining semantic consistency. The code is available at: https://github.com/stefanos50/Hybrid-Sim2Real

Recent advances in deep learning for remote sensing rely heavily on large annotated datasets, yet acquiring high-quality ground truth for geometric, radiometric, and multi-domain tasks remains costly and often infeasible. In particular, the lack of accurate depth annotations, controlled illumination variations, and multi-scale paired imagery limits progress in monocular depth estimation, domain adaptation, and super-resolution for aerial scenes. We present SyMTRS, a large-scale synthetic dataset generated using a high-fidelity urban simulation pipeline. The dataset provides high-resolution RGB aerial imagery (2048 x 2048), pixel-perfect depth maps, night-time counterparts for domain adaptation, and aligned low-resolution variants for super-resolution at x2, x4, and x8 scales. Unlike existing remote sensing datasets that focus on a single task or modality, SyMTRS is designed as a unified multi-task benchmark enabling joint research in geometric understanding, cross-domain robustness, and resolution enhancement. We describe the dataset generation process, its statistical properties, and its positioning relative to existing benchmarks. SyMTRS aims to bridge critical gaps in remote sensing research by enabling controlled experiments with perfect geometric ground truth and consistent multi-domain supervision. The results obtained in this work can be reproduced from this Github repository: https://github.com/safouaneelg/SyMTRS.

Dense 3D reconstruction and tracking of dynamic scenes from monocular video remains an important open challenge in computer vision. Progress in this area has been constrained by the scarcity of high-quality datasets with dense, complete, and accurate geometric annotations. To address this limitation, we introduce Syn4D, a multiview synthetic dataset of dynamic scenes that includes ground-truth camera motion, depth maps, dense tracking, and parametric human pose annotations. A key feature of Syn4D is the ability to unproject any pixel into 3D to any time and to any camera. We conduct extensive evaluations across multiple downstream tasks to demonstrate the utility and effectiveness of the proposed dataset, including 4D scene reconstruction, 3D point tracking, geometry-aware camera retargeting, and human pose estimation. The experimental results highlight Syn4D's potential to facilitate research in dynamic scene understanding and spatiotemporal modeling.

Recent work shows that models that learn to reward hack on RL environments can become broadly misaligned, and that reframing reward hacking as acceptable behavior during training (inoculation prompting, or IP) blocks this generalization. We ask whether synthetic document finetuning (SDF) can inoculate a model against future training we don't intervene on. We add synthetic documents framing reward hacking as acceptable behavior to a model's midtraining corpus, and then train these models with RL on exploitable environments, teaching them to reward hack. Behaviorally, midtraining succeeds: models describe reward hacking favorably and are more approving of reward-hacking outputs they produce. However, they show strong EM after learning to reward hack, while IP in the same setting prevents EM. We show that SDF can predictably steer downstream generalization when inserting new associations, but struggles and has unpredictable effects when overriding existing associations, such as that between reward hacking and misalignment that produces EM. Our results suggest that, at the scales we test, SDF can make a model appear aligned with desired beliefs while steering its generalization from later training in unintended ways.

Vision-based crack inspection depends on segmentation networks whose reliability depends on the quantity, diversity and label quality of their training data. Pixel-level annotations are costly, and crack images of specific structures are scarce. Generative augmentation can supply additional data, but existing methods address isolated steps. They reuse annotated masks, offer limited control over crack geometry, and adopt the conditioning mask as the label without checking it. This paper presents an end-to-end pipeline that produces labelled crack data without manual annotation and assesses the reliability of these data and of the detectors trained on them. Procedurally sampled Bézier skeletons with guaranteed geometric properties are converted into crack masks by a generative adversarial network (GAN). A dual-ControlNet Stable Diffusion model renders the masks as crack images, either on text-described surfaces or on user-provided backgrounds. An ensemble of segmentation networks trained on real images combines its agreement with the inherited label and its internal disagreement into a pixel-wise label confidence. This confidence weights the training loss instead of removing samples with a threshold. The trained detectors are evaluated with image-space probability of detection (POD) and calibration analyses. On CRACK500 and CrackTree200, the pipeline improves five segmentation networks over conventional, diffusion-based and flow-matching-based augmentation, and on CRACK500 confidence weighting yields a higher accuracy than threshold filtering at every tested threshold. On CRACK500, the crack width that U-Net detects with 90% probability at 95% confidence decreases from 8.0 to 4.3 pixels, and the expected calibration error decreases from 14.2% to 9.6%.

Synthetic dialogue generation can support research in privacy-restricted service settings, but generated conversations must preserve communicative intent, affective meaning, and natural dialogue flow. We introduce PragAlign, a feedback-guided framework for controlled synthetic dialogue generation conditioned on service context, target intent, and target emotion, with auxiliary trait-style controls. PragAlign uses a generate--evaluate--revise loop in which an LLM-based evaluator scores intent alignment, emotion alignment, coherence, fluency, and aggregate quality, then provides criterion-specific feedback for up to three refinement rounds. On 800 matched dialogue specifications, PragAlign achieves 99.50% evaluator-defined acceptance, compared with 72.25% for one-shot generation and 95.88% for repeated generation without structured feedback. This indicates that repeated attempts account for much of the gain over one-shot generation, while structured feedback primarily improves last-mile multi-constraint satisfaction rather than broad average quality. Refinement gains are concentrated in emotion alignment, which is also the dominant failure mode in ablations. A separate human evaluation of 1,200 generated dialogues shows that intent expression and dialogue flow are highly recognizable to annotators, while emotion appropriateness is less stable and more subjective. These results support PragAlign as a quality-control framework for improving evaluator-defined communicative constraint satisfaction, while showing that affective realization and independent human-perceived quality remain open challenges.

Real-world multimodal instructions often bundle multiple requirements with unequal importance, yet most multimodal training data still reduce instruction following to answering one self-contained question. We study this gap through rubric comprehension, which casts the model not as a generator measured against rubrics but as an executor that follows them: given an image and a typed, prioritized rubric, the model must verify each rule before producing an overall judgment. To support this setting, we propose PRISM, a four-stage data synthesis framework that produces persona--task pairs, prefix-guided rule sets, quality-filtered rubrics, and structured verification traces. We further introduce PRISM-Eval, whose Loose and Strict metrics use deterministic matching against fixed labels and therefore require no inference-time judge model. With only 10K synthesized samples, PRISM lifts Qwen3-VL-4B from 9.5% to 30.1% Strict accuracy on PRISM-Eval while preserving average performance on general benchmarks, and the gains transfer to four additional open-source MLLMs across dense and MoE architectures, suggesting that structured rubric supervision is a scalable path toward multi-rule, priority-aware multimodal instruction following.

Learning generalizable robot manipulation policies requires large-scale and diverse demonstration data. Egocentric human manipulation videos offer rich scene and task diversity, and prior work has shown that retargeting and rendering such videos into robot-format data can yield effective per-task policies at small scale. However, whether this approach can provide pretraining benefits for vision-language-action models at scale remains unexplored. We present \textbf{Ego2Robot}, a scalable pipeline that converts egocentric human manipulation videos into robot training data through action retargeting, robot-arm visual synthesis, and multi-level quality curation. Ego2Robot supports both curated datasets and in-the-wild videos, producing 18,561 hours of robot training data spanning 15 robot morphologies, making it the largest ego-to-robot dataset to date. To evaluate generalization, we extend RoboTwin2.0 with disentangled perturbation axes covering visual appearance, scene layout, embodiment morphology, and task semantics. Experiments show that joint pretraining on Ego2Robot-synthesized and robot data consistently improves out-of-distribution generalization across multiple perturbation types, with benefits validated on real-robot deployment. Project page: https://www-ye.github.io/ego2robot_blog/

High-quality, diverse data are vital for large language models (LLMs) but remain scarce and costly. Data synthesis is a viable alternative and succeeds on closed tasks, yet the humanities and social sciences (HSS) are overlooked, and their open-ended nature makes synthesis challenging. Moving beyond prior capability-centric, fragmented attempts, we adopt a subject-centric paradigm, define the first HSS domain system covering 14 mainstream fields, and introduce HSS-Synth, the first data synthesis pipeline for HSS. HSS-Synth comprises: (1) constructing seed documents from web corpora via multi-step filtering and text refinement evaluated by a judge; (2) specifying "requirements + persona" to backtranslate seed documents into diverse yet faithful instructions with a strict Q&A alignment check; and (3) breaking LLM response limits via teacher-forced Answering that feeds seed documents during response generation to anchor semantics, reduce hallucinations, and preserve tone and integrity. HSS-Synth yields 237k high-quality, diverse instruction-tuning samples that outperform 14 leading baselines on 16 benchmarks. The fine-tuned Qwen3-8B-Base sets a new SOTA and approaches the official Qwen3-8B, improving both human preference and knowledge capabilities without performance seesaws. Extensive experiments demonstrate HSS-Synth's robustness and transferability. Our code is publicly available at https://github.com/pengr/HSS-Synth.

Future event prediction carries broad social impact yet remains challenging. SOTA approaches augment LLMs with external agent frameworks whose predictive capability vanishes once the harness is removed. While recent Tool-Integrated Reasoning (TIR) internalizes deep search for multi-hop retrieval of facts, forecasting further demands temporal search and reasoning over historical trends and dynamic shifts. The key obstacle is data: historical queries induce temporal leakage that degrades forecasting into retrieval. Prior works either freeze information gathering with static observations, or rely on rejection sampling or unresolved fresh queries that discard vast amounts of data, degrading synthesis efficiency. We propose a time-truncation harness that enforces a temporal cut-off at every turn, enabling TIR-style sampling from historical events, reducing temporal leakage and reliance of rejection sampling or unsolved queries, increasing the sampling efficiency. We further build a large-scale corpus and a process-based metric and show that our harness naturally induces a broader temporal breadth of search and raises the proportion of high-quality data, further increasing the efficiency and reducing the reliance on complex rubrics. Distillation experiments show that students trained on harness-intervened data achieve the best performance, demonstrating harness-assisted model evolving that turns higher quality temporal search and reasoning data into a parametric advancement of the students.

Code search enhances developer productivity by enabling efficient code reuse. Current code search systems often use a retrieve-then-rerank pipeline, where rerankers focus on modeling semantic relevance between queries and code. However, these rerankers overlook critical non-functional qualities like execution speed, memory usage, and maintainability, which are essential for practical software development. Studies reveal developers expect results to maintain high coding standards and satisfy specific needs, such as resource optimization, highlighting the importance of quality-aware code search. Achieving quality-aware code search faces two major challenges: the scarcity of quality-annotated datasets for effective training and the limitations of standard contrastive learning objectives, which fail to capture the ordinal relationships among high-quality, low-quality, and irrelevant code. Although contrastive learning excels in distinguishing relevant from irrelevant code, its binary objective does not support nuanced quality distinctions.To address these challenges, we propose SynH-Rank, a quality-aware code reranking framework that combines LLM-driven diverse data synthesis with hierarchical ranking training. SynH-Rank employs a three-level labeling scheme to explicitly model the hierarchy: high-quality relevant > low-quality relevant > irrelevant. Additionally, we introduce a new benchmark with 4,209 pairs and two novel metrics: Quality Preference Accuracy (QPA) for assessing prioritization of high-quality code and Multi-Condition Accuracy (MCA) for evaluating performance under complex constraints.Experimental results show SynH-Rank improves QPA by 20.15% over backbone models and outperforms standard relevance-only contrastive training by 15.80%, while simultaneously enhancing traditional relevance metrics and multi-condition generalizability.

Large-vocabulary instance segmentation is constrained by long-tailed category distributions and fine-grained inter-class ambiguity. While data synthesis offers a promising alternative, current paradigms have complementary limitations: text-to-image (T2I) methods inherit noisy pseudo-labels and struggle on rare classes, whereas copy-paste methods compromise contextual realism. To address these issues, we propose a hybrid pipeline coupling T2I generation with context-aware image-to-image (I2I) editing. The T2I branch provides broad category and scene diversity, while a teacher-student scheme ensures label reliability by selectively retaining only prompt-specified categories. To strengthen supervision for rare classes, we introduce VRAIN (Verified Rare-class Augmentation via INstructed editing), a novel I2I editor. VRAIN inserts high-confidence instances at semantically appropriate locations within in-the-wild scenes, yielding semantically coherent and visually natural edits that reduce domain gaps and enable targeted augmentation. On the LVIS benchmark, our method surpasses existing baselines, improving overall AP by up to +4.0 points and rare-class AP by up to +9.5 points, while scaling effectively with backbone capacity. Our project page is available at https://seokhunchoi.github.io/TMI

Visual pointing maps a language instruction to pixel co ordinates, a core skill for embodied AI. We describe our PointArena 2026 solution, which achieves 77.2% overall accuracy and ranks second on the benchmark. The ap proach targets three failure modes. First, agent-driven syn thesis builds large semantic and anchor-relative candidate pools; the server inventory contains 55,372 processed out puts, 53,772 de-duplicated sample IDs, and 37,574 train able completed or accepted rows. Second, a determinis tic steerable-data pipeline creates a verified 10,000-sample main set, plus reserve samples, using masks, templates, and path verification. Third, two model-side modules address complementary errors: AttnRes adds gated cross-block at tention for steerability, while ABC correction encodes per turbed coordinates with visual features for general coordi nate grounding. Category-aware routing combines comple mentary specialists; local validation used to select experts records 93.9% Affordance, 82.6% Spatial Relation, 78.2% Reasoning, 70.4% Counting, and 63.0% Steerability.

While image stylization has been studied extensively, video stylization remains a critical and largely unsolved challenge in the field of intelligent content creation. Existing methods, usually utilizing a reference image as the style prior, suffer from content leakage, data scarcity and limited adaptability to long videos, leading to suboptimal results with severe style drift and motion distortion. For these issues, we present EchoStyle, a scalable text-driven framework to achieve high-quality stylization of videos with arbitrary lengths. To start with, we construct a video-to-video architecture to appropriately re-fuse the video content and the text style. To address data scarcity, we pioneer an automatic reverse-synthesis pipeline to establish V-Style20k, a large-scale stylization dataset of 20k high-quality video pairs. To facilitate long video stylization, we devise an init-follow-mode mechanism along with a sliding-window inference strategy. Extensive experiments demonstrate EchoStyle's excellent performance across a wide range of artistic styles, even comparable to leading closed-source solutions.

While large language model (LLM)-based text-to-speech (TTS) systems have achieved high-quality speech synthesis, most existing systems focus on English and Chinese. Japanese, however, remains under-explored, and its unique linguistic challenges, such as widespread context-dependent kanji polyphony, have yet to be adequately tackled. Here we introduce Sarashina2.2-TTS (https://github.com/sbintuitions/sarashina2.2-tts), a Japanese-centric LLM-TTS system that tackles these challenges through a dual approach: data strategy and evaluation methodology. First, we scale training to approximately 361k hours of speech, incorporating a balanced mix of Japanese and English data. Furthermore, we design a targeted data augmentation pipeline covering all 2,136 Joyo (regular-use) kanji designated by Japan's Agency for Cultural Affairs to efficiently address kanji polyphony disambiguation. Second, we introduce the Joyo Kanji Yomi Benchmark (https://github.com/sbintuitions/JoyoKanji-Yomi-Benchmark), covering all 2,136 Joyo kanji and their 4,378 readings. Alongside this benchmark, we propose Kana-CER, a metric that compares synthesized speech against reference readings in the kana space, eliminating orthographic variations to directly measure pronunciation correctness. Experiments demonstrate that our targeted data augmentation significantly improves reading accuracy. Overall, Sarashina2.2-TTS achieves state-of-the-art kanji-level reading accuracy and matches top baselines on general sentence-level pronunciation, while delivering the highest speaker similarity in zero-shot Japanese speech synthesis. Furthermore, cross-lingual evaluation reveals that Sarashina2.2-TTS is the only system that maintains stable Japanese pronunciation regardless of the prompt language, confirming that our balanced training approach improves cross-lingual robustness.