Large Language Model (LLM) agents are increasingly deployed in multi-agent social settings where they must cooperate, negotiate, and adapt to other agents. Measuring these social skills is hard because most social interactions offer no objective ground truth. Evaluations often fall back on LLM judges, which are costly, subjective, and noisy. We introduce SocialGym, an environment of 22 multi-agent social games, from matrix games to hidden-role deduction, with verifiable outcomes determined by game rules. We benchmark ten LLMs across all games and roles using an Elo tournament. Our results show that no model is best in every game, and rankings shift across game families and sides. Secret keeping correlates more strongly with theory-of-mind and privacy benchmarks than with general knowledge benchmarks. Half of the models give away their side's secret in Chameleon and Spyfall, and the smallest models are not reliably better than chance at voting for hidden opponents in most hidden-role games. Finally, we examine whether models can improve their play by reflecting on self-play and writing strategy playbooks, finding mixed results across models and games. SocialGym provides verifiable measures for diagnosing LLM social capabilities across settings and roles.
Auto-research agents now run machine-learning development unattended for hours, revising data pipelines, models, and validation from their own feedback, yet on most competitions they still finish below strong human competitors. Outcome-based benchmarks record this gap but not its cause, because they grade the final submission and discard the development process behind it. We introduce TraceML, which pairs human and agent work on the same competitions under one version-level schema: 4,465 human Kaggle trajectories across 134 competitions, seven of which are also worked by two agent scaffolds, giving 430 paired human and 207 agent trajectories. Every code version carries its score, its timestamp, and labels for the action taken, its intent, the edit size, and the score effect. Read this way, the gap becomes concrete. Experts alternate data work, validation, model changes, and ensembling, and return to approaches they had set aside. Each agent scaffold instead collapses into a narrow loop: Codex spends its steps re-weighting ensembles and tuning submissions, MLEvolve mutates its model in place, and neither pivots at the human rate nor reopens abandoned work. A short planning prompt distilled from human practice moves the behaviors it names toward the human profile and lifts scores, but the effort profile stays agent-shaped: instruction closes only the part of the gap that reduces to instructions. We release the corpus, the schema, and the labeling models, together with an open-source toolkit that turns a run from any command-line agent into a TraceML trajectory and reads it against the human cohorts.
Deformable linear objects such as wires and cables are difficult to segment because they are thin, highly deformable, and frequently self-occluded, while large-scale instance-level annotations are expensive to obtain in real scenes. Existing resources either focus on cable tracing or semantic segmentation under constrained settings, or generate visually plausible images without physically grounded wire deformation. We present WireSeg-32k, a synthetic dataset for wire instance segmentation with 32,000 RGB images, instance masks, depth maps, and a complementary real-world test set with annotations. To generate this dataset, we develop DeformX, a co-simulation pipeline that couples Cosserat-rod dynamics with photorealistic Isaac Sim rendering, enabling physically plausible, contact-consistent wire shapes, CAD-based wire assets, and diverse visually grounded scenes. As a simple baseline, LoRA fine-tuning SAM3 on WireSeg-32k alone improves real-world mAP@75 by 10.2% over the off-the-shelf model, showing that physically grounded synthetic data can transfer to real wire perception.
Language models are increasingly trained to "reason" before answering users' queries, outputting hundreds of intermediate tokens before a final answer. While reasoning is designed with direct model capabilities in mind, we argue that reasoning language models (RLMs) should also be assessed by interaction effects between their reasoning traces and weaker models. In contexts like safety monitoring and distillation, the performance of weak models increasingly depends on legibility--how thoroughly and precisely a reasoning trace externalizes an RLM's actions. Existing efficiency-based metrics for legibility fail to capture "thoroughness", instead focusing on conciseness. Thus, we introduce transfer utility, a method for measuring trace legibility derived from interactions where weak models finish tasks given incomplete traces. Evaluating 84k traces from 12 RLMs across three tasks, we find that reasoning traces generated by the highest-performing and the most efficient models rank the lowest for transfer utility. We also find evidence that transfer utility predicts the ability of weak monitors to verify procedurally dense traces for math or logical problem-solving, though the effect moderates when verifying fact-driven traces. Finally, we show that open-source reward models decouple from transfer utility signals on correctness. Together, these findings surface status quo trade-offs in weak-to-strong legibility, motivating an interaction-first framework for future applications.
Existing low-light image enhancement methods often suffer from local exposure imbalance and color distortion under complex, non-uniform illumination, while mainstream Vision Transformers lack explicit modeling of physics-based priors for illumination degradation. To address these limitations, we propose GLFS, a Gaussian Light Field Splatting-based Vision Transformer that incorporates continuous light-field priors from 2D Gaussian Splatting (2DGS) as an explicit physical bias into the Transformer self-attention mechanism to guide spatial gain estimation under complex non-uniform illumination. Specifically, GLFS represents scene illumination as a superposition of anisotropic Gaussian basis functions and introduces a physics-informed bias into the self-attention mechanism to adaptively infer a spatial gain field, enabling accurate and spatially uniform restoration under complex illumination conditions. To mitigate color shifts and structural degradation during enhancement, we further introduce a color vector angular loss and a luminance edge loss, with the former enforcing hue consistency and the latter improving the structural fidelity of local details. Extensive ablation studies and quantitative evaluations demonstrate that GLFS achieves competitive overall performance in illumination correction, structure preservation, and perceptual quality, validating the effectiveness of the continuous Gaussian light field prior for low-light image enhancement.
As multimodal conversational systems increasingly engage in spoken interaction, their ability to navigate paralinguistic social cues has become a critical bottleneck for natural human-AI communication. However, existing evaluations of machine emotional intelligence assess reasoning exclusively through isolated text or passive acoustic perception, overlooking the complex cross-modal reasoning required for active, multi-turn dialogue. We introduce \textsc{SpeechEQ}, a comprehensive framework designed to evaluate the sociolinguistic reasoning of Speech-Language Models (SLMs). The framework includes a validated dataset of 2,265 dialogues across 15 Emotional Quotient (EQ) subscales grounded in EQ-i 2.0 theory, along with a multi-turn evaluation protocol measured by our proposed Spoken EQ (SEQ) score inspired by human EQ assessments. Experiments show limitations in how both existing Speech Emotion Recognition and end-to-end Speech-Language Models understand and apply paralinguistic cues through speech. While end-to-end architectures outperform cascaded systems, \textsc{SpeechEQ} reveals that current multimodal models remain bottlenecked by a text-reliant ``modality shortcut,'' an alignment-induced ``safety trap,'' and ``contextual amnesia,'' highlighting the barriers to truly emotionally aware AI. Our benchmark can be accessed at https://huggingface.co/datasets/SpeechEQ/SpeechEQ and demo page at https://speecheq.github.io
Social dynamics encode the process in which individual network and discourse interactions aggregate into collective influence, narrative dominance and coordinate behavior. This paper uses the the GhostField architecture, a hybrid LLM-Enabled Agent Based Network-Dynamic (LAND) model as a social simulation framework to build the AuraSight scenario. In the AuraSight scenario, 314,244 heterogeneous cyber social agents and human actors exchange 529,327 messages over 30 days surrounding a fictional international song-writing contest. We methodologically examine emergent social dynamics across four analytical layers: ego-network topology, semantic network evolution, coordination dynamics and influence dynamics. Our results show how generated social simulations do also produce social dynamics, and how the dynamics of coordination and influence emerge not from individual agents but from the recursive interaction between network topology and narrative exchange.
Imperfections in AI-generated code require that software developers modify generations through manual code edits. However, little is known about how developers edit AI-generated code, and to what extent LLMs can infer implicit developer intent from this data, given the lack of high-quality, realistic code editing data. This information is essential for designing and building usable AI-developer interactions. To address this gap, we introduce DECODE (Developer Edits of Code Dataset), a dataset of 53.6K real-world in-IDE code edits of AI-generated code sourced from 1K+ developers. First, we use DECODE to analyze when, why, and how AI-generated code is edited. Developers invest little effort in editing code and abandon AI-generated code quickly, thereby motivating the need for LLMs to understand developer intents through their edits. Therefore, we next use DECODE to understand to what extent LLMs could infer developer edits. Predicting code edits is a challenging task for both frontier and fine-tuned models. Frontier models generate edits that best reflect developer intents, but are prone to over-editing and reverting developer-written edits. Fine-tuning LLMs can improve their ability to predict edits that are more similar to developers' as well as better reflect developer intent, but do not fully bridge the gap to frontier models. Together, our results demonstrate the unique insights on developer and LLM behavior that can be obtained through studying realistic code edit data to promote more better developer-AI interactions.
Most modern optimizers form their momentum as an exponential moving average (EMA) of past gradients, forgetting every direction at one fixed rate. However, the inputs a deep network sees during training can be highly anisotropic, with a few directions queried frequently while most are seen rarely. Preconditioning methods address this anisotropy by wrapping extra processing around this buffer and leave the momentum update itself unchanged. We propose Activation-Keyed Momentum (AK-Momentum), which builds direction-awareness into the momentum update rule. The gradient of a linear layer splits into an input activation that acts as a $\textit{key}$ and an output-side error that acts as a $\textit{value}$. Keying on that activation, AK-Momentum updates the momentum buffer by the canonical delta rule, so each direction is forgotten at a rate set by how often it appears. We prove that it is a valid momentum, that it applies the input-side curvature correction without matrix inversion, and that it clears stale directions faster than EMA under both a fixed and a drifting optimum. It is a drop-in replacement for the momentum buffer of any optimizer, its coefficient transfers across widths under $\mu$P, and its extra compute stays between $22.2\%$ and $25.0\%$ of a gated-MLP block's linear cost with no persistent memory. In FineWeb-Edu pretraining, AdamW with AK-Momentum (AK-AdamW) reaches AdamW's validation loss in up to $46.64 \pm 4.52\%$ fewer steps at 67M and $22.14 \pm 0.81\%$ at 370M over three seeds, and the gain persists at 1B on a Chinchilla-optimal budget. A Muon baseline tuned under the same protocol sits above AK-AdamW at both language-model scales, and the gain holds for SGD, ResNet-18, and ViT-Tiny on CIFAR-10. Training-time diagnostics confirm the predicted mechanism, better gradient tracking and healthier input directions. Code is available at https://github.com/euijinh/AK-Momentum.
Learning visuomotor policies via imitation has proven effective across a wide range of robotic domains. However, the performance of these policies is heavily dependent on the number of training demonstrations, which requires expensive data collection in the real world. In this work, we aim to reduce data collection efforts when learning visuomotor robot policies by leveraging existing or cost-effective data from a wide range of embodiments, such as public robot datasets and the datasets of humans playing with objects (human data from play). Our approach leverages two key insights. First, we use optic flow as an embodiment-agnostic action representation to train a World Model (WM) across multi-embodiment datasets, and finetune it on a small amount of robot data from the target embodiment. Second, we develop a method, Latent Policy Steering (LPS), to improve the output of a behavior-cloned policy by searching in the latent space of the WM for better action sequences. In real world experiments, we observe significant improvements in the performance of policies trained with a small amount of data (over 50% relative improvement with 30 demonstrations and over 20% relative improvement with 50 demonstrations) by combining the policy with a WM pretrained on two thousand episodes sampled from the existing Open X-embodiment dataset across different robots or a cost-effective human dataset from play.
Construction of confidence intervals and hypothesis tests for functionals based on asymptotically normal estimators is a classical topic in statistical inference. The simplest and in many cases optimal inference procedure is the Wald interval or the likelihood ratio test, both of which require an estimator and an estimate of the asymptotic variance of the estimator. Estimators obtained from online/sequential algorithms forces one to consider the computational aspects of the inference problem, i.e., one cannot access all of the data as many times as needed. Several works on this topic explored the online estimation of asymptotic variance. In this article, we propose computationally efficient, rate-optimal, and asymptotically valid confidence regions based on the output of online algorithms {\em without} estimating the asymptotic variance. As a special case, this implies inference from any algorithm that yields an asymptotically normal estimator. We focus our efforts on stochastic gradient descent with Polyak averaging to understand the practical performance of the proposed method.
Sanitizing sensitive text data typically involves removing personally identifiable information (PII) or generating synthetic data under the assumption that these methods adequately protect privacy; however, their effectiveness is often only assessed by measuring the leakage of explicit identifiers but ignoring nuanced textual markers that can lead to re-identification. We challenge the above illusion of privacy by proposing a new framework that evaluates re-identification attacks to quantify individual privacy risks upon data release. Our approach shows that seemingly innocuous auxiliary information -- such as routine social activities -- can be used to infer sensitive attributes like age or substance use history from sanitized data. For instance, we demonstrate that Azure's commercial PII removal tool fails to protect 74\% of information in the MedQA dataset. Although differential privacy mitigates these risks to some extent, it significantly reduces the utility of the sanitized text for downstream tasks. Our findings indicate that current sanitization techniques offer a \textit{false sense of privacy}, highlighting the need for more robust methods that protect against semantic-level information leakage.
We study structured Stackelberg games, in which both players (the leader and the follower) observe information about the state of the world at time of play. Importantly, this information may contain information about the follower, which the leader may use when deciding her strategy. Under this setting, we show that no-regret learning is possible if and only if the set of mappings from contexts to follower types that the leader uses to learn is not ``too complex''. Specifically, we find that standard learning theoretic measures of complexity do not characterize learnability in our setting and we give a new dimension which does, which we term the Stackelberg-Littlestone dimension. In the distributional setting, we give analogous results by showing that standard complexity measures do not characterize the sample complexity of learning, but a new dimension called the Stackelberg-Natarajan dimension does. We then show that an appropriate empirical risk minimization procedure achieves the corresponding sample complexity.
Solving (Stampacchia) variational inequalities (SVIs) is a foundational problem at the heart of optimization, with a host of critical applications ranging from engineering to economics. However, this expressivity comes at the cost of computational hardness. As a result, most research has focused on carving out specific subclasses that elude those intractability barriers. A classical property that goes back to the 1960s is the Minty condition, which postulates that the Minty VI (MVI) problem -- the weak dual of the SVI problem -- admits a solution. In this paper, we establish the first polynomial-time algorithm -- that is, with complexity growing polynomially in the dimension $d$ and $\log(1/\epsilon)$ -- for solving $\epsilon$-SVIs for Lipschitz continuous mappings under the Minty condition. Prior approaches either incurred an exponentially worse dependence on $1/\epsilon$ (and other natural parameters of the problem) or made overly restrictive assumptions -- such as strong monotonicity. To do so, we introduce a new variant of the ellipsoid algorithm wherein separating hyperplanes are obtained after taking a gradient descent step from the center of the ellipsoid. It succeeds even though the set of SVIs can be nonconvex and not fully dimensional. Moreover, when our algorithm is applied to an instance with no MVI solution and fails to identify an SVI solution, it produces a succinct certificate of MVI infeasibility. We also show that deciding whether the Minty condition holds is $\mathsf{coNP}$-complete. We provide several extensions and new applications of our main results. Specifically, we obtain the first polynomial-time algorithms for i) solving monotone VIs, ii) globally minimizing a (potentially nonsmooth) quasar-convex function, and iii) computing Nash equilibria in multi-player harmonic games.
As language model (LM) outputs get more and more natural, it is becoming more difficult than ever to evaluate their quality. Simultaneously, increasing LMs' "thinking" time through scaling test-time compute has proven an effective technique to solve challenging problems in domains such as math and code. This raises a natural question: can an LM's evaluation capability also be improved by spending more test-time compute? To answer this, we investigate employing reasoning models-LMs that natively generate long chain-of-thought reasoning-as evaluators. Specifically, we examine methods to leverage more test-time compute by (1) using reasoning models, and (2) prompting these models to evaluate not only the response as a whole (i.e., outcome evaluation) but also assess each step in the response separately (i.e., process evaluation). In experiments, we observe that the evaluator's performance improves monotonically when generating more reasoning tokens, similar to the trends observed in LM-based generation. Furthermore, we use these more accurate evaluators to rerank multiple generations, and demonstrate that spending more compute at evaluation time can be as effective as using more compute at generation time in improving an LM's problem-solving capability.
Physics-informed machine learning provides an approach to combining data and governing physics laws for solving complex partial differential equations (PDEs). However, efficiently solving PDEs with varying parameters and changing initial conditions and boundary conditions (ICBCs) with theoretical guarantees remains an open challenge. We propose a hybrid framework that uses a neural network to learn B-spline control points to approximate solutions to PDEs with varying system and ICBC parameters. The proposed network can be trained efficiently as one can directly specify ICBCs without imposing losses, calculate physics-informed loss functions through analytical formulas, and requires only learning the weights of B-spline functions as opposed to both weights and basis as in traditional neural operator learning methods. We provide theoretical guarantees that the proposed B-spline networks serve as universal approximators for the set of solutions of PDEs with varying ICBCs under mild conditions and establish bounds on the generalization errors in physics-informed learning. We also demonstrate in experiments that the proposed B-spline network can solve problems with discontinuous ICBCs and outperforms existing methods, and is able to learn solutions of 3D dynamics with diverse initial conditions.
In the past decade, motivated by the putative failure of naive self-play deep reinforcement learning (DRL) in adversarial imperfect-information games, researchers have developed numerous DRL algorithms based on fictitious play (FP), double oracle (DO), and counterfactual regret minimization (CFR). In light of recent results of the magnetic mirror descent algorithm, we hypothesize that simpler generic policy gradient methods like PPO are competitive with or superior to these FP, DO, and CFR-based DRL approaches. To facilitate the resolution of this hypothesis, we implement and release the first broadly accessible exact exploitability computations for four large games. Using these games, we conduct the largest-ever exploitability comparison of DRL algorithms for imperfect-information games. Over 5600 training runs, FP, DO, and CFR-based approaches fail to outperform generic policy gradient methods. Code is available at https://github.com/nathanlct/IIG-RL-Benchmark and https://github.com/gabrfarina/exp-a-spiel .
Large Language Models have proved surprisingly effective at solving a wide range of tasks from just a handful of examples. However, their lack of reliability and modularity limits their capacity to tackle large problems that require many steps of reasoning. In response, researchers have proposed advanced pipelines that leverage domain-specific knowledge to chain smaller prompts, provide intermediate feedback and improve performance through search. However, the current complexity of writing, tuning, maintaining and improving such pipelines has limited their sophistication. We propose oracular programming, a foundational paradigm for building LLM-enabled applications that lets domain experts express high-level problem-solving strategies as programs with unresolved choice points. These choice points are resolved at runtime by LLMs, which generalize from user-provided examples of correct and incorrect decisions. An oracular program is composed of three orthogonal components: a strategy that consists in a nondeterministic program with choice points that can be reified into a search tree, a policy that specifies how to navigate this tree with the help of LLM oracles, and a set of demonstrations that describe successful and unsuccessful search tree navigation scenarios across diverse problem instances. Each component is expressed in a dedicated programming language and can be independently improved or substituted. We address the key programming language design challenges of modularly composing oracular programs and enforcing consistency between their components as they evolve.
The rapid deployment of distributed energy resources (DER) has introduced significant spatio-temporal uncertainties in power grid management, necessitating accurate multilevel forecasting methods. However, existing approaches often produce overly conservative uncertainty intervals at individual spatial units and fail to properly capture uncertainties when aggregating predictions across different spatial scales. This paper presents a novel hierarchical spatio-temporal model based on the conformal prediction framework to address these challenges. Our approach generates circuit-level DER growth predictions and efficiently aggregates them to the substation level while maintaining statistical validity through a tailored non-conformity score. Applied to a decade of DER installation data from a local utility network, our method demonstrates superior performance over existing approaches, particularly in reducing prediction interval widths while maintaining coverage.
Beyond high-fidelity image synthesis, diffusion models have recently exhibited promising results in dense visual perception tasks. However, most existing work treats diffusion models as a standalone component for perception tasks, employing them either solely for off-the-shelf data augmentation or as mere feature extractors. In contrast to these isolated and thus sub-optimal efforts, we introduce a unified, versatile, diffusion-based framework, Diff-2-in-1, that can simultaneously handle both multi-modal data generation and dense visual perception, through a unique exploitation of the diffusion-denoising process. Within this framework, we further enhance discriminative visual perception via multi-modal generation, by utilizing the denoising network to create multi-modal data that mirror the distribution of the original training set. Importantly, Diff-2-in-1 optimizes the utilization of the created diverse and faithful data by leveraging a novel self-improving learning mechanism. Comprehensive experimental evaluations validate the effectiveness of our framework, showcasing consistent performance improvements across various discriminative backbones and high-quality multi-modal data generation characterized by both realism and usefulness.