Scientific figures give research ideas a visual form through choices of abstraction, composition, and emphasis. Generative AI offers new ways to realize these choices, raising questions about explanatory clarity and purposeful visual expression. We examine these questions through a visual retrospective of 371 motivation and framework figures from 200 papers. The corpus comprises a quota-based main sample of 160 papers spanning four temporal periods and a separate purposive style-contrast sample of 40 papers. A single multimodal judge evaluates six dimensions of communication and design, with three repeated judgments per figure. We attempt source-constrained revisions of all 371 figures, of which 369 pass the recorded fidelity checks. Original-figure ratings show no uniform decline across periods. Revisions increase mean clarity ratings by 0.428 points and decrease aesthetic coherence ratings by 0.111 points on a five-point scale. Gains in the six-dimension average are larger for main-sample motivation figures than for frameworks. Case analyses show that added explanation can clarify comparisons while leaving typographic hierarchy and visual economy unresolved. These model-based findings support researcher-led design, with AI assisting exploration and authors retaining responsibility for explanatory purpose, composition, and scientific relationships. The accompanying atlas and evaluation records are available at https://ntu-ai4x.github.io/Rethinking_Figures/index_en.html.
The Forward-Forward (FF) algorithm trains networks layer-by-layer using a local "goodness function," yet the choice of activity statistic remains an important design question. We systematically study how selectivity and activity shape affect FF learning, introducing selective goodness functions based on top-$k$ and entmax-weighted energy alongside shape-sensitive objectives based on normalized higher-order moments. In particular, we propose burstiness goodness, a parameter-free, scale-invariant excess-kurtosis objective that rewards heavy-tailed activity patterns. Our study examines goodness functions together with activation and label-injection choices across six small-image datasets, and uses controlled experiments to isolate their effects. In validation-tuned, five-seed Fashion-MNIST experiments with matched ReLU activation, standard label injection, and per-layer scoring, burstiness improves accuracy from $73.2\pm0.9\%$ to $88.6\pm0.4\%$ over sum-of-squares. On CIFAR-10, burstiness improves block-local ResNet-18 accuracy from $28.9\pm0.5\%$ to $46.8\pm2.5\%$ under the same separate per-block label pathway. These results show that activity-shape statistics can provide effective local learning signals across the tested settings. Code is available at https://github.com/suayptalha/ff-selectivity-shape.
Vision-Language-Action (VLA) models can perform a wide range of robotic manipulation tasks, but often fail during deployment. Teaching them to recover from these failures requires substantial corrective data: existing approaches learn to recover from failures by directly associating corrective actions with the observations in which they were demonstrated. For instance, DAgger-like fine-tuning can require tens to hundreds of interactions, yet the resulting policy may still fail to generalize the learned corrections beyond the demonstrated scenarios. We hypothesize that semantics can capture the underlying reasons for failures and their corrections, enabling policies to learn from only a handful of corrective demonstrations and generalize those corrections to novel objects, and scene variations. We propose OopsieRecover, a framework that uses abduction to infer the intent behind a small set of corrective demonstrations and encodes the inferred knowledge as dense spatial representations and conditional costs. These costs capture when a behavior is undesirable and steer the policy toward more desirable action regions. We experiment across 6 tasks and variations in object poses, scenes, and related tasks, and find that only a handful of corrective demonstrations can enable a VLA to overcome failures across a broad range of novel situations, without requiring new corrective data for each variation. Please refer to https://robin-lab.cs.utexas.edu/oopsierecover for videos.
Yixiao Wang, Peiyi Chen, Mustafa Mukadam, Shreekant Gayaka, Masayoshi Tomizuka, Son Tran
Learning dexterous manipulation from limited visual demonstrations is challenging because policies must acquire both perceptual grounding and precise control, while reinforcement learning (RL) in simulation can acquire rich control behaviors from large-scale interaction with privileged information. We introduce \textbf{Latent RL Prior}, a two-stage framework that reuses this control knowledge for data-efficient visual policy learning. During privileged pretraining, we distill large-scale RL rollouts into a flow-matching action decoder conditioned on proprioception and latent task context. During visual adaptation, we freeze the decoder and learn a visual interface that maps images to its latent context using action supervision alone, without privileged labels or latent targets. The same decoder can further be shared across arms by learning coordinated visual contexts for bimanual manipulation. Across three simulated tasks, our method outperforms full-parameter fine-tuning of $π_{0.5}$ and GR00T N1.7 in the low-data regime; on door lifting and bimanual assembly, it surpasses full-data baselines using $10\times$ and $5\times$ fewer demonstrations, respectively. Further analysis shows that direct visual-to-latent conditioning improves adaptation performance and training stability over routing visual predictions through the pretrained state encoder. Robustness to unseen mass variation and greater linear accessibility of contact-force information in decoder representations further provide evidence of retained RL-acquired control knowledge. Project website: https://yixiaowang7.github.io/latent-rl-prior-page/.
Adversarial Training (AT) is an important method for improving model robustness. We study the dynamics of deterministic Gradient Descent (GD) in AT. Most existing works in this direction focus on the small or vanishing step sizes regime, where GD resembles a continuous-time flow. However, such step sizes can make training inefficient and are rarely used in practice. In this paper, we study the AT dynamics of GD with large constant step sizes. We first study GD in adversarial linear logistic regression, focusing on two aspects: where the model parameters converge (implicit bias) and how quickly the losses decrease (optimization). We then turn to neural networks training and study how the sharpness of losses evolve during AT. Our results advance the understanding of the behaviors of GD with large step sizes in AT and highlight three key differences from its behavior in Standard Training (ST): (i) the implicit bias gap between AT and ST, previously established only for small step sizes, persists for arbitrarily large constant step sizes, (ii) different from ST, there is no uniform rate of convergence for the optimization error in AT, and (iii) sharpness can exhibit two-stage dynamics that differ from the Edge-of-Stability behavior observed in ST.
Yu-Cheng Chou, Shihao Wang, Jin Huang, Chi-Pin Huang, Chiao-An Yang, Xianzhe Fan, Jing Wang, Zhiqi Li 等 19 位作者
Models for physical AI tasks, such as robotics and autonomous driving, need not reason solely in language. They can use visual plans--points, bounding boxes, and trajectories--to represent spatial goals and anticipate motion. However, autoregressive vision-language models generate these plans one coordinate token at a time, making richer predictions increasingly costly. We introduce VFlash, a method for fast, accurate visual planning through Flash Primitive Decoding. VFlash exploits the known structure of visual primitives to predict their coordinates jointly, using a lightweight drafter that generates an entire primitive in one forward pass. The layout adapts to the primitive's type, dimensionality, and length, supporting compact boxes and long, variable-length trajectories. Crucially, spatial accuracy need not require reproducing an autoregressive model's exact token sequence. VFlash therefore trains on annotated primitives and directly accepts structurally valid predictions without target-token agreement. Across the evaluated 2D and 3D robot trajectory tasks, VFlash consistently reduces prediction errors relative to autoregressive decoding. Preliminary single-rollout measurements indicate approximately 25x faster generation for 3D point motion forecasting. A world action model conditioned on VFlash plans achieves 70.2% average success across four MolmoSpaces task suites, compared with 62.6% for the baseline.
Neural networks are likely to be among the most consequential artefacts in human history. As these systems become more capable and autonomous, their potential to cause societal harm also grows, making AI safety one of the defining challenges of our time. Yet we continue to build increasingly powerful systems without fully understanding how they work, how they will behave in unfamiliar situations, or how to reliably control them. Mechanistic interpretability offers a path towards understanding these systems by studying their internal representations and computations, with the aim of evaluating, predicting, and controlling behaviour. We survey mechanistic interpretability for AI safety in order to answer the question: Does understanding how models work actually make them safer? We place mechanistic interpretability in the broader context of interpretability research and compare its mechanistic evidence with behavioural evaluation. We introduce the foundations of mechanistic interpretability, including features, polysemanticity, transformer computation, and circuits, and survey methods for feature extraction and analysis, circuit discovery and causal verification, explaining intermediate representations in natural language, learning dynamics, and training-data attribution. Beyond surveying existing methods, we propose a unified framework for evaluating mechanistic interpretability claims and characterise six research orientations by their goals, assumptions, and trade-offs. Finally, we examine how mechanistic understanding can support AI safety, including knowledge editing, steering, and monitoring, and assess its limitations, failure modes, open problems, and relationship to behavioural safety measures. We also release a repository of code notebooks with worked examples that reproduce key findings from surveyed methods, allowing readers to build intuition through hands-on experimentation.
Junhyuk Kwon, Seungjoon Lee, Hyejin Park, Kyle Min, Jungseul Ok
In human-robot interaction, agentic robots should resolve ambiguous user requests through proactive clarification while keeping user burden low. We study this challenge in interactive instance navigation, where an agent must identify and reach the user-intended target among similar objects. Existing approaches ask open-ended questions and select a candidate that strongly matches the accumulated user-provided information. However, this process can require lengthy user responses and lead the agent to make premature decisions without distinguishing the target from distractors. To address these limitations, we propose Proactive Instance Navigation with Comparative Judgment (ProCompNav), a two-stage framework that first constructs a candidate pool and then identifies the target through Recursive Comparative Judgment (RCJ). RCJ iteratively narrows the pool by selecting an attribute-value pair that divides the candidates, asking the user a binary question, and removing inconsistent candidates, without requiring an attribute unique to the target. ProCompNav substantially improves Success Rate and reduces Response Length on an interactive navigation benchmark, and achieves the highest Success Rate among evaluated methods on a non-interactive benchmark. In two human studies, participants preferred ProCompNav's binary questions to open-ended questions and its batched delivery to interleaved interaction. Our results highlight the need to jointly consider task performance and user-friendly interaction strategies when designing agentic robots.
Michal P. Podolinsky, Yunhao Yang, Daniel Little, Hong Lu, Rohan Siva, Olin Goudey, Neel P. Bhatt, Matthias Scheutz 等 9 位作者
Vision-Language-Action models (VLAs) enable general-purpose embodied intelligence, but their performance is brittle under distribution shift, where unanticipated failures can damage equipment or endanger people. Deploying VLAs in real-world systems therefore requires a runtime monitor that is timely, calibrated, and informative while being efficient enough to run on the edge. Existing methods fail to fulfill these requirements. We present VLA-Tug, a lightweight probe over the latent layers of a VLA backbone that improves early failure detection and bounds the missed-detection and false-alarm rates, with minimal inference overhead. We show that supervising on the pre-failure segment of a trajectory increases early failure detection at matched false-positive rates by up to 30% over the strongest learned baseline. Under shift to unseen tasks, the bounds on missed-detection and false-alarm rates remain valid in up to 86% of runs, versus at most 21% for baselines. Unlike existing methods, VLA-Tug predicts the failure mode to enable corrective actions, raising top-3 accuracy on unseen tasks by up to 44% over a majority baseline. VLA-Tug enables real-time monitoring where the well-timed and informative probe supports lightweight steering that improves task success. We release a dataset of 3,180 annotated manipulation episodes to support future work on latent failure detection in VLAs at https://vla-tug.github.io.
Mohammed Hassan, Zhenyang Chen, Zheng Wang, Zhixin Zhu, Tianlong Chen, Yingyan, Lin, Chaojian Li
Vision-Language-Action (VLA) models enable end-to-end robot control, but inference is increasingly expensive. Each decision processes hundreds of visual tokens, often across multiple cameras, leading to substantial computation while allowing irrelevant background content to distract action prediction under distribution shift. We argue that these issues stem from two complementary forms of redundancy: many visual tokens are spatially irrelevant, while attention over the remaining tokens is often misallocated. Based on this observation, we propose T3R, a Training-Free Two-Stage Token Refinement framework. In the first stage, segmentation-guided token pruning removes irrelevant visual patches to reduce computation and eliminate distracting context. In the second stage, saliency-guided attention biasing reallocates attention toward action-critical visual regions and language instructions within the retained tokens. Together, pruning produces a cleaner candidate set, while attention biasing further emphasizes the most decision-relevant evidence, yielding a better accuracy-efficiency frontier than either stage alone. On LIBERO-Spatial, T3R removes 85% of scene-camera visual tokens while essentially matching the unpruned success rate, and improves success by up to 7 percentage points under out-of-distribution object-placement perturbations. The timing comparison attributes 7.8 ms per inference call to pruning and attention-bias application. As input resolution increases, LLM-forward profiling shows up to a 3.3x speedup over the unpruned baseline. Evaluations across LIBERO, SIMPLER, and RoboTwin with three VLA backbones show the strongest accuracy-efficiency benefits in multi-camera settings. These results demonstrate that training-free token refinement is a practical approach to efficient and robust VLA inference.
Yuanchuan Guo, Jiashun Jin, Zheng Tracy Ke, Bingcheng Sui
The growing use of AI in mathematics raises a question: can we detect AI involvement in proofs, separately from AI assistance with writing? Historical arXiv papers, recent papers disclosing substantive AI use, and manuscripts released by OpenAI provide a rich empirical basis for studying this question. However, several challenges remain: lack of matched case-control pairs; potential incompleteness of disclosure; and lack of ready-to-use features designed to capture signals of AI involvement in proofs. To address these challenges, we construct five cohorts comprising 8,281 papers by combining existing manuscripts with those we produced using generative AI. These cohorts span different levels of AI involvement and include matched pairs of papers addressing the same research problems. We then introduce ProofTrace, a framework that represents each paper as a graph of immediate proof support (GIPS) and extracts interpretable features of mathematical organization. Our study yields three discoveries. First, principal-component projections of our features reveal a qualitative gradient of AI involvement, where AI-reworded papers remain close to human papers, the two AI-generated cohorts depart in broadly shared directions, and papers disclosing substantive AI use occupy intermediate positions which shifts toward AI-generated cohorts over time. Second, we identify recurring signatures of AI-generated proofs, such as fewer externally cited results, greater reliance on intermediate conclusions, and altered graph connectivity; many of these signatures persist under controls for research field, manuscript length, or graph size. Finally, in a separate detection experiment, ProofTrace distinguishes human-written papers from those disclosing substantive AI use. Its performance remains stable under AI rewording, whereas the performance of state-of-the-art text-based detectors declines sharply.
Xiang Gao, Haresh Rengaraj Rajamohan, Weicheng Zhu, Shih-Lun Huang, Long Chen, Gabe Schulman, Huizhen Jin, Shengduo Li 等 13 位作者
While large-scale pretraining has revolutionized language modeling, its potential remains underexplored in healthcare with structured electronic health records (EHRs). We present RAVEN, a novel generative pretraining strategy for sequential EHR data based on Recurrence-Aware next-Visit EveNt prediction. Leveraging a dataset of over one million unique individuals, our model learns to autoregressively generate tokenized clinical events for the next visit conditioned on patient history. We introduce regularization on predicting repeated events and highlight a key pitfall in EHR-based foundation model evaluations: repeated event tokens can inflate performance metrics when new onsets are not distinguished from subsequent occurrences. Furthermore, we empirically investigate the scaling behaviors in a data-constrained, compute-saturated regime, showing that simply increasing model size is suboptimal without commensurate increases in data volume. We evaluate our model via zero-shot prediction for forecasting the incidence of a diverse set of diseases, where it rivals fully fine-tuned representation-based Transformer models and outperforms both standard simulation-based next-token approaches and a prompted medical large language model baseline. Finally, without additional parameter updates, we show that RAVEN can generalize to an external patient cohort under lossy clinical code mappings and feature coverage gaps.
Paired multi-omic cancer cohorts are costly to assemble and difficult to share, motivating synthetic cohorts as a complementary resource. As generative model development accelerates, we shift the focus from proposing yet another architecture to defining evaluation criteria that determine whether synthetic data preserve biologically relevant structure. We introduce a biology-informed and patient-disjoint framework for evaluating the generation of paired RNA expression and copy-number variation (CNV) profiles, and apply it to the longitudinal DECIDER high-grade serous ovarian cancer cohort. The framework assesses modality-specific fidelity and coherence between RNA and CNV pairs using agreement between real training and held-out data as a reference, alongside complementary training-data proximity diagnostics. We adapt three multi-omic integration architectures based on variational autoencoders to generate paired profiles. The model leading several fidelity and coherence measures also shows the strongest training-data proximity. The protocol and case study offer the community a reference for comparing future multi-omic generators.
Shanghao Shi, Jian Xiao, Ching-Hsiang Chan, Sanjoy Baruah, Sastry Kompella, Yevgeniy Vorobeychik, William Yeoh, Ning Zhang
Latent communication enables agents to collaborate efficiently by directly exchanging intermediate model representations rather than raw sensor data or text messages. Existing approaches, however, are primarily designed for LLM agents and do not consider the bandwidth constraints and heterogeneous task deadlines faced by cyber-physical agents. In this work, we propose SCALE, a real-time latent communication method tailored for cyber-physical agents. Motivated by the spatio-temporal continuity of the physical world, SCALE matches tokens across consecutive latent messages using similarity scores to identify reusable information and avoid redundant transmission. It then profiles the semantic gain and communication cost of transmitting different amounts of latent information. Based on these profiles, SCALE employs a real-time scheduler that jointly determines which pending messages to transmit and how many latent tokens to send, subject to available bandwidth and task deadlines. We evaluate SCALE across three cyber-physical benchmarks and a physical spider robot case study under diverse network conditions and heterogeneous task deadlines. Our results show that SCALE significantly improves both task performance and on-time delivery in bandwidth-constrained networks. Our code is available at: https://anonymous.4open.science/r/SCALE-4E80/.
Taha Asghar, Bilal Mufti, Syed Qasim Zaheer, Luqman Ahmed
Landing-phase safety-event prediction requires the detection of rare abnormal approaches within strongly imbalanced multivariate flight data. Capturing coupled temporal behavior while distinguishing event-specific signatures across airport operating contexts remains a central modeling challenge. This study proposes an interpretable hybrid feature-fusion framework for multiclass landing-phase safety-event prediction. A long short-term memory (LSTM) autoencoder trained on nominal flight windows characterizes deviations from routine approach behavior. Its continuous reconstruction-error descriptor is fused with parameter-specific statistical flight-history features and supplied to a class-weighted XGBoost classifier to distinguish Normal, Speed High, Path High, and Flaps Late windows. The framework is evaluated using 53,750 NASA DASHlink approach windows from three arrival airports. Pooled and airport-specific learning strategies are compared alongside component-level benchmarks, repeated pipeline retraining, and feature-attribution analysis. The results demonstrate strong class-balanced detection, with mean macro recall above 0.90 across all airports under repeated partitioning and retraining. Component comparisons establish the importance of class weighting and show that continuous feature fusion outperforms reconstruction-error-only and hard-gate alternatives. Airport-level comparisons reveal benefits from local specialization at the two airports with larger datasets and from pooled training at the airport with the smallest dataset. Feature attributions associate predictions with operationally relevant vertical-path, configuration, speed, and approach-context variables. The framework provides an interpretable approach to rare-event detection and practical guidance for adapting flight-data monitoring models to airport-level data availability.
Pengkai Wang, Wei-Wei Zhang, Yan Li, Min Tang, Zhitian Hou, Zeyu Liu, Guanghao Zhu, Yuanyi Wang 等 16 位作者
Current medical benchmarks typically break down clinical competence into isolated factual questions, single-image tasks, or static case vignettes. These tests score individual answers, but reveal less about whether a model can integrate subspecialty knowledge with multimodal evidence and track a patient's condition throughout a course of care. To close this gap, we introduce NPCBench, a clinical apprenticeship benchmark for nasopharyngeal carcinoma (NPC). NPCBench mirrors specialist training, from guideline and foundational knowledge to case planning, longitudinal review, CarePath integration, and open-ended consultation. It contains 4,777 component and consultation items and a 100-patient CarePath cohort with 417 scored diagnostic and care tasks. The episode-level evaluation tests whether diagnostic and care answers are jointly correct within a care pathway. By linking stepwise tasks to patient-level care trajectories, NPCBench evaluates whether models ground decisions in visual evidence, interpret changes in patient state, and meet clinical requirements in consultation. Across 30 evaluated models, we observe a composition gap: strong performance on individual tasks does not ensure jointly correct diagnostic and care answers for the same patient. Among 21 multimodal models, strict CarePath success reaches at most 23%. Paired controls and physician calibration further examine evidence grounding, longitudinal revision, and consultation judging. Overall, NPCBench broadens the evaluation of medical artificial intelligence beyond static question-answering paradigms to an apprenticeship-oriented assessment of clinical workflows, thereby offering a testbed for evidence-based and temporally coherent reasoning in oncology care.
Robotic perception increasingly relies on heterogeneous sensor suites combining cameras with emerging modalities (event, SPAD, iToF/dToF, and thermal). Yet existing simulators idealize these sensors, hiding critical modality-specific failure modes that limit real-world deployment. We present VisionSIM, a unified, physically grounded simulation platform that elevates sensor realism to a first-class concern by modeling the raw physical measurement process for diverse conventional and emerging modalities. We expose fine-grained control over physical sensor parameters, reproducing failure modes overlooked by general-purpose simulators: pileup and glare for LiDAR; phase wrapping, saturation, and motion artifacts for iToF; refractory drops for event cameras; and history-dependent emission for thermal cameras. This enables principled cross-modality comparison on identical scenes and trajectories. We release the platform and a large indoor dataset with synchronized raw measurements and rich ground-truth annotations.
Online 3D bin packing with a processing buffer requires choosing which box to pack, its orientation, and its placement while keeping the stack stable under unknown future arrivals. A collapse ends the episode and forfeits the remaining volume. Existing methods use geometric rules or hard stability masks, limiting how risk can influence the joint choice of box, orientation, and placement. We propose Stability-Hybrid On-Policy Reinforcement Learning (SHORE), which combines a learned stability probability of a placement with an on-policy estimate of its return conditional on stability. A 3D U-Net learns stability from executed outcomes and simulated alternatives, revealing outcomes of actions underrepresented in early rollouts. A discrete on-policy actor--critic uses stability predictions in both its action values and policy prior, with gates that limit the effect of an untrained or changing model. In deployment, SHORE does not require additional search or simulation overhead. In Isaac Gym, SHORE reaches 73\% space utilization by 0.4M environment steps, while the same learner without stability modeling reaches that level at about 1M. On the expanded 60-box set, SHORE achieves 0.734 space utilization versus 0.588 for the GOPT state-of-the-art baseline, with a higher placement-failure frequency (1.03\% vs. 0.63\%), trading more risk for greater packed volume. The results show faster learning and higher utilization on the more demanding realistic box arrival set. Ablations suggest that simulated alternatives provide useful early supervision, while executed outcomes alone sustain later performance. We also demonstrate the method on a real robot to verify the simplicity of integrating the method to real bin packing pipelines.
Oral cancer and the oral potentially malignant disorders (OPMD) that can precede it are visible on the oral mucosa, which makes smartphone photographs a promising tool for screening. Deep learning models for this task classify individual photographs. However, screening refers patients, not photographs: a patient is usually photographed several times, and the referral decision must miss no more than a set share of patients with disease. We propose OraRefer, a patient-level screening model that reads each photograph with a segmentation-guided foundation model that conditions its class decision on the predicted lesion, combines all photographs of a patient with risk factors, and refers through a threshold with a finite-sample sensitivity guarantee. The model localizes the mouth itself, so inference needs no annotation. On 3000 photographs of 714 patients, patient-level fusion reaches an AUROC of 0.919 and outperforms attention-based multiple-instance learning. At a 90% sensitivity target, OraRefer detects 95% of patients with OPMD or cancer and misses 3 of 61 patients with cancer, at a specificity of 0.610. Because the threshold is calibrated without retraining, each clinic can set its own guaranteed operating point on a local calibration sample.
Robert Ren, Matthew Bronars, He Zhu, Yikai Tang, Jianing Zhang, Sergey Zakharov, Pavel Tokmakov, Katerina Fragkiadaki
We introduce GPWM, a Generative Particle World Model that predicts how 3D scenes evolve by generating the motion of particles and rigid bodies over time. Unlike video world models that generate scene appearance, GPWM operates directly in 3D physical space. A single representation and architecture cover rigid, deformable, cloth, and granular materials, in both passive physics and robotic interaction. GPWM is an object-aware multi-resolution transformer that combines flow matching with diffusion forcing to generate coherent physical trajectories far beyond its training horizon. We train and evaluate GPWM on ParticleWorlds, a 25.5-million-frame corpus spanning diverse geometries, materials, interactions, and robot embodiments. GPWM outperforms the best prior learned-dynamics baseline in each of six physical domains, reducing prediction error by 31% on average. Breaking from the domain-specific setting of prior methods, a single Unified GPWM models these heterogeneous regimes with one set of weights, largely outperforming per-domain baselines and even surpassing our own specialist models in two domains. GPWM further generalizes to unseen geometries, physical properties, object counts, and rollout horizons. Beyond prediction, GPWM enables contact-rich planning by learning a generative prior over joint actuator-object motion. Instead of searching over unconstrained actions, planning searches the model's latent noise for interactions that achieve a desired object-space goal. With the same world model, noise-space planning substantially outperforms action-space search on pushing and picking, where complex contact sequences are hard to discover. Executed open-loop in simulation, the resulting plans achieve goal-directed manipulation without reference action trajectories or goal-specific policies.
Much of human culture is tacit and is typically learned through situated social interaction. As LLMs are deployed as interactive agents across diverse communities, assessing their cultural competence demands moving beyond existing question-answering approaches, shifting from what they know to how they adapt. In this work, we broaden the scope of cultural adaptation in NLP by investigating whether LLMs can adapt through repeated interactions via a sandbox environment. By reframing cultural adaptation as a continuous process of norm assimilation through social interaction, our work demonstrates that memory mechanisms foster longitudinal learning across repeated interactions, facilitate transfer to different social settings, and enable multilingual norm assimilation across five languages. Together, our findings lay the groundwork for continuous, interaction-driven cultural adaptation in LLMs.
Fine-tuning large language models on narrow data with harmful content produces broadly misaligned behavior on unrelated prompts, a phenomenon known as emergent misalignment. We propose that emergent misalignment involves persona-model collapse: a deterioration of the internal representations that encode and distinguish personas, undermining the nuance and consistency of their expression. We first test this hypothesis behaviorally, by comparing how models fine-tuned on harmful data and benign controls answer questionnaires while role-playing diverse personas. Across four model families, harmful fine-tuning produces more polarized, all-or-nothing answers, suggesting more binary, less nuanced persona expression. Histograms show increased concentration at the scale endpoints, and Kullback--Leibler (KL) divergence quantifies the distributional shift. We also find reduced within-persona consistency: repeated answers to the same question under the same persona become less stable. We reproduce these observations across diverse harmful datasets and questionnaires that probe different aspects of persona-conditioned behavior, while capability and simple instruction-following controls argue against a uniform loss of general ability. Finally, a mechanistic analysis of persona-conditioned activations in Qwen3-235B shows that representations elicited by different persona prompts span fewer effective dimensions in the misaligned variant than in the original and benign-control variants, most clearly in the final layers. Together, these findings provide behavioral and representational support for persona-model collapse as an aspect of emergent misalignment.
Heyuan Huang, Jirui Dai, Alexandra DeLucia, Sonal Joshi, Mahsa Yarmohammadi, Jie Gao, Bernal Jiménez Gutiérrez, Mark Dredze
Retrieval-based factuality evaluation, where LLM-generated claims are verified against evidence from authoritative medical corpora, has become the dominant paradigm for scalable hallucination detection in high-stakes clinical settings. Despite the urgency of reliable and transparent medical fact verification, most systems measure performance with aggregate metrics like F1, which obscure where and why failures occur. Existing RAG diagnostics require gold answers or annotated gold evidence, neither of which exists in this regime. We introduce two comprehensive taxonomies, grounded in a case study on the open-ended MedExpert dataset and 3 closed-ended datasets, decomposing failures into retrieval-stage errors along five quality dimensions, and verifier-reasoning errors into six consecutive steps. We adapt an automatic pattern induction pipeline using LLM-as-Judge to label evidence quality and classify verifier reasoning errors at scale, and then stress-test our findings across 4 retrieval methods and 6 frontier verifier models. Our analysis reveals that scaling model size, adding reasoning effort, expanding to authoritative web sources, and applying medical fine-tuning do not resolve these failure modes, demonstrating that they represent fundamental limitations of the retrieve-then-verify paradigm in open-ended medical settings rather than artifacts of outdated systems. We release our code and data at https://anonymous.4open.science/r/Medical_RAG_eval-4AB5 for the full reproducibility of our results.
Niclas Lietzow, Danielle Bitterman, Carsten Eickhoff, William Rudman, Michal Golovanevsky
Vision-language models must reconcile visual evidence with memorized world knowledge when the two conflict. How they resolve this conflict shapes the reliability of multimodal systems, yet prior work characterizes it behaviorally without a component-level causal account. We combine activation patching across three granularities (residual stream, attention heads, and MLP sublayers) with model-component ablation studies and mechanistic analysis. Across three VLM families, we find that visual grounding emerges by default, whereas prior grounding depends on a small set of causally necessary attention heads (2.5-4.8%) concentrated in the second half of the network. These heads enable answers from stored world knowledge (e.g., "red" for a strawberry) despite conflicting visual input. Ablating them flips predictions from knowledge-grounded to visually grounded answers in 68-96% of cases under prior-knowledge prompts, but changes only 0.8-7.5% of visually grounded predictions, establishing an asymmetric causal structure. The identified heads decompose into routing heads, which modulate information flow, and writing heads, which directly project answer tokens into the residual stream. This structure is consistent across model families and scales, revealing a sparse causal circuit underlying perception-knowledge conflict in VLMs.
Deterministic benchmark scores show that an agent received credit, but not whether that credit was earned, reported honestly, or would hold on a second run. We introduce a Trust Layer for Agent Evaluation, an additive post-hoc framework that reports, beside each recorded score, whether it should be believed. It verifies four properties: whether the result is supported by the benchmark's own grading logic, whether a passing answer was earned through traceable computation, whether the agent's completion claim matches what occurred, and whether the result is stable under repeated execution. The first three use only saved artifacts; the fourth re-runs the agent. Model judgments only label evidence under majority voting; all verdicts follow deterministic rules and never modify the recorded score. Applied to five agent configurations on 108 tasks from Agents' Last Exam, every model shows passing runs with no traceable computation (at rates varying tenfold), confirmed false completion claims, and unstable results: 18-46% of tasks do not stay in one score band over five runs. Only 22.6% of recorded passes clear all four checks (95% CI 15.0-32.6, n=84). Measuring what an agent can do and verifying that it did it are different problems, and current benchmarks address only the first.
Models of collective motion usually reduce social interactions to pairwise forces, despite empirical evidence for many-body effects. We introduce the directional atomic cluster expansion (dACE), which adapts the atomic cluster expansion (ACE) of interatomic potentials to self-propelled agents. dACE provides a systematically improvable, exactly equivariant expansion of an agent's acceleration that splits uniquely into 1-, 2-, 3-body and higher contributions, along with a self term for the agent's own past motion that is fitted so that it does not absorb the interactions. On simulated collectives in two and three dimensions with known rules, dACE recovers the interaction law body order by body order from little data, while the neural model SocialNet, the current closest competitor, needs more data, has much larger errors on many-body or chiral rules and is restricted to 2D. On real data sets of tracked zebrafish and golden shiners, the laws learned by dACE agree better between disjoint halves of the data than SocialNet's. The learned laws are interpretable, and in the shiners the 3-body term opposes the pairwise response as neighbourhoods get dense. Finally, we show that next-step prediction accuracy mostly measures self-propulsion and noise rather than the interactions, and suggest more robust combinations of error measures for interaction learning.
Automatic sleep stage classification from electroencephalography (EEG) signals is a critical task in clinical sleep medicine. While deep learning models have shown promising results in this domain, their performance is fundamentally constrained by the quality of training labels, which human experts manually annotate and that are subject to inter-rater disagreement and systematic labeling errors. In this paper, we propose a novel LLM-assisted label-noise filtering pipeline that operates exclusively during training to identify and remove unreliable sleep-stage annotations before retraining the classifier. Specifically, TinySleepNet is first trained on the original dataset to extract per-epoch SoftMax confidence scores, then low-confidence epochs are submitted to LLaMA 3.1 along with their EEG spectral features and surrounding sequence context. The LLM evaluates each epoch against AASM sleep-scoring guidelines and assigns a verdict of RELIABLE or SUSPICIOUS. Suspicious epochs are removed from the training set, and TinySleepNet is retrained on the cleaned data. Evaluated on the Sleep-EDF Expanded dataset using 20-fold cross-validation, our method improves overall accuracy from 81.8% to 84.6% and Macro F1-score from 75.6% to 77.4%, with particularly notable gains in REM detection. Crucially, because the LLM operates only during training, the lightweight TinySleepNet architecture (1.3M parameters) is preserved at inference, making the proposed pipeline suitable for real-time, wearable sleep monitoring applications.
Daniel Salwasser, Lars Gottesbüren, Laxman Dhulipala, Jakub Łącki, Guy Blelloch, Rajesh Jayaram, Tobias Rubel, Kshiteej Sheth 等 10 位作者
We present \textsc{Ginger}, a state-of-the-art open-source C++ library for graph-based approximate nearest neighbor search (ANNS). We developed \textsc{Ginger} at Google to push the speed and memory efficiency of graph search on large and high-dimensional embedding datasets. \textsc{Ginger} surpasses existing ANNS libraries with regards to query throughput and indexing time while matching the memory footprint of the most space-efficient competitors. \textsc{Ginger} builds upon recent advances in proximity graph construction (PiPNN) and vector quantization (TurboQuant). We introduce Nested TurboQuant (NTQ), a two-level quantizer which supports low precision graph navigation and high precision reranking in a single code. This representation achieves near floating point precision with $2.6\times$ less memory. We also compress the size of our proximity graphs by up to $2.5\times$ without impacting query throughput. We combine graph reordering with a new lossless compression codec called AdaptiveGap, which builds upon StreamVByte. At $95\%$ recall, \textsc{Ginger} speeds up query throughput by an average of $1.8\times$ over each dataset's fastest competitor and by at least $2.3\times$ over any single library. This lead grows to $2.9\times$ over compact libraries on high-dimensional embeddings.
Shuyue Stella Li, Priyanka Kargupta, Haocheng Wang, Jinu Lee, Dean Light, Shan Chen, Orevaoghene Ahia, Thomas L. Griffiths 等 12 位作者
Large language models (LLMs) solve complex problems yet fail on simpler variants, suggesting they achieve correct outputs through mechanisms fundamentally different from human reasoning. To understand this gap, we synthesize cognitive science research into a taxonomy of 28 cognitive elements spanning reasoning invariants, meta-cognitive controls, representations for organizing reasoning and knowledge, and transformation operations. We introduce a fine-grained evaluation framework and apply it to 192K traces from 18 models across text, vision, and audio, alongside 1,346 human think-aloud traces on a new, domain-agnostic dataset for human-model comparison. We find that models are able to deploy cognitive elements seen in human reasoning but differ in how they use them. Leveraging these patterns, we develop test-time reasoning guidance that automatically scaffolds successful reasoning topologies, improving the accuracy of the strongest models by 9.7 percentage points on ill-structured problems. Meta-analysis of 1.6K LLM reasoning papers reveals the research community concentrates on easily quantifiable elements but rarely targets self-awareness, whose use differs between human and model reasoning. By establishing a shared vocabulary between cognitive science and LLM research, our framework enables diagnosis of reasoning failures and principled development of models that reason through robust cognitive mechanisms rather than spurious shortcuts, while providing tools to test theories of human cognition at scale.
Sheldon Liang, Ali Taghibakhshi, Saurav Muralidharan, Sharath Turuvekere Sreenivas, Ashwath Aithal, Pavlo Molchanov, Nima Tajbakhsh
LLM pruning must preserve quality while meeting hardware- and workload-specific constraints on latency, throughput, and memory. Existing pruning workflows often target parameter counts, requiring practitioners to manually select model sizes that meet latency, throughput, and GPU memory limits. However, parameter count alone does not predict these metrics, which depend on the model architecture, workload, hardware, and serving engine. We introduce AutoCompress, a hardware-guided compression framework that directly incorporates such constraints into differentiable architecture search. AutoCompress jointly optimizes global and per-layer structure to produce heterogeneous architectures, guided by serving-engine cost models validated against end-to-end latency, throughput, and memory measurements. We evaluate AutoCompress on long-context local coding, batched datacenter serving, and latency-sensitive in-game interaction. Compared to the state-of-the-art Minitron pruning framework, AutoCompress improves average accuracy across eight benchmarks by 6.57, 2.35, and 0.57 percentage points on RTX 5070, RTX 5080, and H100, respectively. AutoCompress also compresses Nemotron 3.5 Super from 120B to 74B parameters for single-B200 BF16 serving at 128K context, retaining 95% of its mean benchmark score.