让下一篇重要论文，不再错过。

User embeddings are central components of modern recommendation, personalization, and behavioral modeling systems. Though these embeddings can compactly encode rich behavioral information, they are inherently dense and opaque, limiting interpretability. To address this gap, we argue that user embeddings can be understood as sparse superpositions of interpretable behavioral modes. This structure is directly predicted by generative models of user behavior for linear embeddings and extends approximately to deep-learning representations by the superposition hypothesis. Superimposed modes can naturally be recovered by sparse dictionary learning methods, such as sparse autoencoders (SAEs), and recovered modes can be automatically named and queried in natural language. We validate mode recovery against known ground truth on synthetic data, where SAEs recover modes better than classical sparse baselines, including on transformer-encoded embeddings. We then apply the method to classical and transformer-based user embeddings from two real-world datasets. We also show how the corresponding natural-language labels enable zero-shot audience retrieval and characterization over embeddings never trained on text. The framework is post-hoc and model-agnostic, requiring no modification to the upstream embedding system.

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.

Grounded reasoning, where a model outputs text coordinates of relevant locations in chain-of-thought reasoning, is a common strategy for solving complex visual reasoning problems. However, the benefits and limitations of grounded reasoning are understudied. We study where and why grounded reasoning is most useful. To do so, we introduce a chain-of-thought analysis method to measure how much a model uses its outputted grounding in its thinking trace and perform a detailed ablation study of a grounded reasoning training recipe. We find that grounded reasoning improves counting but shows no significant gain over baselines without coordinate grounding across a wide range of spatial reasoning and perception benchmarks. Using several synthetic tasks, we show that grounded reasoning is most useful when a task requires undescribable references, which are visual concepts that are difficult to describe in language. However, we observe that many existing visual reasoning benchmarks are solvable with only describable text references. To close this gap, we introduce BrickVision, a new benchmark that requires undescribable visual references to solve.

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.

Modern video models possess strong visual priors, yet do not natively represent human action in metric 3D space. We present EgoGeneralist, which adapts pretrained video generators to unify action-conditioned egocentric video generation, pose estimation, and joint action-video generation. Our role-adaptive action renderer provides video-native representations for multidirectional action-video learning, making metric motion usable as both a spatial control signal and a recoverable generation target. Cross-stream action-video coupling establishes close correspondence between metric motion and visual dynamics through lightweight adaptation while preserving the pretrained visual prior. Together, these designs establish a generalist framework spanning five action-video tasks. Despite adaptation only on controlled laboratory recordings, EgoGeneralist generalizes to diverse unseen scenes and interactions with effective action control and visual quality, bridging pretrained visual knowledge and metric 3D human motion.

Existing home energy management systems conceptualize occupants as passive recipients of energy information and control, which limits their ability to effectively support informed decision-making and sustained engagement. This paper presents Home Energy Management Assistant (HEMA), the first open-source, multi-agent system enabling sustained human-AI collaboration - multi-turn conversational interactions with preserved context - across diverse home energy management (HEM) tasks - from energy analysis and educational support to smart device control. HEMA combines large language model (LLM) reasoning capabilities with 36 purpose-built domain-specific tools through a three-layer architecture: a web-based conversational interface, a backend API server, and a multi-agent system. The system features three specialized agents - Analysis (energy consumption patterns and cost optimization), Knowledge (educational queries and rebate information), and Control (smart device management and scheduling) - coordinated through a self-consistency classifier that routes user queries using chain-of-thought reasoning. This architecture enables various energy analyses, adaptive explanations, and streamlined device control. HEMA also includes a comprehensive evaluation framework using an LLM-as-simulated-user methodology with 23 objective metrics across task performance, factual accuracy, interaction quality, and system efficiency, allowing systematic testing across diverse scenarios and user personas without requiring extensive human subject testing. Through demonstrations using real-world household energy consumption data, we show how HEMA supports informed decision-making and active engagement in HEM, highlighting its potential as a user-friendly, adaptable tool for residential deployment and as a research platform for HEM innovation.

Temporary traffic control redirects vehicles through work zones, requiring automated vehicles to connect an interpretation of the visible devices to a continuous passing path. Geometric predictors provide several plausible paths, yet selecting the one consistent with the temporary control remains difficult. We propose SAGE (Semantic-to-Action Grounded Execution), a read-decide-execute-check framework that separates path construction, path preference, and control compliance within a shared candidate space. The central idea is to use a vision-language model's (VLM's) trajectory as a preference over intact geometric candidates. A role-decoupled generator constructs the candidates, and rank fusion combines each candidate's geometric score with its agreement with the VLM trajectory. Supervision structured around the Manual on Uniform Traffic Control Devices (MUTCD) trains SAGE-VLM to state the lane status and hazard side before its passage decision and trajectory. The predicted hazard side enters image-space predicates derived from MUTCD provisions, which support a final intervention bounded in score and displacement. On 2,306 real-world ROADWork validation episodes, the candidates include a path within 10 px of the human demonstration in 87% of episodes, yet selection regret accounts for 53% of the geometric ranker's error. SAGE reduces average and final displacement errors by 8.3% and 12.0% relative to the ranker and hard-constraint violations by 37%. The trajectory opinion improves all four tested generator-selector combinations. SAGE is comparable to state-of-the-art VLM modules in accuracy, with a faster-responding VLM, and outperforms four other adapted driving VLMs in displacement error and safety measures. Project page: https://hhjj233.github.io/SAGE-page/.

ML code should run where researchers choose, not where it was first written. Today, machine learning code mainly lives in two ecosystems, PyTorch on NVIDIA GPUs and JAX on Google TPUs. Migrating repositories between them means building and testing across frameworks and hardware. This migration is (1) a practical need, since it unlocks different hardware for researchers, and (2) a natural benchmark: it involves different frameworks and hardware, covers whole repositories, and is long-horizon and challenging. We introduce XStack, a benchmark that evaluates agents on migrating ML repositories between these two ecosystems. Each agent receives a PyTorch repository and a clean TPU machine and must deliver a JAX repository. Its output passes if it meets our grading criteria, based on the source repository's published result, authenticated and calibrated by us. Migration by agents remains unreliable: Claude Opus 5 passes 75%, GPT-5.6-sol 45%, and every other agent at most 40%. XStack is a step toward freeing ML code from its ecosystem. We release XStack to make progress toward reliable migration across frameworks and hardware measurable.

Ground-to-space astronomical super-resolution requires recovering space-quality images from ground-based observations that are simultaneously limited by pixel sampling resolution and atmospheric seeing, which imposes a stochastic, spatially varying PSF that cannot be resolved through upsampling alone. Existing methods rely on synthetic training pairs that fail to capture real atmospheric statistics and are prone to either over-smoothed reconstructions or hallucination sources with no physical counterpart in the observed sky. We propose FluxFlow, a conservative pixel-space flow-matching framework that incorporates observation uncertainty and source-region importance weights during training, and a training-free Wiener-regularized test-time correction to suppress hallucination sources while preserving recovered detail. We further construct the DESI--HST Dataset, the large-scale real-world benchmark comprising 19,500 real co-registered ground-to-space image pairs with real atmospheric PSF variation. Experiments demonstrate that FluxFlow consistently outperforms existing baseline methods in both photometric and scientific accuracy. Project webpage: https://i2wm.github.io/fluxflow/.

The canonical cut hierarchy recursively partitions a graph along Tutte-Nash-Williams partitions, choosing the partition with the most number of parts at each step. We give a randomized almost linear-time algorithm for computing this hierarchy in undirected graphs with polynomially bounded capacities via reduction to max-flow. This yields randomized almost linear-time algorithms for computing partition connectivity, strength, both arboricity and fractional arboricity, both integral and fractional spanning tree packing numbers, and implicit representations of the principal partition sequence and a maximum fractional spanning tree packing.

To interpret a direction in a language model's residual stream, we can look for text that strongly activates it. But when such text is found by scanning a corpus, it is limited to what that corpus contains. We develop Maximally-Activating Example Metamodels (MAEMs), language models which map residual stream directions from a subject model to text exemplars that strongly activate those directions. Our objective is to maximise the cosine similarity of a supplied residual stream direction with the activations of the text generated by the MAEM. This label-free RL objective helps to scale MAEMs on diverse sources of residual stream directions. Empirically, we find the resultant metamodel produces informative exemplars: a Qwen3.6-27B MAEM inverts held-out activations better than corpus search or a natural-language autoencoder (NLA), beats corpus search (at 10M tokens) on a majority of held-out languages/domains, and activates 89% of held-out SAE features. It also identifies concepts behind steering vectors about as well as NLAs, recovers the payloads of LoRA backdoors more often than corpus search, and can even verbalise some model workspace inner monologue. Unlike explanations produced by other metamodels (e.g. NLAs), MAEM exemplars can be verified with a forward pass on the subject model, making the MAEM objective a grounded alternative for scaling the training of metamodels.

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.

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/.

Creating images from noise is image generation; reconstructing fine details from coarse inputs is super-resolution. Despite their practical differences, both can be understood as reversing information loss across scales. We introduce $\textbf{SKILD}$, a $\textbf{S}$cale-invariant $\textbf{K}$-Space $\textbf{I}$mage $\textbf{L}$earning $\textbf{D}$iffusion model that unifies generation and continuous super-resolution within a single unconditional framework. Both natural images and critical physical systems exhibit scale invariance, and we leverage it to design a forward process that attenuates image content from fine to coarse scales while injecting spectrum-matched Gaussian noise, making scale an explicit coordinate of the diffusion dynamics. The same trained reverse process performs generation and continuous super-resolution by varying only the starting timestep: $\textit{no task-specific architecture, no conditioning branch, no classifier-free guidance, no retraining per scale factor}$. Empirically, SKILD reaches FID $2.61$ and Inception Score $9.65$ on unconditional CIFAR-10, performs $2\times$--$8\times$ super-resolution on ImageNet from a single unconditional checkpoint while outperforming conditional models across perceptual metrics, and reconstructs critical Ising models whose connected four-point correlations closely track the ground truth.

Generative AI and agentic coding are accelerating software development, leading to concerns about technical debt. But two less visible forms of debt, cognitive debt and intent debt, are emerging as key risks to software health. Cognitive debt is the absence or erosion of a team's shared understanding where no one can confidently explain how a system works or predict the impact of a change. Intent debt is the lack of clearly articulated goals, constraints, and rationale that explain what the system is for and guide how it should evolve. This paper presents Cognitive Tours, a tool designed to mitigate and repair cognitive and intent debt. Cognitive Tours is a web-based prototype for creating and following curated paths connecting user needs, design decisions, code, tests, and open questions. AI-assisted skills help audit existing waypoints and tours for possible problems or staleness and suggest candidates for new ones; humans decide what matters, verify relationships, and construct the narrative. This division of work deliberately preserves productive friction: the sensemaking involved in selecting, explaining, and revising a tour, which is intended to foster and share understanding. Our demonstration uses a tour of Cognitive Tours itself to introduce the tool and illustrate its design decisions and interaction mechanisms. A proposed evaluation agenda concludes the paper. A demo is available at https://cognitive-tours-demo.vercel.app/

On-policy distillation (OPD) trains a student on its own rollouts under a frozen teacher's per-token supervision. Does this dense signal reduce the training FLOPs needed to reach a target accuracy compared with reinforcement learning? We systematically compare vanilla reverse-KL OPD with Group Relative Policy Optimization (GRPO) across task difficulty, student and teacher size, domain, and optimization choices, measuring held-out accuracy against training FLOPs. With an external post-trained teacher, OPD generally loses to GRPO in accuracy per training FLOP at matched settings, even when the teacher is further trained with GRPO, and often suffers response-length inflation. Yet a different teacher changes the picture: when the teacher is the student's own GRPO checkpoint, OPD recovers that checkpoint's accuracy on math with $50\times$ fewer distillation FLOPs than GRPO needs to reach the same accuracy. Following the same teacher setting, we consistently observe this high efficiency with a larger student, on code, and in a second model family. On a single domain, however, the GRPO checkpoint is already the desired model, so distilling it adds to total compute. Across domains, separate GRPO checkpoints hold different capabilities, and no single checkpoint serves every domain. Multi-teacher OPD (DeepSeek-AI, 2026) consolidates per-domain GRPO experts into a single student that matches each expert's accuracy, with the distillation stage costing only 5.2% (on 2 domains) and 8.0% (on 4 domains) of the specialist training FLOPs. Counting specialist training and distillation together, this pipeline reaches joint GRPO's performance on the 2-domain mix with 42.0% fewer total FLOPs.

Reproducing machine learning experiments requires reconstructing procedures from information scattered across papers, released code, and supporting artifacts. For coding agents, reproduction success can therefore depend on their ability to fill in missing details, rather than on the released artifacts alone. We ask whether preserving recovered procedures and decisions improves subsequent independent reproduction. We define a standardized artifact design comprising a refined codebase, environment setup, experiment launchers, and a reproduction guide, and introduce ReproBridge, an agent workflow for constructing these refined reproducible artifacts. ReproBridge-Bench provides 1,029 tasks from 387 experiments across 36 papers, anchored by five human-verified reference scores per task. Agents attempt tasks in fresh sessions with raw code or refined artifacts; a separate Auditor checks execution evidence. Across four evaluated models, refined artifacts raise mean reproduction success from 76.4% to 92.6% and reduce downstream API spending by 19.7% on average. The cross-model success-rate spread narrows from 21.6 to 3.9 percentage points, with all four models achieving mean success rates above 90%. Removing any component reduces success in the ablation study, with the largest losses from code and environment removal. These results suggest that preserving execution knowledge helps even models with lower raw-code success reproduce experiments more reliably and at lower API cost.

State-of-the-art feedforward reconstruction models perform poorly on in-the-wild object-centric photo collections. They are trained largely on single-session object captures whose images vary in viewpoint while the object and its environment remain nearly fixed. Public photo archives contain a different form of multi-view data. They hold found photos of the same object or design: photos taken independently by many people across years, environments, cameras, and sometimes physical units. In this work, we introduce Trainspotting, a dataset of 12,068 objects and 311,886 photos from Wikimedia Commons. We reconstruct each object to recover camera poses, multi-view stereo depth, and object masks. We categorize the objects into three regimes: objects fixed at one site, objects that move between sites, and designs shared by many physical units. Evaluating recent feedforward reconstruction models on Trainspotting shows that they fail most severely in the last two regimes. Fine-tuning VGGT-$Ω$ on Trainspotting significantly improves camera pose, depth, and 3D geometry on held-out objects. We observe similar improvements on the independent NAVI in-the-wild benchmark, which contains photos of each object in different environments. Performance on conventional benchmarks remains largely unchanged.

Lee-Yang theorems are a powerful tool for studying many-body systems, with applications ranging from analyzing phase transitions to proving the efficiency of certain classical and quantum algorithms. In this work, we prove a Lee-Yang zero-freeness theorem for the partition function of a broad class of interacting fermion models, implying the existence of a provably efficient quantum algorithm for estimating their ground-state energies. This class includes several well-known models such as the attractive Hubbard model, repulsive Hubbard model on bipartite graphs, and the interacting Hofstadter model. Our results also rigorously establish the nonexistence of phase transitions in these models in the presence of a nonzero local external field.

Customer experience (CX) agents use tools and large language models to address customer requests and guide conversational interactions with an organization's products. Improving these agents, especially in regulated industries, is difficult: they must detect intent, follow complex operational policies and use tools reliably. Manual end-to-end testing offers limited coverage, while live experiments expose customers to failures that can erode trust. We present a hypothesis-driven simulation workflow for screening candidate CX agents before deployment. Synthetic customers react to agent responses and simulated tool outputs enable multi-step agentic workflows without invoking production backends. We use the Snowglobe simulator on Nubank's Card Delivery agent and its expanded successor, Card Management - Nubank's highest-volume chat-support agent in Brazil. Across 4 deployed versions, simulated and production version-level binary evaluator scores show high correlation. Simulation-guided iteration increased transactional net promoter score (tNPS) by 36.69 points in a live A/B test. We also screened open-weight configurations in over 16,000 simulated conversations. In a subsequent live A/B test, the selected model increased self-service rate (SSR) by 8.82 percentage points to the highest level observed at Nubank, with no statistically significant change in tNPS. Simulation made broad exploration of models, reasoning settings, and prompts feasible without customer exposure, enabling production improvements that would have been impractical to pursue through live experimentation alone.

A defender anticipating an attacker needs a probability distribution over the attacker's next action from information available before that action occurs. Two structures complicate this: the attacker's behavioral regime persists and can switch within an episode, and each action combines an operation with state-dependent arguments whose feasible values change over time. We capture both with a hidden Markov model whose latent state is the active regime and whose regime-conditioned action law is defined on a feasibility-masked tree of executable actions. The same likelihood forms the next-action distribution and updates the regime belief once the action is seen, and a partially observed prefix contributes its resolved local factors. Estimation separates into convex logistic fits that Baum-Welch reuses when regimes are unlabeled. We derive a reach-weighted nodewise KL expression for regime information, a sequence-wise log-loss bound for switching regime paths, and a uniform fallback guarantee. On the CAGE 2 cyber defense benchmark the filtered latent state is worth 0.8 nats per step, and after a controller switch the loss begins to fall in three steps against ten for a recurrent baseline.

The speech capabilities of mobile devices have grown, but their displays still rely on a text-based chat. When resuming a voice chat after a context switch, the last messages may not convey the key points of the conversation. Autobiographical use and sessions with external participants led to design goals for a mobile-first, speech-driven, glanceable display that helps users continue their reasoning from where they left off. We implement these goals in Souffleur and evaluate it in a between-subjects study with 48 participants. Compared with a conventional chat transcript, we found no reliable difference in how easily participants followed and resumed short voice conversations, nor in effort, attention or recall, although Souffleur reduces the conversation to a few short phrases.

Standardised post-quantum cryptography (PQC) is critical for protecting software systems against future quantum threats, but standardised algorithms alone do not provide a pathway for their secure integration into software. This paper presents a Systematisation of Knowledge (SoK) of PQC integration in software development, synthesising existing approaches across implementation, migration, and developer education. We analyse these approaches against NIST PQC standards and PQC migration guidance, software development lifecycle coverage, cybersecurity practices, deployment maturity, and testing and evaluation evidence. Our analysis shows that existing work establishes important technical and organisational foundations but provides disconnected support across the software development lifecycle, with limited coordination between preparation, implementation, migration, deployment, maintenance, and developer practices. Evaluation evidence also remains concentrated in controlled or early-stage settings, providing stronger evidence of technical feasibility than operational readiness. We identify cross-cutting gaps and research directions for developing lifecycle-oriented, empirically validated approaches that connect technical, developer, and organisational requirements for secure PQC integration.

Cyclist fatality rates remain high despite policy support for active travel, and autonomous vehicles add complexity to mixed traffic. Neither crash statistics nor multi-model detection pipelines scale to continuous monitoring of perceived safety along a route. We present an edge-assisted crowdsensing system: handlebar cameras supply street-level frames, a vision-language model (VLM) turns each frame into structured traffic attributes through a single prompt, and an explainable boosting machine (EBM) combines them with OpenStreetMap road infrastructure and on-bike kinematics to predict a crowdsourced 1-9 safety score. Extraction belongs at the network edge: the fastest self-hosted model needs 2.49 s per frame and scoring takes 0.4 ms on a CPU. On 1,362 annotated frames from two London routes, the EBM regressor reaches 72.2 percent binary accuracy and a mean absolute error of 1.011 under five-fold cross-validation. It is within one accuracy point of black-box ensembles. Its per-feature shape functions give exact attribution: physical lane separation is worth +0.97 rating points, a painted lane +0.70 and a construction site -0.65. Planners can therefore price an infrastructure change in rating points, and the results enter cost-benefit analysis directly. We also report inter-annotator agreement (alpha = 0.4695), an attribute audit, and a sensitivity analysis linking attribute error to accuracy. With re-encoding, a 20 s sampling interval serves 1,000 riders at 54 Mbit/s at a cost of 1.4 accuracy points, making continuous monitoring practical.

Typed-decision models score options that the caller supplies at request time, so their accuracy depends on label wording that the model's developer does not control. We compare two releases of a 307M-parameter multilingual cross-encoder, Ru-Jev, trained on thirteen Russian-language tasks; the later release adds an augmentation that presents labels in several surface forms. Holding questions fixed and varying only how eighteen routing labels are worded, we evaluate ten label vocabularies, produced by three separate processes and checked for overlap with the training data, with five candidates per question and, for eight of them, also with eighteen. On held-out vocabularies made of ordinary or borrowed words (English-derived jargon, office slang, department names), the augmented release is more accurate in every setting, by up to 0.24. Eleven of these fourteen gains survive a Holm correction (slang with eighteen candidates and descriptive sentences with five do not), and for jargon on all labels the label-level interval excludes zero in both settings. On abbreviations and opaque codes it gives no usable gain: on abbreviations it is not better, and worse on the clean slice, and on opaque codes both releases stay at or below chance once labels related to training options are removed. In most cells where accuracy rises, the risk-coverage area falls, but with eighteen candidates confident errors rise on jargon and slang. On two datasets neither release saw, accuracy moves by -8.4 and +5.6 points. We also report four evaluation hazards from comparing the two releases.

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.

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.

We introduce SM4RT, a multi-task cascade feed-forward model for dynamic scene reconstruction from RGB videos. While existing feed-forward approaches typically address geometry reconstruction and motion estimation separately--and often predict motions in a single coordinate frame that must account for camera movement--SM4RT solves these tasks sequentially by first reconstructing 3D geometry and then bootstrapping motion estimation from the predicted structure. By expressing each pixel's 3D motion in its corresponding camera frame, SM4RT reduces ambiguity arising from camera motion. To address the sparsity of real-world motion labels, SM4RT further incorporates a weakly supervised loss that regularizes unlabeled point motions using labeled point positions at the other time step. We evaluate motion estimation on three dynamic-scene datasets--Kubric, Waymo, and Point Odyssey--and SM4RT achieves the best performance on most metrics compared with prior baselines. We additionally benchmark geometry reconstruction on five dynamic scene datasets and show that SM4RT achieves competitive performance with state-of-the-art methods. Overall, our unified framework demonstrates that combining geometry reconstruction and motion estimation in a cascade not only avoids performance trade-offs but also enhances motion estimation through improved geometric reasoning.

LLM-based medical assistants that follow a patient across visits need a memory that keeps the patient's clinical state correct as it changes. We examine the errors of five memory systems on longitudinal medical dialogue and find three recurring causes: the retrieved evidence gives an outdated state, includes another person's health facts, or lacks a safety fact the answer depends on. We present MedGraph, a graph memory built for these three cases. Construction extracts dated facts from each session, attributes each fact to the person it describes, and links medication, diagnosis, and measurement records into chains that keep earlier states. Retrieval ranks the patient's chains as units, reads each chain at the time the question asks about, and adds the original dialogue text; a patient profile lists every recorded safety fact with every question. On MedMemoryBench, MedGraph has the highest overall accuracy with both a GPT and a Qwen backbone, 10.5 and 10.4 points above the strongest of five baselines, with efficient memory construction and answering in the GPT setting.

As society, ethics, and professional practice become increasingly integrated across the computer science curriculum, technical courses need practical ways to support ethical reasoning. Ethical essay assignments offer one approach, but little is known about how students use GenAI when writing such essays. This experience report examines GenAI use across three years of ethical essay assignments in a large first-year computer systems course. Across six post-essay surveys from 2024 to 2026, we analyzed 1,894 responses, linked survey data with essay grades in 2025 and 2026, and introduced a scaffolded responsible-use workflow in 2026. Under voluntary-use policies, AI adoption remained steady at around two-thirds of students, rising above 90% when use was scaffolded. Students reported mainly using AI for help with grammar and style, while tool use diversified beyond ChatGPT in 2026. Grade-linked analyses showed that AI use did not independently predict essay quality, while marker evaluations indicated high levels of critical engagement when use was scaffolded. Overall, GenAI's pedagogical value appears to depend on how assessment design guides students to evaluate and critically engage with AI-generated suggestions.