让下一篇重要论文，不再错过。

Computer-use agents (CUAs) have recently achieved superhuman performance on several long-horizon benchmarks. However, their computational cost limits their use in long-horizon tasks and real-time domains such as video games. A key bottleneck is representing each screenshot as thousands of visual tokens, which accumulate into long contexts as agents reason over extended interaction histories. We observe that most of these visual tokens are either semantically irrelevant to the agent's next action or temporally redundant across screenshots. Building on this observation, we introduce Visual Token Routing (VTR), a joint training framework for learning sparse visual representations that preserve the information needed for action prediction. In VTR, we train the vision-language model together with a lightweight spatio-temporal router that examines each image region in the current and previous screenshots and decides whether its visual token enters the language model. We optimize the router using an adaptive loss to keep the number of visual tokens as small as possible while minimizing performance degradation. With just 2,000 training steps, VTR matches the performance of the full-resolution model on OSWorld, CUA-World, and video game action prediction, while using 2.5-5.8x fewer visual tokens. These savings reduce inference cost by up to 3.3x and speed up prefill by up to 2.9x. VTR also outperforms lowering the image resolution and other baselines for dropping tokens, by up to 12.8 points on OSWorld. Overall, VTR enables computer-use agents to process only the visual tokens they need, allowing them to be used in long-horizon tasks and real-time domains at a fraction of the computational cost. Code and models are available at https://pranjal2041.github.io/vtr

Video generation models synthesize realistic motion, but can they respond correctly to changes in initial geometry? We introduce GeoFlip, a video generation benchmark centered on physical outcome flips induced by geometry. With shared physical settings and textual prompts held fixed, GeoFlip applies controlled interventions to support extent, object placement, and mechanism configuration, constructing paired scenes with contrasting reference events, such as remaining stable versus falling or tipping. These pairs test whether models translate geometric differences in the first frame into correct physical evolution. The benchmark comprises 482 scene pairs and 964 reference videos across six physical categories, with reference outcomes established through procedural scene construction, physics simulation, and numerical validation and verification of physical events. We evaluate 16 models for image to video generation along three dimensions: event correctness, structural integrity, and similarity to reference videos. We examine evaluation consistency with independent vision language judges. On the 442 conditions shared by all models, the highest event correctness averaged across tasks is 27.98%, revealing substantial limitations in physical generation conditioned on geometry. Model strengths vary across physical mechanisms, and the three evaluation dimensions capture distinct capabilities. Typical failures include ignoring geometric changes, generating incorrect object motion, bending rigid objects, and introducing additional components. Rank correlations between judges reach 0.906 for event correctness and 0.929 for structural integrity. By combining controlled geometric interventions, verifiable physical references, and multidimensional evaluation, GeoFlip assesses progress from visual realism toward physical consistency conditioned on geometry.

Common methods for monitoring features in large language models (LLMs) such as linear probes and sparse autoencoders (SAEs) assume that features maintain a consistent representation across contexts. LLMs, however, can flexibly adopt different perspectives or "personas" that suggest different underlying systems of conceptual relations. In this paper, we examine whether prompting an LLM to adopt a persona alters the representational geometry of the model's latent space. We extract representations of over 9,000 common words and 33 semantic axes from models adopting liberal and conservative political personas and compare these against two neutral baselines. We find that persona conditioning systematically alters both the projections of words onto semantic axes and the alignments among the axes themselves in directions consistent with the specified ideology. We then test the implications of this reorganization for two safety-relevant features: truthfulness and morality. Linear probes, SAE-based classifiers, and classifiers built from manually selected word representations, all trained without a persona, produce divergent results when applied to models conditioned on political personas, producing politically slanted ratings of truth and morality on controversial statements. These findings suggest that internal representations, as they are conventionally extracted, can exhibit different geometric relations when the model adopts personas, potentially undermining monitoring efforts.

Faithfulness requires that algorithmic explanations accurately reflect the model's behavior. Yet existing operationalizations of faithfulness can disagree about whether the same explanation is faithful, leaving developers and auditors to choose among conflicting evaluations. In this paper, we investigate disagreement among evaluation criteria and offer a practical path forward for evaluating explanation faithfulness. We first identify how operationalizations of faithfulness differ. We introduce a faithfulness functional that locates their disagreements in what an explanation claims about the model, which comparisons bear on that claim, and how agreement across those comparisons should be aggregated, and we reconstruct existing evaluation criteria within it. We argue these reconstructions support understanding faithfulness as a common though incompletely specified relation between the model and explanation, so the differences reflect choices left open by the general requirement. Justifying an operationalization then comes down to justifying these choices, which, on our view, draws on explanatory, institutional, and normative considerations. We support this argument with a case study of adverse action notices, reconstructing the evaluation implicit in an industry-proposed procedure and identifying which of its choices the setting justifies and which remain open. Our work gives auditors and practitioners a way to navigate disagreeing evaluations of faithfulness. It also explains why the literature should expect multiple operationalizations, while showing how concrete settings can narrow the choice among them. More broadly, our analysis shows how a shared requirement of faithfulness can support contextually justified evaluations despite disagreement among existing metrics.

A central yet under-explored question in Multimodal Large Language Models (MLLMs) is where to connect---how visual information from a hierarchical vision encoder should be wired into the layer-wise semantics of a large language model. Mainstream MLLMs inject a fixed single-layer visual representation into the LLM and treat all LLM layers as a single consumer, ignoring that different layers demand different visual granularities. Recent multi-layer fusion methods either compress hierarchical features on the encoder side or rely on predefined sparse or hierarchical layer-to-layer connections, and therefore stop short of modeling adaptive cross-layer interactions between the ViT and LLM hierarchies. We address this gap with DGP (Dynamic Gated Pathways), a routing mechanism that establishes all-to-all connections between every ViT layer and every LLM layer, with each LLM layer dynamically gating its preference over visual layers conditioned on its own semantic state. Built on top of LLaVA-1.5-7B and 13B, DGP delivers consistent gains over the LLaVA-1.5-7B baseline (e.g., +4.2 on SQA$^{I}$, +3.3 on MM-Vet, +3.1 on VizWiz and LLaVA$^{W}$) and surpasses prior multi-layer visual fusion methods on most benchmarks. Layer-wise routing analyses further show that different LLM layers do select different visual granularities on demand, supporting both the effectiveness and interpretability of the proposed dynamic gated pathways. The code is available: https://github.com/YingyingYan1/DGP.

Grain growth during heat treatment strongly affects the microstructure of polycrystalline materials and, consequently, their mechanical properties. Although numerical methods such as front-tracking or front-capturing simulations can accurately describe grain growth, their high computational cost limits their use for long or non-isothermal thermal cycles. In previous work, grain growth was successfully predicted under isothermal conditions by treating grain size distributions as time series and forecasting their evolution using recurrent neural networks, allowing complete grain growth trajectories to be generated in seconds rather than through costly PDE-based simulations. In the present work, this forecasting framework is reformulated to handle non-isothermal conditions by explicitly incorporating thermal information into the model architecture. Grain size distributions are encoded using a convolutional neural network, thermal descriptors are embedded through a dedicated multilayer perceptron, and temporal evolution is predicted using a recurrent network. Instantaneous temperature and thermal rate are provided at each time step, allowing the model to adapt its predictions to changing thermal conditions. The framework is trained and evaluated on datasets generated from validated front-tracking simulations across a wide range of non-isothermal thermal paths. The results show that the thermally conditioned model accurately reproduces grain growth dynamics under complex thermal histories, achieving, in k-fold cross-validation over all simulations, mean relative errors of 2.99% (one-hour cycles) and 4.51% (four-hour cycles) with coefficients of determination exceeding 0.99, and mean relative errors between 1.91% and 14.66% on ten simulations following four unseen thermal histories of increasing complexity.

Large language model (LLM) coding agents develop software through experimentation, offering an opportunity to design the serving systems supporting their own operation. Effective serving requires coordinating implementation and operating decisions under session dependencies, latency constraints, and correctness requirements. We introduce ServerBench, a benchmark for agentic design of complete, specialized LLM serving systems. Agents experiment against a common objective, dollars per dollar: latency-qualified token value per dollar of GPU cost. Stateful replay preserves session dependencies, and independent verification checks the submitted system. Across 11 coding-agent models and three deployments, Claude Sonnet 5.5 achieves 2.97 dollars per dollar on Gemma E4B, $4.87\times$ the vLLM reference; 22 of 33 runs yield verified systems, and 13 exceed the strongest baseline. Our analysis identifies three findings: high-value designs coordinate capacity improvements with load selection; agents interpret performance feedback through workload-relevant experiments; and reliable delivery requires verification of the execution paths introduced by optimization. Overall, ServerBench supports systematic study of how agents translate service objectives and feedback into effective, verified serving systems for given task specifications.

Large language model (LLM) assistants increasingly help users with everyday activities that involve time-dependent constraints. However, users may not mention these constraints in later requests, requiring assistants to recognize and adjust their advice accordingly. To evaluate such capability, we introduce TASC (Temporal Action Safety in Conversation), a benchmark that evaluates whether LLMs apply time-dependent constraints when responding to conversational requests. TASC consists of 3,418 inputs derived from 128 templates, along with a separate set of 33 scenarios designed to compare models' ability to calculate elapsed time with their ability to apply that information. Across 13 models, assistants account for time-dependent constraints in only 30.5% of cases where the proposed action violates a rule, while rarely warning users unnecessarily. More importantly, models fail to apply these constraints in 62.1% of cases despite correctly calculating the elapsed time. Explicitly providing the timing resolves only about half of these failures, revealing a gap between temporal reasoning and its application in conversational advice.

Autoregressive image generation has recently emerged as a competitive visual generation paradigm, yet its generation order is often predefined or heuristic, such as next-patch prediction, random-order prediction, or one-step generation for specific scales. We observe that these strategies can cause models to overlook important structural dependencies in images, thereby limiting both structural robustness and reasoning capability. In this work, we propose a context-aware autoregressive generation framework, where the model dynamically determines the next token location based on global context and current local information. Rather than following a fixed or random order, our method allows the generation process to adapt to image structure and semantic dependencies. Extensive experiments show that context-aware generation significantly improves structural robustness across different image AR paradigms, including masked autoregressive and scale-wise autoregressive models. Beyond quantitative improvements, we further observe emerging reasoning properties: the model can better understand complex semantic instructions and generate images that more faithfully follow the given guidance. These findings suggest that generation order is a critical yet underexplored factor in image autoregressive modeling, and that context-aware decoding can unlock stronger reasoning-aware visual generation capabilities. Code is available at https://github.com/Hpesojyjx/Saccades-AR.

Under stochastic decoding, a single context can elicit many different answers from a language model. Characterizing this variation typically requires many expensive autoregressive rollouts, each revealing only one realization. We ask whether the activations produced while reading the context can directly forecast the model's rollout activations. We train context--answer metamodels that map a context activation sequence to point or distribution forecasts of mean-pooled answer activations. We find that (i) the four metamodel families we study forecast mean-pooled answer activations with high fidelity, and the distribution metamodels also capture rollout variability; (ii) performance improves as the training set grows to 500,000 prompts, and the metamodels generalize zero-shot to out-of-domain datasets; (iii) their forecasts retain behavior-predictive signal; and (iv) context activations from Qwen2.5 predict mean-pooled answer activations from Qwen3.5 and Gemma 4, suggesting that activation forecasts can transfer across models. Together, these results suggest that context activation traces can support scalable forecasts of behaviors in LLM answers before the first answer token is generated.

Looped Transformers increase effective depth by re-executing shared layers, but three limitations hinder their adoption: memory grows with the loop count, compute can only be adjusted in whole-block steps, and the loop count is hard to change after training. We propose LooM (Loop-invariant Memory), which addresses each in turn. In a causal encoder-decoder (CED), LooM loops only the decoder layers after the first, which read a global memory written once per token, and each of them publishes the key-value (KV) entry for a token only on its first visit, so the cache does not grow with the loop count. These layers loop independently, providing every budget from 3 to 21 layer executions per token, and one model trained on randomly sampled budgets covers all of them without a router. With enough loops, LooM outperforms a single-pass baseline trained on the same data on language modeling, zero-shot accuracy and long-context perplexity, while using the same cache size. Our analysis further shows that the training budget distribution decides what extra depth buys. These results suggest that a CED is a natural place to loop: memory is written once, and depth becomes a budget chosen at deployment.

We present a toolchain for development of ML models on AWS EC2 F2 cloud FPGA infrastructure from high-level language, implementing a complete Vitis HLS to HDK bridge to AFI workflow that has been absent from the academic literature despite eight years of AWS FPGA availability. We introduce a four-level methodology (FP32 PyTorch, bit-accurate quantized Python, HLS C-simulation, physical FPGA) to aid in localizing errors, including FPGA-specific circuit errors. Together these tools allow coding agents to effectively implement FPGA designs from user directions, testing and debugging as needed. We document the deployment workflow, timing closure strategies, and practical HLS engineering decisions that proved crucial. We validate using TinyBERT fine-tuned for address Named Entity Recognition, implementing the embedding layer and the first of four encoder layers on the FPGA hardware. Our implementation streams weights from host DRAM, yielding poor performance compared to CPU/GPU implementations; we discuss forward paths.

Gated-delta linear attention layers such as Gated DeltaNet (GDN), Kimi Delta Attention (KDA), and GDN-2 are tuned for GPUs, where small tensor-core tiles and fast vector registers make per-channel decay rescaling and sequential in-block solves cheap. On Google TPUs, which concentrate compute in wide $256 \times 256$ systolic matrix units backed by a slower vector unit, porting these layers directly leaves half of each matrix tile empty and bottlenecks on vector operations. We present \textbf{Systolic Linear Attention (SLA)}, which reshapes gated delta attention around the matrix unit by widening keys to $256$ with grouped value heads, grouping decay gates into four $64$-channel tiles, and computing the chunk solve entirely with block matrix products in a single fused kernel. On TPU v7x, a full \texttt{SLA-R4} layer runs $1.6\times$--$2.5\times$ faster than the open-source Tokamax KDA kernel ($1.4\times$--$2.4\times$ with exact chunk inversion), driven primarily by head sharing and kernel fusion. Across a $2{,}048$-chip TPU v7x pre-training study of sparse Mixture-of-Experts (MoE) models from $6.5\text{B}$ to $63.1\text{B}$ parameters, SLA trains $11\%$--$24\%$ faster than KDA at similar perplexity, and reaches slightly lower perplexity than KDA in smaller matched-seed runs.

Whether navigating a building, operating a robot, or playing a game, an agent that acts effectively in an environment must first learn an internal model of how that environment works. Partially-observable Markov decision processes (POMDPs) provide a flexible modeling class for such internal world models, but learning them from observation-action trajectories alone is challenging and typically requires extensive environment interaction. We ask whether language-model priors can reduce costly interaction by leveraging prior knowledge, and introduce Pinductor (POMDP-inductor): an LLM proposes candidate POMDP models from a few observation-action trajectories and iteratively refines them to optimize a belief-based likelihood score. Despite relying on strictly less information, Pinductor achieves comparable performance and sample efficiency to LLM-based POMDP learning methods with privileged access to hidden states. Moreover, it is up to 330 times more sample-efficient than standard POMDP baselines such as DreamerV3, at comparable computational cost. Further results show that performance scales with LLM capability and degrades gracefully as semantic information about the environment is withheld. Together, these results position language-model priors as a practical tool for sample-efficient world-model learning under partial observability, and a step toward generalist agents in real-world environments. Code: https://github.com/atomresearch/pinductor.

Naive on-policy RL is often bottlenecked by the capabilities of the base model --- insufficient coverage can result in an inability to learn from certain rewards. At the same time, rewards in modern post-training systems often come from rich textual sources, e.g. judge models with a rubric of criteria. Can we utilize such structures to provide a rich directional signal that enables improving on otherwise difficult to optimize rewards? In this work, we examine a framework we refer to as targeted guidance that learns from per-turn textual feedback via an on-policy self-distillation objective. Our focus is not only on directly improving hard-to-optimize behaviors, but on how to do so throughout the training process without degrading baseline RL performance. Over a range of settings across an agentic multi-turn software engineering environment, we find success by combining targeted self-distillation with a per-criteria RL reward, and by limiting feedback to turns where the policy specifically fails. We further examine the design choices that lead to stable entropy and reduce oscillatory behavior throughout training.

We study which training and calibration data help three open models solve ARC-AGI-2 tasks when the solver runs offline on four L4 GPUs for at most 12 hours. We compare data offline on public tasks and report 189 runs on the 120 hidden semi-private tasks. For the NVARC Qwen3-4B grid model, we built a corpus of 12,000 executable rules in a typed language and 95,993 tasks; a rule is admitted only if its behaviour on a fixed probe set differs from that of every admitted rule, and a task only if its pairs refute a shortlist of nearest competing rules and defeat 12 shortcut solvers. Continued fine-tuning on 1,920 real task records with revised four-GPU scheduling raised the semi-private score from 28.96 (mean of the released model) to 33.06 (mean of five runs, best 34.86); no checkpoint trained with synthetic tasks, even all 95,993 of them, improved on the real-task model. For the Qwen3.8-27B reasoning model, we fine-tuned on 1,036 of its own reasoning traces with correct final answers and compared sets that add traces of other models and synthetic tasks. With an inference procedure built around the model (a thinking cap enforced in the serving engine, a Python program checked on the demonstrations after each answer, a long first attempt per input), the score rose from 26.81 to 44.03, the level the ARC Prize Foundation measured for the base model without a time limit (42.36); most of the gain came from the inference procedure. For the Qwen3.8-Flash-Next mixture-of-experts model (125B parameters, 6B active), GPU memory limited concurrency. Pruning from 512 to 384 experts per layer by router-weighted activation saliency grew the key-value cache from 0.55M to 1.39M tokens; a 296-expert list adapted with ARC routing statistics, keeping experts for program tokens, plus an FP8 indexer cache, reached 1.96M. With a tool that checks programs, the model scored 56.67, our best result.

We present CivBench, an open-source benchmark for evaluating language model agents in long-horizon, tool-mediated environments through the Model Context Protocol (MCP). A single episode can span over 300 turns and produces thousands of tool calls over a large action space, requiring sustained planning, state monitoring, and execution under partial observability. The environment exposes 76 MCP tools and a deterministic narration layer that retrieves and formats game state as structured text. We use CivBench to characterise agent behaviour across four model families in 23 admissible runs. The sample is a pilot, not a model ranking: aggregate outcomes do not reliably discriminate models at this scale. Instead, we introduce two interface-level metrics that the environment makes measurable: Proactive Monitoring Rate (PMR), capturing whether agents actively query latent strategic state, and RAG@10, capturing whether commitments stated in structured planning reflections are executed within ten subsequent turns. Across runs we observe two consistent patterns under a shared playbook protocol. Agents under-monitor strategically relevant state that is available but requires explicit querying: despite playbook guidance to query victory progress every 20 turns, agents do so only every 30-75 turns, and in 8 of 20 detectable defeats they failed to query within the 20-turn warning window before game end. Agents also frequently fail to execute near-term commitments stated in their own planning reflections (RAG@10 between 48.2% and 65.8% across models). Both patterns arise despite tool access and explicit guidance, and we interpret them as deviations under instruction rather than absences of capability. We release the environment, scenarios, logs, metrics, and analysis pipeline at https://github.com/lmwilki/civ6-mcp.

Guiding pretrained flow-based generative models for conditional generation or to produce samples with desired target properties enables solving diverse tasks without retraining on paired data. We present ESS-Flow, a gradient-free method that leverages the Gaussian prior of the source distribution in flow-based models to perform Bayesian inference directly in the source space using elliptical slice sampling. ESS-Flow only requires forward passes through the generative model and observation process, making it applicable even when gradients are unavailable, such as with non-differentiable simulation-based observations or quantization in the generation process. We demonstrate its effectiveness on scientific applications including generating materials with target properties, protein backbone structures with improved designability, and synthesizable, drug-like molecules.

Erasures are used to model faults at a subset of known locations in various platforms used to construct fault-tolerant quantum computers. Beyond the utility in erasure recovery, efficient decoders for erasures are useful to both understand error decoders and as a subroutine for error decoding. Classical expander codes, which belong to the family of low-density parity-check (LDPC) codes, can be equipped with the Peeling decoder which works in linear time. Our main result is the algorithm Small-Set-Peeling, a quantum generalization of the classical Peeling decoder for erasures, that applies to quantum expander codes (Tillich, Zémor, IEEE T-IT 2013; Leverrier, Tillich, Zémor, FOCS 2015). This decoder can correct up to $D-1$ erasures, where $D$ is the (design) distance of the code and, importantly, is the first decoder that is guaranteed to do so in linear time. Combining with the erasure-conversion algorithm Small-Set-Find (Krishna, Livni-Navon, Wootters, SODA 2024), our result also implies the existence of a novel graph-local linear-time decoder for Pauli errors.

Diffusion and rectified flow (RF) models generate high-fidelity images and videos, but their iterative velocity-field evaluations are computationally expensive. Existing caching methods accelerate sampling by skipping timesteps, yet their coarse approximations introduce accumulated errors over long skip intervals and degrade quality under aggressive acceleration. We propose TACache (Trajectory-Aware Cache), a training-free acceleration framework following a skip-then-compensate paradigm. TACache performs an orthogonal decomposition of discrete velocity acceleration along the RF trajectory into a parallel component and an orthogonal residual, isolating the magnitude and directional sources of per-step approximation error. The framework operates in two stages: offline, cumulative variation thresholds on the magnitude and direction indicators yield the skip schedule and bound how far each skip interval may extend; online, at each skipped step the offline statistics are combined with the sample's historical orthogonal direction to reconstruct the skipped velocity without additional model evaluations. Main comparisons on BAGEL, FLUX.1-dev, and Wan2.1-1.3B show that TACache achieves up to $4.11\times$ speedup on text-to-image generation and $2.11\times$ speedup on text-to-video generation, with consistent improvements over prior cache-based methods on most quality metrics. Our code is available at https://github.com/XiaoLiu1237/TACache.

A natural way to compare two graphs is to align their vertex sets by a bijection and to measure the difference of the aligned adjacency matrices in a matrix norm. Depending on the norm, this yields, for example, the edit distance, the cut distance, the spectral distance, and the local edit distance. To make these distances applicable to graphs of different order, one can consider blowups of the graphs or, equivalently, relax bijections to fractional alignments, that is, optimal transport plans between the uniform distributions on the vertex sets, in the spirit of Gromov-Wasserstein distances. We investigate how much the optimal transport versions can differ from their bijective counterparts, that is, we quantify their integrality gaps. For the edit distance, we improve the best known upper bound on the integrality gap from 3 to 2 and the lower bound from 6/5 to 4/3. For the cut distance, we approach Lovász's conjecture that $δ_\square \le 2δ_\square^{\rm OT}$ from both sides. We improve the previously best known upper bound $δ_\square \le 32 \cdot (δ_\square^{\rm OT})^{1/67}$ (Borgs, Chayes, Lovász, Sós and Vesztergombi, Adv. Math. 2008) to $δ_\square \le C_\varepsilon \cdot (δ_\square^{\rm OT})^{1-\varepsilon}$ for every $\varepsilon > 0$, and we construct graphs $G,H$ with $δ_\square(G,H) = 2δ_\square^{\rm OT}(G,H)$, establishing a factor-2 lower bound that matches Lovász's conjecture. For the local edit and spectral distances, we obtain upper bounds of order $\sqrt{t\ln(1/t)}$ and $\sqrt{t}$, respectively, where $t$ is the corresponding optimal transport distance, and we show that, unlike for the edit and cut distances, no linear upper bounds are possible. Finally, we show that computing the optimal transport distances is NP-hard, and we give hardness of approximation results for the cut, spectral, and local edit distances.

With the growing need for evaluating Large Language Models (LLMs) and their applications to speech, challenges persist in summarizing and evaluating conversations that lack a clear end goal. We introduce SAUCE -- a reference-free, fact-based evaluation pipeline for cross-lingual conversational speech summarization. It measures the accuracy and the fact coverage of a summary through the entailment between conversation and text. We compare SAUCE against several popular summarization metrics and demonstrate the effectiveness of capturing information loss due to transcription and translation error and identifying broken summaries. Crucially, unlike black-box LLM evaluators or dense embedding metrics, SAUCE is inherently explainable: it maps summary scores to discrete, verifiable facts, allowing users to pinpoint exact hallucinations or omissions. We illustrate how this interpretability helps developers systematically profile LLM behaviors and gives end-users an actionable tool to verify summary accuracy in noisy, real-world conditions. Preliminary investigations show SAUCE strongly align with human judgment.

Structure property prediction is a classical task in machine learning, ranging from molecular activity and materials properties to local functional responses measured in microscopy. For image-based regression, two choices define the information available to the predictor: the size of the image neighborhood and the spatial resolution retained within it. A small neighborhood can omit relevant surrounding structure, whereas a large, finely sampled neighborhood can introduce redundant or irrelevant information. Here, we develop a framework for constructing spatial descriptors for prediction of local spectra and hysteresis loops. We benchmark regression models, use SHAP maps to examine their spatial attribution, and explore the effect of the patch size and compression to establish the relationship between predictive accuracy and descriptor dimensionality. We then use a genetic algorithm to optimize spatial partitions and the assignment of mean, minimum, and maximum pooling operations. Each retained region contributes an explicit measurement of the image, allowing the search to balance prediction error against the number of descriptor values. Cartesian and polar constructions provide complementary ways to organize this information. The resulting descriptors make it possible to examine where spatial detail is needed, where regional summaries are sufficient, and how these choices depend on the predictor and the measured response. The same approach can be used for molecules and solids once the genetic algorithm rules for these are established.

Financial backtests of large language models (LLMs) are meaningful only if the model did not already know the outcomes it is asked to predict. Practitioners guard against this by trusting a vendor's stated knowledge cutoff, by asking the model what it knows, or by instructing it to answer as of a past date. To verify whether any of these safeguards can be trusted, we introduce FinDendro, a benchmark of 2,226 dated U.S. facts from 1999 to 2026, covering monthly macroeconomic series, month-end stock prices and Federal Reserve decisions. Each fact yields four prompts: a value question, Recall and Predict questions that pair each candidate value with a claim of knowing or guessing without a declared date, and a Predict question with a historical declared date. Controls for persistence and dominant outcomes keep chance accuracy near 0.27. Results show that for most frontier models, self-reports understate knowledge. GPT-5 answers 76% of controlled facts correctly but claims to know 38%, and answers 61% of the facts it says it does not know. Knowledge claims can persist beyond the measured knowledge boundary, by up to seven quarters, and a declared date removes most of the claims of prior knowledge while leaving accuracy unchanged. Backtests inside a model's training window therefore cannot be certified leak-free by self-report, instruction or stated cutoff. These findings motivate behavioral measurement of temporal knowledge boundaries, and FinDendro provides a benchmark and procedure for selecting backtest periods based on demonstrated factual knowledge.

World models learn to predict the temporal evolution of visual observations given a control signal, potentially enabling agents to reason about environments through forward simulation. Because of the focus on forward simulation, current world models generate predictions based on factual observations. For many emerging applications, such as comprehensive evaluations of physical AI behavior under varying conditions, the ability of world models to answer counterfactual queries -- such as "what would happen if this object was removed?" -- is of increasing importance. We formalize counterfactual world models that additionally take interventions as explicit inputs, predicting temporal sequences under hypothetical modifications to observed scene properties. Traditional world models operate directly on entangled pixel-space representations where object properties and relationships cannot be selectively modified. This modeling choice prevents targeted interventions on specific scene properties. We introduce CWMDT, a framework to overcome those limitations, turning standard video diffusion models into effective counterfactual world models. First, CWMDT constructs digital twins of observed scenes to explicitly encode objects and their relationships, represented as structured text. Second, CWMDT applies large language models to reason over these representations and predict how a counterfactual intervention propagates through time to alter the observed scene. Third, CWMDT conditions a video diffusion model with the modified representation to generate counterfactual visual sequences. Evaluations on two benchmarks show that the CWMDT approach achieves state-of-the-art performance, suggesting that alternative representations of videos, such as the digital twins considered here, offer powerful control signals for video forward simulation-based world models.

Efficient benchmarking estimates a system's full benchmark performance from a small number of task executions. Standard approaches rely solely on the binary correctness of executions from a collection of already-evaluated reference systems. An agentic system, however, produces a rich trace for each execution that can reveal substantially more about its capabilities. Leveraging this information requires useful trace representations. We propose representing a trace by extracting a structured summary of it with an LLM and embedding the summary with a standard text encoder. These representations preserve correctness- and task-related signal while minimizing the presence of system identity-, model family- and harness-related signals that dominate raw trace embeddings. Given this representation for each task execution from all reference systems, we construct low-dimensional representations of the systems and predict a new system's score from its position among the reference systems. On SWE-bench Verified and Terminal-Bench 2.0, augmenting an item-response model with this geometry-based prediction reduces mean absolute error from 0.13 to 0.08 (Wilcoxon $p < 10^{-4}$) and from 0.14 to 0.12 (Wilcoxon $p < 10^{-3}$), respectively, when provided a single random executed task. Further, given access to the full set of reference systems, our approach dominates baselines for any given number of executed tasks on both benchmarks.

Hamilton-Jacobi (HJ) Reachability identifies the unsafe states of a dynamical system from which it can be forced into failure under worst-case disturbance. However, classical grid-based solvers scale exponentially with state dimension. DeepReach improves scalability by replacing the grid with a neural value function trained on the HJ-Isaacs (HJI) variational inequality (VI), but penalizes both sides of the residual equally. This allows the learned value to lie above the true one and unsafe states to be misclassified as safe. We propose Conservative DeepReach (ConDR) which aims to learn a subsolution of the HJI VI such that the learned unsafe set contains the true one. ConDR enforces the obstacle branch and terminal condition by construction, pushes the HJI residual to a safety margin $ε$ below zero with a one-sided loss, and prevents the degenerate solution of classifying everything as unsafe. We prove that the exact solution of the margin-shifted VI that ConDR trains toward is a subsolution of the original VI so its unsafe set contains the true one and its value lies within $ετ$ of the true value function. We compare against DeepReach and MADR in the established systems of Air3D, Air6D, a 9D multi-vehicle system, and a 13D quadrotor. ConDR reports the lowest rate of unsafe state misclassified and the lowest closed-loop collision rates with at most a small increase in intervention. On Air3D and Air6D, ConDR tolerates model mismatch that scales with $ε$. At matched false-negative rates, ConDR's unsafe sets are smaller than the baselines showing the safety is not obtained by trivial over-approximation.