论文库

Existing adaptive-reasoning methods shorten the explicit reasoning a model writes, or skip it by switching modes; either way the answer is still written out token by token. \emph{Rather than deciding how much to think}, we explore a different paradigm, \emph{zero-or-$T$ thinking}: a single model either answers from the implicit thinking inside one forward pass, generating no token, or performs full autoregressive reasoning at a cost of $T$ tokens when that is worth paying. We realise it in a \emph{hybrid decision model} (HDM), where one frozen backbone serves both modes and a lightweight \emph{self-routing block} chooses between them. The block estimates the \emph{value of thinking}, the utility the slow mode is expected to add over the fast one, from what that implicit thinking has already produced, and the slow mode runs only where that value exceeds a price $λ$ set per request. Where no per-decision label exists, as in episodic agentic tasks, we introduce \emph{Calibrated Decision Policy Optimisation} (CDPO), a reinforcement-learning recipe that trains the block from episode-level returns. We evaluate HDM on MMLU-Pro, AGIEval, GPQA-Diamond, JevBench, BANKING77 and MiniWoB. Specifically, on JevBench HDM-4B scores $0.857$ against Jev's $0.851$ ($+0.7\%$) at a median $0.179$\,s against $0.652$ ($-72.5\%$). On the agentic venue MiniWoB HDM-27B cuts token cost by $80.1\%$.

We study algorithmic matroid intersection coloring. Given $k$ matroids on a common ground set $U$ of $n$ elements, the goal is to partition $U$ into the fewest number of color classes, where each color class is independent in all matroids. It is known that $2χ_{\max}$ colors suffice to color the intersection of two matroids, $(2k-1)χ_{\max}$ colors suffice for general $k$, where $χ_{\max}$ is the maximum chromatic number of the individual matroids. However, these results are non-constructive, leveraging techniques such as topological Hall's theorem and Sperner's Lemma. We provide the first polynomial-time algorithms to color two or more general matroids where the approximation ratio depends only on $k$ and, in particular, is independent of $n$. For two matroids, we constructively match the $2χ_{\max}$ existential bound, yielding a 2-approximation for the Matroid Intersection Coloring problem. For $k$ matroids we achieve a $(k^2-k)χ_{\max}$ coloring, which is the first $O(1)$-approximation for constant $k$. Our approach introduces a novel matroidal structure we call a \emph{flexible decomposition}. We use this to formally reduce general matroid intersection coloring to graph coloring while avoiding the limitations of partition reduction techniques, and without relying on non-constructive topological machinery. Furthermore, we give a \emph{fully polynomial randomized approximation scheme} (FPRAS) for coloring the intersection of two matroids when $χ_{\max}$ is large. This yields the first polynomial-time constructive algorithm for an asymptotic variant of Rota's Basis Conjecture. This constructivizes Montgomery and Sauermann's recent asymptotic breakthrough and generalizes it to arbitrary matroids.

General-purpose language models are trained mostly on English: over open models we review, at most, non-English pretraining data makes up only a 12% share. Models built for European languages give them far more, but score well below general-purpose models of the same size. A common concern is that, in small models, languages compete for capacity. We present Lara Mini base, a model trained from scratch on 6T tokens in English and 44 other languages, which together cover all 24 official languages of the European Union. As training progresses, the 44 other languages together come to outweigh English. To give these languages enough capacity at low cost, Lara Mini base uses a mixture of experts architecture with 7.2B total and 0.6B active parameters. Among open base models with up to 2B active parameters, it achieves the highest average score across standard English, multilingual, math, and code benchmarks, ahead of models trained on 4 to 6 times as many tokens. Its margin is largest on multilingual benchmarks in the languages it supports and on math benchmarks. It supports contexts of up to 128k tokens. We find that our MoE architecture reduces capacity competition between languages. For a single user on one GPU, without quantization, it generates text at least 1.8 times faster than any model with a higher average score. We release the weights of Lara Mini base under the CC BY-NC-SA 4.0 license. In addition, we will release T-HellaSwag, a higher-quality machine translation of the commonsense reasoning benchmark HellaSwag into 110 language variants.

Age-related macular degeneration (AMD) is a leading cause of irreversible vision loss in older adults. While chronological age is a primary risk factor, it inadequately reflects individual variations in retinal aging, limiting effective AMD risk stratification and timely intervention. Here, we propose and validate fundus-based AMD age (FAA), which quantifies disease-specific retinal aging in AMD and captures accelerated AMD-related retinal aging, thereby improving prognostic precision over chronological age. Trained on 36,782 images from 2,192 participants in the Age-Related Eye Disease Study (AREDS) dataset and validated across AREDS and AREDS2 cohorts, our model demonstrates significant improvements over traditional chronological age-based approaches. Specifically, integrating the proposed FAA into Cox proportional-hazards models significantly enhances prediction performance. FAA improved the C-index from 0.836 to 0.888 and reduced the 5-year Brier score from 0.050 to 0.046 in AREDS, with consistent gains observed in AREDS2 (C-index 0.715 to 0.756; 2-year Brier 0.078 to 0.076). FAA showed improved prognostic performance across the two datasets and under different phenotype definitions (reading center-defined vs. specialist-defined grading), supporting its potential utility for AMD risk stratification. These findings highlight the advantage of incorporating FAA into AMD risk prediction frameworks, supporting further evaluation of FAA for AMD risk stratification.

Diffusion autoencoders (DAs) are variants of diffusion models that use an input-dependent latent variable to capture representations alongside the diffusion process. The latent variable is typically used for tasks such as downstream classification, controllable generation, and interpolation. However, the latent variable has an effect on unconditional generation, with performance strongly dependent on how this variable is modelled and learnt and how it is used in the denoising process. Here, we draw a connection between DAs and a particular class of improved diffusion models---those that learn their forward (noising) process---to argue that DAs have unrealised potential as generative models. We develop a variant of DAs, termed diffusion models with $z$ (DMZ), to show that the adoption of certain design decisions, such as the choice of latent variables and conditioning method, can help realise this generative potential. Through extensive experiments and ablations, we show DMZ converges in fewer iterations, generates high-quality samples with fewer denoising steps, and learns effective latent representations for downstream tasks like classification and domain transfer.

AI coding agents such as GitHub Copilot, OpenAI Codex and Devin now open pull requests (PRs) that are merged into open-source projects. Industry reports suggest that AI-assisted code is revised or reverted soon after it is written, which raises the concern that agent code is "disposable". However, little is known about what happens to agent code after it is merged. Knowing how long this code survives, and why it is changed, can help maintainers decide how much review it needs. We performed a survival analysis of the 648,959 lines that 2,794 merged PRs added outside vendored and generated paths in 215 open-source repositories from the AIDev dataset, comparing agent-attributed and human code within the same repositories. Agent-attributed code is not modified sooner than human code. When each PR is weighted equally, agent-attributed lines are modified later (HR = 0.83, 95% CI [0.75, 0.93]). This difference comes from the supervised assistants, mainly Copilot and Codex (HR = 0.76). Code from Devin does not differ from human code over the whole period (HR = 0.95), but among PRs merged after mid-May 2025, code from both the supervised assistants and Devin is modified later. Without PR weighting, the difference is not statistically significant. When agent-attributed code is modified, the intent of the change is similar to that of human code (Cramer's V = 0.080). A file's size and recent changes predict whether agent-attributed code will be modified better than its code tokens do (AUC-ROC 0.78 vs. 0.70), and process features predict when it will be modified only weakly (Macro F1 of 0.31 against 0.25 for a random guess). Our findings suggest that agent code is not disposable, but whether it survives longer than human code depends on the agent, on when the code was merged and on how the comparison is computed.

SignalAudit-QD is a reproducible audit resource for testing whether a dated public signal adds information beyond query history in a fixed 20-term U.S. AI panel. It aligns two archived Google Trends extractions with eight USPTO AI-patent publication streams and releases origin-indexed predictions, paired losses, provenance, executable tests, and tamper-evident run manifests. Both inputs are retrospective reconstructions obtained after the modeled 2020--2023 origins; the task is therefore a pseudo-origin assay within frozen measurement vintages, not a historical deployment simulation. In the designated one-month comparison, adding a two-component patent block to a level ridge changes macro mean absolute error by +0.0012 Trends points, with conditional 95% moving-block interval [-0.0363,+0.0387]. This establishes neither aggregate benefit nor equivalence. Alternative dimensions, semantic mapping, patent definitions, a second extraction, and a three-month horizon remain unresolved; persistence-anchored and nonlinear hosts instead show model-specific worsening, tiny for the anchored fit. Matched timing controls expose availability assumptions, while a planted-signal assay detects only the largest tested injection. A within-query high-interest alert demonstrates the released interface but not online utility. The contribution is a bounded negative case study and an extensible evidence schema for auditing external-signal claims in temporal information retrieval.

We use TabPFN, a recently developed tabular foundation model, to predict three properties of energetic materials: impact sensitivity, decomposition temperature and crystal density. Without any hyperparameter tuning, TabPFN outperforms the best published models for all three properties. It also substantially outperforms a wide range of established machine-learning models, including tuned gradient boosting (XGBoost), random forests and Gaussian processes. This holds both when the models are given hand-crafted molecular descriptors and when they are given learned representations from Chemprop's message-passing neural network. In particular, for all three properties, TabPFN applied to Chemprop's own embedding is more accurate than Chemprop's predictions. TabPFN is also among the fastest models we consider, more than an order of magnitude faster than XGBoost. As TabPFN predicts the entire distribution of the target rather than just a single value, it offers built-in uncertainty quantification, which we compare with the empirical distributions of compounds with repeated measurements. Finally, we interpret the predictions using Shapley values, which rank the features by their importance.

The promised benefits of data centers are global, but their negative impacts are very much local, affecting the energy bills, aquifers and living environments of citizens across the United States. While governments and authorities have historically provided data centers with incentives such as tax credits and exemptions, AI's aggressive growth in scale has changed this dynamic, and recent months have seen them issue bills and file lawsuits in order to ensure that data centers remain a net positive influence on their legislatures. Existing pieces of federal legislation -- such as the Public Trust Doctrine, the Pollution Prevention, Clean Air and Safe Drinking Water Acts -- exist to manage natural resources, but their scope remains contested, and the current federal government's stance on competitiveness and growth of AI has put the onus on regional governments to endeavor to protect local resources. There are also state- and regional laws and regulations that are relevant for building and operating data centers since, in practice, they require substantial land, electricity, water, and infrastructure at a local level. Given this situation, communities are currently provided with the opportunity to raise their concerns regarding the potential negative impacts of how data centers are built and operated, and have existing regulations that can be used to put pressure on data center operators.

Typed System-1 (S1) decision models are increasingly deployed in LLM agents for routing, moderation, and authorization to cut inference cost, yet untrusted context can still manipulate typed outputs and alter execution branches. Unlike autoregressive System-2 (S2) models, S1 interfaces score dynamic rules and options without token-by-token decoding, mismatching attacks targeting text continuation or explicit override commands. We propose S1-Evo, a self-evolving framework tailored to S1's contract-based option competition and margin feedback. S1-Evo perturbs untrusted fields via contract-aware evidence channels to induce \emph{evidence competition} with original task evidence while preserving fact invariance. Offline on seed tasks, it optimizes candidates via feedback search guided by directional margins and re-verification, distilling verified flips into experience cards. At test time, S1-Evo retrieves relevant cards to construct perturbations for unseen tasks. Across four benchmarks and three S1 models, S1-Evo reaches 71.99\% mean ASR and 57.47\% high-confidence ASR on \texttt{jev-1.14}, and causes up to 24.00\% E2E agent violations.

Public health agencies need to anticipate how outbreaks will unfold, when they will peak, and how long demand on health services will remain elevated. We previously developed a foundation model, Mantis, which was trained entirely on simulated outbreaks---enabling us to generate the diversity and scale of training data needed to support forecasting across diseases. However, our previous work left unresolved when simulation improves predictions and how it should complement observational data. Here, we varied the balance of simulated and observed training data while holding forecasting tasks fixed and varying the available surveillance history. We found that simulation-rich training was particularly valuable with limited local history, and its relative advantage diminished as more observations became available. We also found that simulation-based training reduced exceptionally large forecast errors, suggesting that mechanistic knowledge can help compensate for limited local evidence. Building on these findings, we developed Mantis 2, which combines simulated and observational training to jointly forecast up to six variables sixteen weeks ahead. Across 30 forecasting tasks, Mantis 2 achieved the best aggregate probabilistic accuracy over short, medium, and long horizons among the machine learning, statistical, and baseline models evaluated. It missed roughly 30\% fewer outbreak peaks than the strongest comparison model, with successfully anticipated peaks identified more than six weeks in advance on average. We publicly release Mantis 2, enabling public health agencies and researchers to forecast disease outcomes without developing a separate model for each disease or location.

Time series foundation models have demonstrated strong zero-shot forecasting performance, but computational efficiency remains a central architectural challenge. Temporal self-attention incurs quadratic cost in the number of temporal tokens, while nonlinear recurrent state transitions restrict parallel computation along the temporal dimension. We introduce ci-proto, a family of time series foundation models based on Mamba-3, whose linear state recurrences support chunk-wise parallel-in-time computation during training, and inference costs that scale linearly with sequence length. The architecture alternates recurrent temporal processing with cross-variate attention, supporting multivariate probabilistic forecasting with past-only and future-known covariates. Across fev-bench's 100 forecasting tasks, ci-proto XL sets a new state of the art among time series foundation models. The five ci-proto variants also make up the entire Pareto frontier for mean scaled quantile loss and mean absolute scaled error against both parameter count and median runtime.

Full-duplex spoken dialogue models learn turn-taking, overlaps and backchannels from conversations recorded with one channel per speaker, but most available conversational audio is a single mono mix. We present Kezek, a model that turns a mono recording of exactly two people into one track per person with the original length and timing; overlaps and short replies ("yeah", "mm-hm") stay on the track of the person who said them. Kezek decides who speaks when and who is who once for the whole recording (stage 1: a two-voice activity model guided by a speaker fingerprint trained to be insensitive to phase), and a conditioned time-domain separator only carries out that decision (stage 2), with outputs that always sum to the input. Standard separation metrics miss the failures that matter for dialogue data, so we report short-reply retention, leakage onto the other person's solo, channel swaps and coverage. Against the open pipeline pyannote -> DialogueSidon on the same recordings, Kezek keeps more short replies (0.83 vs. 0.69 on Open Yap test; 0.80 vs. 0.71 over the other's speech on unseen telephone speech from CallHome English), mostly because it returns the whole recording while the external pipeline drops part of it (on the audio both return, reply retention is about the same: 0.859 vs. 0.865 on Open Yap); it also leaks less, swaps channels less often, and runs 4-5 times faster (about 230 hours of audio per GPU-hour). We also report negative results and an ideal-mask ceiling that locate the remaining gap: quiet replies under a louder voice. Code and weights are released under permissive licences.

Understanding the physics of many-body complex dynamical systems may be a non-trivial task. High-dimensional analysis approaches are often deemed necessary to prevent losing important information. Typically, these use order parameters or descriptors capturing information related to, e.g., relative positions, symmetries, etc., of the units in the studied system. However, in many cases, gaining information related to the relative positions of the constitutive units (or their velocities) alone may be insufficient, and to reach a more complete physical knowledge, one should ideally learn and correlate with each other both structure and dynamics. Here we demonstrate how to achieve such a goal efficiently by building and navigating high-dimensional Time-Derivatives (TiDe) spaces. A TiDe space can be generated for virtually any type of system/phenomenon from the time-series data collected along its observation over time. Each TiDe's dimension corresponds to a growing-order time-derivative of the extracted data, thus containing information related to different physical phenomena/events, which can be easily extracted via unsupervised approaches. We demonstrate how, by definition, TiDes can be directly analyzed without a need for prior dimensionality reduction, providing results that are intrinsically intuitive to interpret. We show the potential of the method by analyzing two prototypical example datasets extracted from molecular dynamics simulations or experimental tracking of different types of complex dynamical systems. Our results demonstrate how efficiently one can navigate and learn in information-rich TiDe spaces, which provide a robust general framework for data analysis and for studying complex dynamical systems from the data collected along their observation over time.

Generative AI has entered the workplace faster than organizations can govern it. A large share of employees now use AI tools their employer has not sanctioned, and many actively conceal that use, but the psychology of this covert technology behavior remains poorly understood. A common assumption, in popular accounts and early commentary, treats unauthorized (shadow) AI use as curiosity-driven exploration by adventurous individuals. We test this against a competing account grounded in the psychology of habitual technology use, in which automaticity rather than exploratory disposition sustains the behavior. Using survey data from 510 employed generative AI users across multiple countries, we adjudicated among individual and contextual predictors (workplace curiosity, habit, technology-acceptance beliefs, and organizational learning capability) with hierarchical regression, relative weights analysis with bootstrap confidence intervals, and full-subset dominance analysis. Habit was the strongest predictor, accounting for 28.3% of explained variance, and completely dominated joyous exploration, the curiosity dimension most aligned with the exploration account, across all 512 possible predictor subsets. Joyous exploration contributed 3.4% (95% CI [1.6, 7.4]) and fell to nonsignificance once use behavior and habit were modeled. Learning-supportive organizational conditions accounted for a further 24.1% and operated additively rather than by moderating dispositions. Parallel models decomposing the outcome showed the same dominance pattern for unauthorized use and concealment, and perceived social support for AI use was protective specifically against concealment. These findings reframe shadow AI as a habit-and-environment phenomenon rather than an individual-exploration phenomenon. The cross-sectional design means causal claims are exploratory.

The easy access to Large Language Models (LLMs) has made AI-generated texts on the web increasingly ubiquitous, making AI-text detection a crucial task. Prompts greatly influence the generation process, but existing detectors focus solely on the generated texts --- supervised methods learn from labeled texts only, and zero-shot methods rely on the text's probability distributions --- ignoring the underlying generation process, which limits their robustness in open-world scenarios. We introduce SPADE: Soft Prompts for Ai-text DEtection, a novel framework for detecting AI-generated text that transfers better across datasets. Specifically, we measure the prompt's influence and use it as a discriminative signal. We characterize the prompt's influence by introducing a Pointwise Mutual Information (PMI) term that captures how much a prompt explains a text. Based on this quantity, SPADE trains one soft prompt per class to amplify the separation, and predicts the class whose prompt best explains the candidate text. Extensive experiments on several benchmarks against various detectors demonstrate the robustness and effectiveness of our method, achieving state-of-the-art (SOTA) performance on out-of-distribution (OOD) datasets.

Provably secure linguistic steganography hides a message inside text sampled from a language model, so that without the key, a stegotext is computationally indistinguishable from an honest model sample. Nearly every such scheme assumes the receiver can recompute the sender's token probabilities exactly bit for bit. We show this assumption is fragile: across three model families, bit-exact schemes decode almost none of our messages once the receiver's floating-point precision, quantization, or GPU architecture differs from the sender's, or even when the receiver merely scores a text in parallel versus one token at a time, with the same model, precision, and hardware held fixed. We present STEALTH, a scheme built for exactly this setting. The sender draws each token with an exponential race driven by keyed pseudorandom noise, so a receiver's probability mismatch flips a token only when the race is close; a rateless spinal code, decoded by beam search, absorbs the resulting flips without the sender and receiver ever agreeing on a single distribution. We prove stegotexts remain indistinguishable from honest samples under this construction. Empirically, STEALTH decodes every message we sent for every receiver in the sender's model family, at every precision and quantization level, on smaller models, and on a different GPU architecture. STEALTH is even robust across different model families with separate vocabularies and tokenizers, decoding every message we tested. Our approach carries far more payload per token than prior robust schemes, and trained steganalyzers and a targeted statistical test both fail to distinguish its stegotexts from honest model output.

This report summarizes the Mobile Unified Multimodal Understanding (MUMU) track of the 8th Large-scale Video Object Segmentation (LSVOS) Challenge, held in conjunction with ECCV 2026. MUMU asks for a single unified model that jointly performs multi-concept image tagging over quality, scene, and event vocabularies, open-vocabulary object detection, and concise image captioning, within a budget of 0.5B parameters and 8 GB of peak inference memory. Submissions are ranked by a weighted task score scaled by an efficiency factor. We describe the tasks, rules, and evaluation protocol, report the final results, and review the methods of the top three teams. All three leading solutions build on the compact Florence-2-base model, but extend it in different ways: lightweight tagging heads with cross-task inference refinement, a shared low-rank adapter trained on coverage-aware multi-source data with task-dependent decoding, and a training-free pipeline that maps captions to tags and applies detection test-time augmentation. The results show that a compact prompt-based vision-language model can serve all three outputs when supervision, decoding, and cross-task cues are designed for each task, while open-vocabulary detection on novel categories remains the main bottleneck.

Predicting the thermodynamic properties of an alloy requires sampling its configurations across many chemical conditions and recovering free energies on a common absolute scale. Markov chain Monte Carlo (MCMC) is the standard tool, but it requires separate simulations at different conditions, and auxiliary free-energy methods such as thermodynamic integration are used to place results on a common absolute scale. Modern discrete neural samplers typically use reverse KL divergence as the objective and can be mode-seeking or biased. We present Free energy Offering Generative Sampler (FrOGS), a hybrid discrete neural sampler that couples an autoregressive model to a continuous-time Markov chain (CTMC) to be trained jointly under a single shared loss. FrOGS draws i.i.d. configurations, returns an unbiased estimate of the partition function, and gives consistent estimates of thermodynamic observables. We train a single model across a wide range of chemical conditions to produce estimates on a common absolute free-energy scale. FrOGS matches exact finite-size results on the 2D Ising model and reference phase diagrams for AgPd and CuAu, without mode collapse. We additionally compare to SEGAL, a published autoregressive baseline, and find that only FrOGS recovers the stability range of the CuAu$_3$ phase.

We present scentree, an open-source Python package for constructing scenario trees for multistage stochastic programming models using either historical data or samples drawn from a probability distribution. It combines machine learning and multivariate time series models to obtain a scenario fan that captures dependencies in the stochastic processes. This scenario fan is subsequently reduced into a scenario tree that preserves representativeness of the stochastic process, providing a flexible and extensible framework for uncertainty modeling in stochastic programming problems. The framework can also be naturally applied to two-stage stochastic programming problems. Key contributions are the joint modeling of the stochastic process, preserving correlations between all random variables, and the automation of the complete workflow, including forecasting model selection, parameter estimation, scenario fan generation, and scenario tree construction. In addition, scentree does not assume any underlying data distribution, reducing the statistical expertise required to produce a scenario tree, and is modular by design, allowing external forecasts and other partial executions of the workflow.

An important challenge in homomorphic encryption (HE) is ensuring the integrity of the encrypted computations. Although verifiable computation can address this in theory, making it practical is far from trivial. [Cascudo et al. CRYPTO 2025] recently proposed a solution to this problem for the CKKS scheme. In particular, they propose new succinct arguments that efficiently prove arithmetic circuits and range checks over the ring $R_q = \mathbb{Z}_q[Y]/(Y^N+1)$, and combine them to prove computations on CKKS ciphertexts. Despite the important progress, their work faces scalability challenges: the size of the polynomial commitment proofs, a large $O(N)$ overhead in range proofs, and the lack of an explicit method for efficient -- sublinear in circuit size -- verification. In this paper we present new techniques that remove all these bottlenecks. First, we provide a more efficient arithmetization of CKKS computations. Second, we introduce BatchFold, a new polynomial commitment for $R_q$ that supports batched evaluations and achieves shorter proofs. Third, we design packing, a novel algebraic technique that eliminates the need to commit to massive polynomials and removes the $O(N)$ overhead in range proofs. Finally, we introduce SparsePack, an efficient protocol for proving the evaluations of sparse, circuit-dependent polynomials, leveraging both packing and batched evaluations. We evaluate the efficiency of our solution on a generic neural network architecture, achieving proofs several orders of magnitude smaller and faster than prior approaches.

Efficiently locating rare targets requires balancing the exploration of novel space with the exploitation of local vicinities. While this tradeoff is well-documented in simplified environments, its behavior in realistic three-dimensional environments remains unclear. Here we analyze intermittent Lévy walks in three dimensions and show that search efficiency becomes strongly dependent on target geometry, with volume, surface area, and elongation governing performance across the Lévy spectrum. Against this geometric complexity, the Cauchy walk ($μ= 2$) emerges as a robust, near-optimal solution across a wide range of target geometries. Specifically, we prove that for convex targets of surface area $Δ$ in a domain of volume $n$, the optimal expected detection time scales as $n/Δ$, and that the Cauchy walk achieves this scaling up to polylogarithmic factors, while other Lévy strategies are suboptimal in different geometric regimes. Our findings provide a theoretical framework for biological movement patterns, including marine predator foraging and immune surveillance, and inform the design of efficient search algorithms for engineered systems such as autonomous nanoscale explorers.

Speech recognition (ASR) systems are usually compared by word error rate (WER) on recorded files, after punctuation, capital letters and number format are removed. In dictation, the audio is streamed live and the text goes into the user's message as written, so what matters is which mistakes the user has to fix. The test set, streamed to eleven ASR systems, comprises 106 scripted English dictations from one speaker (read in a quiet room, with a fan on, and whispered), 20 clips in each of eleven other language and dialect sets, and 400 recordings that join clips in two languages. Every error was sorted by a fixed rule into four levels: recognition, output formatting, language handling and, for the self-hosted models, audio dropped before the model. Systems with similar WER failed in different ways. Zoom Scribe and OpenAI's GPT-Live-Transcribe have almost the same WER on quiet English (7.4% and 7.3%), but 34.7% and 8.8% on whispered English. Zoom and both AssemblyAI models left out 26-28% of whispered words, and still 21-26% with their end-of-turn detection relaxed or off; seven other systems left out at most 3%. Qwen3-ASR's 32% came from the speech detection in our own deployment (6% when relaxed). Across 107 EARS speakers, the same four systems lost 7-31% of whispered words, while six others lost under 2%. With only English selected, Gemini, Deepgram and our self-hosted Whisper lost the non-English part of a two-language recording in 86-99% of recordings, against 33-46% with both languages selected, while GPT-Live-Transcribe, OpenAI's realtime Whisper and Meta lost it in at most 20%. Running Whisper behind a Silero voice activity detector that drops audio judged to be silence raised whispered WER from 10.1% to 51.4%. Most English results come from one read speaker. The code, the recordings, both references and every transcript are released.

State-of-the-art LiDAR-inertial odometry methods, including those on richer state manifolds, achieve strong global accuracy, yet their pose stream alternates between propagation and correction and can oscillate over short horizons even when globally accurate. Most existing methods follow these approaches: they either modify the filter to enable faster updates or estimate a continuous-time trajectory from raw LiDAR and inertial data. The latter involves extra optimization, delaying state updates and becoming less robust during rapid rotations or accelerations. We propose CT-SGAL-LIO, a continuous-time refinement module that smooths the posterior pose stream of a discrete SGal(3) estimator without modifying the estimator itself. The key novelty is to perform continuous-time trajectory refinement directly on the projected SGal(3) posterior: we fit a cubic trajectory over a bounded window of recent poses and evaluate it at a small fixed lag. This places the query time within the observed interval, so the refined pose is obtained by interpolation rather than extrapolation, while preserving causal operation. CT-SGAL-LIO improves relative trajectory accuracy on 17 sequences from three public benchmarks. Across 17 sequences from GRAND-TOUR, MCD, and GEODE, CT-SGAL-LIO improves short-term motion accuracy while largely preserving global accuracy. RTE improves by up to 70% on GRAND-TOUR and on all seven GEODE sequences, with 9.9-51.1% reductions. A 19.5-min R-Campus run also reduces posterior jumps by 85.4% with only a 4.7 mm change in the end-to-start gap. We plan to release our source code and results to support community validation and reproducibility.

Large language model (LLM) agents increasingly support long-horizon tasks by leveraging external memory and past experience. However, existing agents mainly retrieve records or summaries, lacking the ability to form and refine causal understanding of the world. Inspired by human causal cognition, we introduce CauProphet, an agent that develops and updates causal knowledge within its Causal Cognition. For each task, CauProphet combines relevant knowledge in Causal Cognition with current evidence to construct a Causal Belief, a structured belief capturing causal factors and the effects of possible actions. It uses this belief for decision-making and continuously updates both the belief and Causal Cognition as new observations emerge, enabling causal knowledge that persists beyond individual experiences. Evaluations demonstrate that CauProphet progressively acquires causal knowledge and transfers it to future decision-making. In financial research, CauProphet leverages prior discoveries to improve downstream analyses and outperforms all baselines in mean rank correlation with next-day returns, achieving a 25.4% improvement over a baseline agent. It also improves prediction accuracy from 21.3% to 28.6% in dynamic real-world settings. Further experiments validate its ability to perform reliable causal analysis and action selection across scientific and everyday tasks, demonstrating the potential of causal experience learning for next-generation LLM agents. The project repository is available at https://github.com/OpenCausaLab/CauProphet.

Printed circuit board (PCB) routing is the stage of physical layout that finds paths connecting the placed pads of each net. Public datasets in this domain are few, and most contain only the final states of routed boards, not demonstrations of the electronic design automation (EDA) engine actions that produce them. To close this gap, we propose CADET (Constructing executAble Demonstrations for Engine-grounded Training), which jointly generates synthetic boards and demonstrations built from EDA engine actions. The resulting RouteDemos provides 100,000 synthetic boards routed with zero design rule violations (DRVs), together with their demonstrations. Their distributions over 17 statistics resemble those of 142 real reference boards. CADET trains a routing policy on these demonstrations with behavior cloning and fine-tunes it with reinforcement learning (RL). On real test boards, the policy raised the rate of complete DRV-free routing from 38.0% to 67.2% over an RL-only baseline. This rate was comparable to that of the open-source autorouter FreeRouting and exceeded those of the other baselines, including GPT-6 Astra. We release the dataset and its generation code as open source at https://github.com/LGAI-Research/CADET.

We assess the descriptive complexity of *bisimilarity* or "equality of behavior" on a family of Markov decision processes over uncountable standard Borel spaces, namely *nondeterministic labelled Markov processes* (NLMP). We show that bisimilarity is analytic for processes with a uniform assignment of finitely-supported measures to each state and label. More precisely, we obtain that bisimilarity on the space of countable Kripke frames (or labelled transition systems) is classifiable by countable structures. We show that bisimilarity of well-founded ("terminating") processes is Borel. We also provide a lower complexity bound by reducing the relation of eventual equality of binary sequences, $E_0$, to the former. As a consequence, there is no countable fragment of basic modal logic with denumerable conjunctions that characterizes bisimilarity for processes of small rank. We finally apply the previous Borel-definability to study the well-founded part of discrete uniform processes over continuous spaces.

We develop a quantitative theory of the Random Language Model (RLM), an ensemble of stochastic context-free grammars, in a scaling limit where the number of hidden symbols $N \to \infty$ while the grammar temperature $\tildeε_d \to 0$ at fixed $x = \tildeε_d \log N$. In this limit, the model admits a controlled description based on a large-deviation principle over rule-usage patterns. A semi-annealed approximation maps the problem to a class of Random Energy Models with nontrivial combinatorics. We show that the RLM exhibits a freezing transition at a critical value $x_c=1/8$, below which rule usage condenses on a subset of preferred rules and language statistics acquire a nontrivial dependence on corpus length. A second characteristic scale at $x=1/2$, associated with the breakdown of Gaussian fluctuations, separates transition and babbling regimes. Across regimes, we derive explicit scaling laws for the number of distinct rules, entropy, and energy, identifying distinct frozen and critical regimes controlled by the interplay of grammar size, corpus length, and temperature. The theory resolves previous ambiguities regarding the existence of a thermodynamic transition and explains the slow approach to the large-$N$ limit as a consequence of the dependence on $\log N$. It further provides a unified framework in which universal statistical properties of language emerge from typical realizations of generative grammars, with implications for both natural language statistics and the behavior of large language models. Heaps' law and Zipf's law are both derived, and the dependence of predictive entropy on context length is rationalized. Comparison with empirical measurements of Heaps' and Zipf's laws yields parameter estimates for real languages.

Linear Temporal Logic (LTL) provides a rigorous framework for specifying long-horizon robotic tasks, yet existing approaches face a trade-off: model-based synthesis relies on accurate labeled transition systems, whereas learning-based methods often require online interaction, task-specific rewards, or specification-conditioned training. We study LTL-specified robotic planning and execution in an offline, model-free setting, where the agent is given only fixed, task-agnostic trajectory fragments, with no dynamics model, task demonstrations, or online data collection. To address this setting, we propose SAGAS, a framework that combines the compositionality of symbolic synthesis with the data-driven reachability structure learned from offline trajectories. SAGAS first learns a reusable latent reachability graph and a frozen goal-conditioned executor from fragmented offline data. For each new LTL formula, it performs task-time semantic graph augmentation to ground state-defined propositions on the learned graph and applies Büchi product search to synthesize a cost-aware accepting prefix--suffix waypoint plan executed by the frozen executor. This separation of learning and planning enables zero-shot generalization to unseen LTL formulas and new predicates supported by the offline data, without task-specific rewards, additional data collection, or policy retraining. Experiments on LTL task suites constructed from OGBench locomotion domains show that SAGAS achieves higher success rates and lower execution costs than baselines on diverse unseen tasks, with further evaluation in OGBench Scene demonstrating its applicability to robotic manipulation.