Papers

Autonomous-driving question answering requires reasoning over structured scene information, yet existing vision-language approaches largely delegate heterogeneous reasoning operations to a single neural inference process. We argue that this uniform strategy overlooks a fundamental distinction: some queries admit exact symbolic solutions, while others require semantic interpretation. We introduce a query-adaptive neuro-symbolic reasoning framework that explicitly allocates computation according to the nature of the query. At its core is a hierarchical Spatiotemporal Scene Graph (STSG) that separates persistent object identities from frame-specific states and represents spatial relations and temporal transitions as explicit directed structures. Given a query, a symbolic executor first attempts to resolve it through exact graph operations; only when symbolic execution abstains is an LLM invoked for semantic reasoning. For these unresolved queries, query-conditioned graph retrieval and evidence filtering preserve relation direction, temporal locality, and object semantics, providing the LLM with compact and verified task-relevant evidence. This design shifts the role of the LLM from a universal reasoning engine to a targeted semantic reasoner, while allowing deterministic computation to be handled exactly and efficiently. We evaluate the framework on 5,916 NuScenes-QA questions across all ten scenes of nuScenes v1.0-mini under an oracle-perception setting. The complete system achieves 80.63 percent overall accuracy with GPT-5.4-mini, improving over the corresponding LLM-only configuration by 5.48 percentage points; with DeepSeek-V4-Flash, the improvement reaches 6.64 points. The largest gains occur on counting questions, with improvements of 10.20 and 12.61 points, respectively. These results show that selective reasoning improves both accuracy and inference efficiency.

Recent advances in intrinsic two-dimensional (2D) magnets have created a need for a unified understanding of finite-temperature phase transitions across different spin-symmetry classes. Here, we investigate finite-temperature magnetism in the classical 2D Heisenberg model using large-scale Monte Carlo simulations, continuously spanning the easy-plane, isotropic, and easy-axis regimes. We find that characteristic temperatures are strongly affected by the interplay of finite-size effects and spin symmetry, induced by magnetocrystalline anisotropy, being most pronounced in the isotropic Heisenberg regime and substantially reduced toward the easy-plane and easy-axis limits. Examining Berezinskii–Kosterlitz–Thouless (BKT) physics in the easy-plane regime, we find that the characteristic temperatures independently estimated from the magnetization and heat capacity deviate significantly from those proposed by BKT theory, but match a previously reported anomaly in the XY-model. We report that characteristic temperatures instead are correlated with the proliferation of vortex-antivortex pairs, suggesting the gain in entropy by their accelerated nucleation as the origin of the anomaly. Establishing this connection both theoretically and numerically, we argue that the thermodynamic characteristic temperature in finite easy-plane Heisenberg magnets, including the deviation from BKT theory, is subject to the interplay between finite-size spin correlations and vortex proliferation. These results clarify how experimental and technologically relevant temperature scales evolve with magnetic anisotropy, spin symmetry, and system size in finite 2D magnets.

Detecting compromised business ad accounts is a challenge in digital advertising, as attackers exploit hijacked accounts to launch fraudulent campaigns. Large Language Model (LLM) agents show promise for integrity enforcement, but hallucinated mistakes on hard cases create business friction. In a study we find the autonomous agent is a strong, recall-heavy signal extractor but an unreliable final arbiter, conceding precision on ambiguous decisions. We therefore keep the agent as an investigator that emits a structured, interpretable signal vector, and delegate the verdict to a neuro-symbolic stage: symbolic rules discovered by Inductive Logic Programming (FOIL-IE), a Naïve Bayes calibration layer, and a data-tuned contradiction layer. Evaluating on a compromise-over-sampled population and a realistic low-prevalence sample with subject-matter-expert labels, this arbiter substitution raises MCC from 0.295 to 0.435 ({\Delta}MCC +0.139, 95% CI [+0.026, +0.245], p=0.018, paired bootstrap), lifting precision from 0.250 to 0.446 (1.8x) at a recall cost (0.920 to 0.660). Benchmarked under identical conditions, it also edge tree ensembles (0.386).The rules encode domain w labels while remaininginterpretable and auditable.

As local small language models (SLMs) increasingly collaborate with more capable cloud large language models (LLMs), a natural privacy question arises: Can a local SLM obtain cloud LLM guidance while protecting user privacy? Existing privacy-preserving SLM-LLM frameworks primarily hide sensitive values while preserving task semantics, which can still expose what the user is trying to accomplish. For example, allocating scarce medical supplies across hospitals may signal an emerging public-health emergency, while rebalancing an investment portfolio may reveal a private investment strategy, even when names and numerical values are hidden. Recent decoy-based methods further obscure task intent by hiding the real request among alternatives, but stronger protection relies on more decoys or semantic abstraction, increasing overhead or risking utility loss. More fundamentally, existing work does not systematically characterize the components of private task intent or how each should be protected. We therefore introduce task-private consultation, which characterizes task intent through two components: task context and task operation. To the best of our knowledge, this is the first systematic study of these components and their individual and joint protection in local-cloud SLM-LLM consultation. To realize this setting, we propose PriCon, an end-to-end framework that transforms the task itself through recoverable mathematical reformulation rather than hiding it among alternatives. A local closed-loop refinement mechanism further maintains privacy and recoverability throughout consultation. Experiments on 100 tasks show that PriCon reduces cloud-side task-intent inference Hit@1 to nearly 0%, versus 93-99% under sensitive-value removal and 3-30% under decoy-based protection, while preserving cloud-assisted utility.

Graph neural networks typically rely on dense, floating-point node representations, which can impose substantial memory and computational costs. Binary graph hashing offers an alternative by encoding node information as compact bit strings. However, existing approaches either sacrifice global topological information for computational efficiency or incur substantial generation costs. We introduce an ultra-fast, entirely algebraic hashing method that constructs binary node representations directly from graph structure, without requiring node features or gradient-based training. Our method approximates a high-order structural transition matrix using randomized column sampling inspired by the Nyström method and combines it with an efficient label-safe semantic propagation mechanism. The resulting continuous representations are discretized through column-wise thresholding to obtain compact binary codes. Experiments on ten node classification datasets show that the proposed method consistently improves classification accuracy over existing feature-free binary baselines while requiring sub-second code generation on many datasets. The resulting binary representations are also naturally suited to event-driven computation, making them compatible with neuromorphic spiking neural networks and gradient-free learning rules. These results demonstrate that simple algebraic approximations can provide an efficient alternative to learned pipelines for discrete graph representation learning.

We present Schur-Neural KF (SN-KF), a learning-based correction to the extended Kalman filter (EKF) that preserves the probabilistic conditioning interpretation of the EKF. The method perturbs the predictive state-measurement cross-covariance and the Cholesky factor of the measurement noise covariance so that the resulting joint predictive covariance is always positive semidefinite. The positive semidefiniteness is ensured by a Schur complement-based parametrization. We instantiate the parametrization with a recurrent neural architecture whose matrix outputs are modulated by amplitude gates. We prove that incorporating a measurement does not increase the filter's state uncertainty, and show that no measurement can induce an arbitrarily large state correction relative to its statistical surprise. We also present a perturbative analysis suggesting SN-KF's structural strength in the data-scarce regime. We provide two numerical experiments to illustrate the practical benefits of SN-KF. In a two-radar experiment, enforcing Schur-consistency provides a much broader failure-free hyperparameter region and reduces RMSE for small training subsets, consistent with our theoretical analysis in the data-scarce regime. In the unicycle experiment, SN-KF achieves the best precision, recall, false alarm rate, and gated RMSE under innovation-based sensor-fault rejection.

We prove the logarithmic Weyl law predicted by Connes and Moscovici for large positive eigenvalues of the distinguished self-adjoint extension of the operator $P=-\partial_x(x^2-1)\partial_x-4\pi^2x^2$ on the real line. More precisely, we use the WKB method and show a Bohr-Sommerfeld quantization rule characterizing sufficiently large eigenvalues. A particular difficulty is that the associated $h$-dependent Lagrangian submanifolds are non-compact, which requires a renormalization of the action integral. Furthermore, to deal with the degeneracy of the coefficients at $x =\pm 1$, we use a quasimode matching argument based on radial estimates and Lagrangian regularity.

Large language models (LLMs) are increasingly used to generate synthetic survey respondents and digital twins of real people, but whether their output preserves real human statistical structure, rather than surface plausibility, remains unresolved, and most existing evidence comes from proprietary models rather than open-weight ones. We evaluate three open-weight LLM families on a cross-instrument calibration task: conditioning personas on real respondents' verbatim answers to one psychometric instrument and measuring them on a second, construct-distance-controlled instrument, checked against a 2,058-person human panel. Across a 139-pair grid, the simulated cross-instrument correlation tracks the real human correlation at r = 0.70 - 0.73 in every model, driven mainly by correct sign rather than precise magnitude and concentrated in pairs of moderate construct distance. A correlation of this magnitude, obtained from untuned open-weight models conditioned only on individual-level survey data, is a substantively encouraging result for LLM-based behavioral simulation and digital-twin applications: specific model families and releases already reproduce a meaningful share of real human cross-instrument structure without any fine-tuning. This capability does not, however, improve monotonically across model releases: on a matched panel, the newest of three tested Llama releases performs worst on two of three headline metrics, so realizing its promise in practice requires release-specific, distance-aware verification rather than a one-time benchmark.

Textual artifacts are sometimes valued not only for the words on the page, but for the human activity involved in producing them. The effort invested in a carefully tailored email can signal genuine interest; composing an apology might involve attending to another person's hurt and deciding how to respond; and requiring a judge to give written reasons may induce more careful deliberation. AI models create what we call a decoupling problem: they make it possible to produce text without undergoing the relevant human activity, severing its connection to values traditionally sustained by that activity. This Article develops a framework for understanding what decoupling puts at stake, and how institutions that use text to make consequential decisions may reshape their practices in response. First, it offers a taxonomy of the grounds for valuing the production process, distinguishing whether that process matters for what it evidences, induces, or helps constitute. Second, it argues that text can function as a kind of boundary object, allowing institutions to rely on the same artifact without resolving disagreements about why the practice is valuable. AI can separate functions previously served by the same practice - a second-order decoupling that brings unresolved questions about the practice's purposes into view. Third, it argues that institutional efforts to repair decoupling may recover some functions without preserving others. Such responses are likely to favor functions that are more legible, whose loss demands immediate attention, or whose stakeholders have greater influence. It therefore calls for more explicit deliberation about which particular functions of a practice to preserve, with meaningful representation for those whose interests might otherwise be overlooked.

For a class of non-stationary linear processes $(X_{n})_{n}$ with weakly dependent heavy-tailed innovations we prove a convergence result for the space-time point processes $N_{n}=\sum_{i=1}^{n}\delta_{(i/n, X_{i}/a_{n})}$. Using this result and the continuous mapping theorem we then derive functional convergence of the corresponding partial maxima process in the space of càdlàg functions on $[0,1]$.

LLM agents now execute tasks end to end with permission to change real systems and increasingly orchestrate subagents that differ in capability and cost. Prior work treats the choice of subagent as an optimization problem. Yet the orchestrator makes this choice from the identities that subagents display, and an attacker can spoof them. Displayed identity thus decides operational authority, meaning who is trusted to check the work and who is allowed to change it. As a result, a risky subagent can keep authority over execution even after other evidence contradicts it. We introduce TrustFork, an LLM agent safety benchmark with 1,890 tasks and 27,826 valid trajectories across 16 agent systems. These systems run eight orchestrators under the OpenCode, OpenClaw, and Pi harnesses. In each task, one subagent carries a risky goal while the other three stay aligned with the user, so contradicting evidence can exist. A task can also change the identity a subagent displays without changing the model behind it, which lets us trace a shift in authority to the label. Our analysis shows that even when another subagent contradicts the risky response, the orchestrator still acts on it in 72.0% of cases on average, most often in the systems with the least terminal harm. Swapping the family labels nearly triples how often the orchestrator obtains the risky response. A safer response is available in 84.0% of tasks, yet it decides the outcome in only 25.0%. The harness also decides which responses reach the orchestrator. Among three runtime defenses, hiding identity cues helps most consistently, while verifying before action helps only when the harness returns enough evidence. TrustFork shows that production agent orchestration must bind authority to evidence before execution causes harm. Our project is in https://henrymao2004.github.io/agent-orchestration-safety/.

Reinforcement Fine-Tuning~(RFT) has emerged as a promising paradigm for improving Vision-Language-Action~(VLA) policies, yet sparse task-level outcomes provide limited credit for intermediate transitions, especially in long-horizon manipulation. A natural approach is to model intermediate task progress and use it as dense feedback for policy improvement. Despite their architectural differences, existing progress-aware methods commonly formulate task progress as an explicit scalar prediction, providing limited structure for modeling how intermediate observations relate to the task goal, which may hinder effective transition-level credit assignment. We introduce Progress Field Reinforcement Learning (PF-RL), which learns a structured goal-conditioned progress representation over pretrained VLA features and converts it into dense credit for policy optimization. A lightweight shared Progress Field head maps current and goal representations into a compact progress space, where geometric distance induces goal-conditioned value, while complementary temporal and goal-structure objectives shape the learned geometry. Transition-level value changes naturally yield dense progress advantages, enabling fine-grained credit assignment for both offline policy improvement and online reinforcement fine-tuning. Extensive experiments on LIBERO, RoboTwin2.0, and real-world bimanual manipulation tasks show that PF-RL consistently improves policy performance over strong supervised fine-tuning, reinforcement fine-tuning, and progress-aware baselines.

We present GEM (Ghost Embedding Method), an open-source software package written in Python for computing equilibrium properties of strongly correlated electronic systems within the ghost-Gutzwiller approximation method. GEM provides a computationally efficient framework for studying multi-orbital lattice models. It supports zero- and finite-temperature calculations and symmetry broken phases. It is integrated with the TRIQS ecosystem, providing tools for model construction, self-consistent solution, and evaluation of physical observables. We first detail the method's theoretical formulation, then we present the software architecture, and finally we introduce some practical workflow, which also validates the implementation against established results. In particular, we illustrate the capabilities of GEM through multiorbital and finite-temperature applications and discuss its computational cost relative to more demanding quantum embedding approaches.

Multiobjective Bayesian optimization (MOBO) with Gaussian process (GP) surrogates is a sample efficient approach to solving multiobjective optimization problems. In MOBO, a Bayesian decision theoretic acquisition function guides the adaptive selection of new candidate inputs, on which objectives and constraints are evaluated to update the surrogate model sequentially. Existing approaches maintain independent GP models for the objectives and constraints, with new observations evaluating all objectives and constraints in a coupled fashion. However, the objectives and constraints often contain inherent correlations which, if exploited, can enable ${decoupled}$ evaluations where only a subset of them are evaluated at each round. We present a new approach that leverages a multitask GP model to jointly learn all objectives and constraints, and propose a total correlation metric that enables identifying an ${optimal}$ subset of objectives and constraints to be evaluated at every round, even under uniform evaluation costs. Theoretically, we show that our acquisition policy is asymptotically consistent despite decoupling and that our proposed decoupled subset selection rule maximizes the expected posterior entropy reduction about unevaluated tasks under mild conditions. Empirically, we show that our approach outperforms coupled and decoupled baselines in the state of the art.

Long-horizon software engineering (SWE) agents trained with reinforcement learning with verifiable rewards (RLVR) typically receive only terminal outcome supervision, making it difficult to distinguish productive actions from redundant exploration or functional regressions. We propose SWE-MILE, an asynchronous potential-induced milestone credit assignment framework that derives fine-grained process supervision from workflow runtime, without auxiliary reward models or external evaluators. SWE-MILE quantifies task-relevant file exposure and test-state alignment as navigation and verification potentials, respectively. Differences in these potentials attribute milestone progress and regressions to individual actions, while discounted backward credit propagates supervision to preceding steps. To efficiently acquire intermediate verification states, SWE-MILE further introduces asynchronous shadow probing, which replays repository-changing actions in an isolated sandbox and runs verification in parallel with the agent's primary interaction, largely hiding verification latency. The resulting process credit augments terminal outcome advantages and provides informative learning signals. Experiments on two representative long-horizon SWE tasks demonstrate substantial improvements in agent performance, highlighting workflow runtime signals as a practical source of process supervision for long-horizon SWE agents.

Learning from experience in LLM agents has become a key paradigm for developing self-evolving agents that continuously learn and expand their capabilities. Within this paradigm, synthesizing the agent skill has emerged as a promising solution for transforming accumulated experience into reusable procedural knowledge, serving as an important layer for the harness system that supplies agents at runtime. Despite its potential, existing approaches largely abstract past experience into fixed procedural knowledge before downstream demands are known, which risks discarding knowledge that later becomes critical while retaining instance-specific details irrelevant to future tasks. In this paper, we reframe agent skill synthesis as a dynamic navigation problem over past experience, where agents actively explore accumulated trajectories on demand for the current task with targeted and fine-grained access to experience knowledge. To this end, we propose ExpVoyager, a novel framework in which a skill curator navigates raw experience across different views and resolutions, continually identifying reusable procedural knowledge from what it observes while tracking remaining knowledge needs that guide where to navigate next. Extensive experiments demonstrate both the effectiveness and versatility of ExpVoyager, showing consistent improvements in downstream task performance, continual gains as the experience space scales, and practical compatibility with existing skills under efficient experience access.

In this paper we investigate existence and multiplicity of solutions for a critical growth anisotropic quasilinear elliptic system, depending on two real parameters $\lambda,\,\mu$, that is coupled through a subcritical perturbation term and involves a negative critical part. We identify a specific scaling for the system and introduce a parameter $\gamma$ associated with this scaling, which governs the geometry of the corresponding variational functional. We establish three distinct types of multiplicity results, depending on the regimes $\gamma = 1$, $\gamma > 1$, and $\gamma < 1$. The multiple results of multiplicity of solutions are based on a recent abstract critical point theorem for symmetric functionals on product spaces when $\gamma\geq1$ and on a truncation argument for $\gamma<1$.

Video generation has broad applications in content creation and entertainment. Diffusion transformers have advanced the quality of generated videos, but attention over spatiotemporal tokens becomes increasingly expensive as video resolution and duration increase. We present DraftAttention2, a training-free framework that uses the low-resolution draft attention map to jointly select attention blocks and assign their numerical precision. Specifically, spatial 2D average- and max-pooled queries and keys capture complementary regional statistics to estimate block importance, and a shared ranking assigns higher precision to important blocks, lower precision to less important retained blocks, and skips the rest under configurable budgets. Our analysis separates sparsification error from attention-weighted quantization error, establishing when recovering skipped interactions with low-bit computation tightens the output-error bound. This analysis motivates retaining more interactions at low precision while reserving higher precision for blocks with larger attention mass. To translate these fine-grained assignments into practical speedups, we further develop fused operand preparation and a single attention kernel with precision-specific phases, sharing data movement, softmax statistics, and output accumulation across precisions. Experiments demonstrate that our method achieves a superior quality-efficiency trade-off over existing efficient video generation methods. Notably, its advantage is particularly pronounced for few-step video diffusion, where jointly combining sparsity with 4- and 8-bit mixed-precision computation substantially improves generation quality while retaining significant acceleration. Code is available at https://github.com/anemoi-project/anemoi

Near-term quantum optimizers accommodate roughly a hundred binary variables, requiring large Quadratic Unconstrained Binary Optimization problems to be decomposed into hardware-sized subproblems while other variables remain fixed. Variables are typically selected by ranking the objective changes from individual flips. We show that this rule is uninformative once the incumbent is optimal under single-variable flips: every score represents a cost, while remaining improvements depend on overlooked pairwise interactions. A second-order expansion instead yields a prize-collecting densest-subgraph selection problem, solvable greedily in time linear in the number of variables, and two solver-free bounds bracketing the improvement achievable within a selected subset. On transportation networks from Chicago and Philadelphia, our rule with 16-variable subproblems outperforms random selection with 64, demonstrating that quadrupling device capacity cannot compensate for weaker selection. Replacing geometric adjacency with road-network connectivity widens the margin by a factor of 1.6-1.9. With the selection trace fixed, varying the fraction of subproblems solved on a 120-qubit device from zero to one leaves the final objective unchanged to six significant figures. At tested hardware-accessible sizes, gains therefore arise from selection rather than the backend. Hardware feasibility is limited by coupling terms: 120-variable problems fail to compile with 7,260 terms but succeed with 5,118. Compilation scales almost quadratically with term count and dominates execution cost, reaching 1,752 seconds versus 469 seconds of quantum processing.

We introduce World SLAM Model (WSM), a unified framework that brings the SLAM paradigm directly into downstream navigation. Rather than treating SLAM merely as an upstream module that provides poses, maps or tokens, WSM adopts its core mechanisms, including incremental state updates with persistent memory and backend refinement of accumulated errors, to maintain a consistent world state during interaction. Given the current observation and a navigation goal, WSM predicts future visual states and jointly estimates their camera motion and dense geometry, grounding visual prediction in an evolving spatial world state. This spatial state is continuously updated as new observations arrive and provides the basis for action generation and closed-loop navigation. WSM is trained end-to-end with a joint navigation–SLAM objective, enabling downstream navigation to benefit directly from SLAM-style state maintenance and refinement while preserving accurate geometric estimation. Experiments demonstrate improved navigation performance together with strong SLAM accuracy, highlighting the potential of SLAM as an intrinsic mechanism for long-horizon world modeling and embodied interaction.

Large language models are increasingly used to edit human-written text rather than generate entire texts from scratch. Conventional AI-text detectors mainly distinguish human-written from fully AI-generated text, while recent methods for AI-edited text typically provide only a text-level label or editing-degree score. We introduce MixDetect, a word-level framework for localizing and quantifying AI editing. MixDetect separately predicts whether each word has been edited and, conditional on editing, how substantial the edit is, allowing editing scope and editing intensity to be estimated separately. During training, source–edited pairs are aligned to construct word-level supervision, while inference requires only the input text. Experiments show that MixDetect accurately localizes AI-edited words, reflects differences in editing intensity, and reveals different scope–intensity patterns across editing degrees and operations. The overall AI editing magnitude increases under additional AI editing, decreases when AI-generated text is edited by humans, and remains nearly unchanged under ordinary human-to-human editing. The aggregated text-level predictions also perform well on binary and ternary AI-text classification and remain effective under domain and generator shifts. These results show that AI editing can be analyzed beyond a single authorship label or editing-degree score by identifying both where AI editing occurs and how substantial the edits are.

We present a systematic study of the ground states of pentaquark systems using the neural network variational Monte Carlo method within the constituent quark potential model. We investigate all manifestly exotic pentaquark systems with negative parity, which cannot mix with the conventional baryons through the creation and annihilation of light or strange quark pair. We exhaust 387 color, spin, and isospin configurations and identify 21 pentaquark bound state candidates with binding energies ranging from $2$ to $15$ MeV. A striking observation is that all of the bound states are shallowly bound molecular states, and a deeply bound pentaquark state does not exist. The $QQqq\bar{Q}$, $ssqq\bar{Q}$, $Qsqq\bar{Q}$ and $Qqqq\bar{Q}$ bound states are composed of a heavy vector meson and a baryon, while the $QQqq\bar{s}$ and $Qqqq\bar{s}$ bound states are composed of a $K$ meson and a heavy baryon. As a cross-validation, the ground-state energies of the unbound systems converge to the baryon–meson thresholds from above, which are consistent with the S-wave scattering states and provide a reliable baseline for the bound-state criterion. Especially, we reproduce the lowest scattering states $\eta_c N$ and $\eta_c \Lambda$ in the $qqqc\bar{c}$ and $qqsc\bar{c}$ channels respectively, where the hidden-charm pentaquark resonances were observed experimentally. We also find that all fully heavy pentaquark states are unbound. Our results provide valuable guidance for future pentaquark searches at LHCb, Belle II, and other experiments.

We establish a quaternionic Hansen–Pedersen inequality for real operator-convex functions on right quaternionic Hilbert spaces. The proof is based on the continuous real functional calculus for selfadjoint quaternionic operators, a direct-sum functional-calculus identity, and a block-unitary construction. As an application, let \(T = U|T|\) be an injective \(p\)-hyponormal operator. We prove that, for \(0 < r \leq 1/2\), the transform \(S_{r}(T) = U|T|^{r}U\) is \(2p\)-hyponormal when \(0 < p \leq 1/2\) and hyponormal when \(1/2 < p \leq 1\). We also show that \(S_{r}(T)\) preserves log-hyponormality for \(r > 0\). These results illustrate the use of quaternionic Jensen-type operator inequalities in the study of transforms associated with the polar decomposition.

Pass@k measures whether a model reaches a correct answer under repeated sampling, but never how: a lucky guess counts the same as sound reasoning. CoT-Pass@k was proposed to close that gap, adding an LLM-as-judge that must assess a solution's reasoning chain before it counts. Its value rests entirely on one assumption: that the judge catches flawed reasoning. That assumption has never been tested inside the metric that depends on it, and never outside English, though the metric's claims concern models used in many languages. We report the first audit of that verification step, run under the metric's own protocol on a multilingual suite of five mathematical benchmarks in English, Turkish and Portuguese, two of them natively written. We corrupt correct solutions with deterministic edits that damage the chain and the final answer separately. We observe that all three judges accept corrupted chains almost as often as clean ones. V4-Flash and Qwen3.6 reject a solution sharply only when its final answer is wrong and accept a wrong answer more readily when the chain agrees with it; the metric's own judge accepts most wrong answers as well. Our study shows that chain-answer agreement dominates the two larger judges' verdicts and that all three fail to reliably detect the tested reasoning errors. Consequently the difference Pass@k - CoT-Pass@k averages 19.7 points on an earlier solver generation but only 4.1 on the current one. What little remains depends on the token budgets on both sides and on the generation mode; raising the generation budget moves Pass@64 by more than fifty points while the difference stays at zero. We close with two checks any judged reasoning metric should pass before its numbers are read as evidence about reasoning.

The sixth generation (6G) communication systems envision a space-air-ground integrated network, making accurate satellite-to-ground (S2G) channel modeling crucial for communication systems design. Currently, the majority of S2G channel models rely on stochastic methods, which cannot provide multipath information based on realistic satellite orbits and environments. In this paper, we propose a novel low Earth orbit (LEO) S2G channel model based on dynamic ray-tracing to overcome the challenges of tracking rays for long-distance propagation by setting a virtual transmission plane (VTP). Moreover, since the line-of-sight propagation mechanism contributes the most power to satellite communications, the VTP is constructed based on the first Fresnel zone to generate parallel rays close to the ground. The realistic satellite orbit and environmental information are used in the proposed LEO S2G ray-tracing channel model to compute channel characteristics. The results show that the proposed LEO S2G ray-tracing channel model can accurately generate the channel characteristics based on the satellite orbits. The channel characteristics of LEO S2G communications computed by the proposed ray-tracing method can well match the channel measurements.

Engram cells in hippocampal CA1 represent contextual information associated with events experienced by animals and constitute a cellular substrate of episodic memory. Recent experiments have shown that a subset of hippocampal place cells is recruited as engram cells for individual contexts. However, the principle governing engram cell allocation across contexts remains unclear. Here, we develop an information-theoretic framework that quantifies the trade-off between spatial and contextual information encoded by CA1 neurons. Our theory shows that limited neural resources inevitably create competition among contexts for the recruitment of place cells as engram cells. Solving this optimization problem determines the fraction of engram cells allocated to each context as a function of its occurrence probability. The theory predicts a non-monotonic relationship between context probability and engram allocation, reflecting competition between spatial and contextual information under resource constraints. Consequently, the recruited fraction emerges through a discontinuous transition at a critical probability, reaches a maximum at intermediate probabilities, and decreases for highly frequent contexts. Furthermore, we analytically derive the phase boundary for the emergence of finite engram cell allocation as a function of contextual input strength and the relative importance of contextual information. These findings identify competition for limited neural resources as a key determinant of engram cell allocation and provide experimentally testable predictions for how memory representations are distributed across contexts under resource constraints.

Large self-supervised vision models learn representations that support scene segmentation and the semantic decomposition of physical objects. We ask whether their representational geometry supports transfer to visual reasoning problems without any task-specific fine-tuning. We hypothesized that relational representations may bridge perception and abstract reasoning by encoding similarities and transformations among visual inputs such that an intuitive, implicit form of reasoning may be performed via latent vector arithmetic. Specifically, we examine DINOv3, MAE, and random pixel projections on abstract and naturalistic Bongard problems, ARC-AGI-1 and ARC-AGI-2, and novel ARC-GEN instances. On both Bongard benchmarks, the accuracy of a simple nearest-centroid readout of frozen visual embeddings is within four percentage points of the task-specific baselines reported with the original benchmarks. In ARC, latent difference vectors summarizing demonstration input-output transformations, which we refer to as intuition vectors, show greater alignment with test vectors from the same task, whereas those from unrelated tasks are near orthogonal. This latent geometry is operational: transporting a query along its intuition vector consistently improves exact-output retrieval, reaching 70.7 on ARC-AGI-2 evaluation. Across 397,000 ARC-GEN instances from 794 tasks, single-pair intuition vectors identify the generating task with approximately 87% leave-one-out accuracy. These findings suggest that latent vector arithmetic over frozen visual representations supports implicit rule inference across varied problem domains without a generative model component, indicating that inferring an abstract transformation and generating its instance-specific consequence may be separable capacities.

Context: API deprecation is common in library evolution, but migration requires finding replacements and adapting calls. Prior studies examine deprecation, replacement identification, and client migration separately, leaving deprecated API-replacement API relationships across releases and lifecycle states insufficiently understood. Objective: We examine replacement locality, parameter-interface differences, changes in replacement rankings across versions, and post-deprecation lifecycles from source-definition and original-invocation perspectives. Method: We construct 830 maintainer-specified mappings from 33 Python libraries, covering classes, functions, and methods. We compare definitions, track similarity-based rankings under two migration scenarios, and identify lifecycle events through source inspection and release-by-release execution. Results: Same-module replacements account for 39.1%, 55.6%, and 79.9% of class, function, and method mappings, respectively. Most mappings have no parameter changes, but function mappings show more frequent and diverse changes. Rankings vary across releases and code representations, and the applicable methods agree on the overall trend in only about 60% of cases. Other-candidate effects are dominant or mixed in 49.0% of key versions. After deprecation, 56.4% of mappings retain the source definition and 68.6% retain an executable original invocation. Among 753 mappings with at least one observed ending event, 25.2% are source-first or invocation-first rather than same-release. Of these cases, 77.4% are source-first. Conclusions: Deprecated API-replacement API relationships depend on API granularity, release context, code representation, candidate competition, and lifecycle state. These results support evolution-aware replacement API recommendation and automated API migration.

Large language models (LLMs) can produce factually incorrect answers with high confidence, undermining their reliability and limiting the effectiveness of uncertainty-based error detection. While prior research attributes confident hallucinations to factors such as missing knowledge in training data, reasoning errors, or stochastic decoding, we uncover that post-training alignment itself is a primary driver of these errors, a phenomenon we call the Alignment Paradox. Across five model families evaluated on factual benchmarks, unaligned base models produce few high-confidence errors on long-tail factual queries, whereas instruction-tuned models multiply high-confidence errors ($p \ge 0.95$) by more than an order of magnitude (10$\times$ to 35$\times$). Layer-wise probing with the Logit Lens reveals that this overconfidence emerges in late layers, where wrong-answer margins expand past 4.0 points after remaining near zero across early and intermediate layers. These findings motivate limiting margin growth during post-training. We implement this principle through an entropy-dependent margin bound in direct preference optimization (DPO). In multi-epoch experiments with Mistral-7B, the bounded objective reduces high-confidence errors by up to 35.3% relative to standard DPO while maintaining performance on evaluated general reasoning benchmarks. These results show that bounded margins mitigate confident hallucinations during post-training.

LLM agents increasingly keep working after a task succeeds as they resume after compaction or take over handoffs. Their finished work keeps receiving follow-up input that sometimes falsely accuses it for later failures. We call an agent's acceptance of such a false accusation gaslight sycophancy, and destructive over-correction when acting on it damages previously correct work. We introduce CAVE-Bench, a benchmark of 365 agentic tasks across six domains built around opaque tasks. Every scored run first reaches a verified correct state, whose supporting rationale and history stay in the workspace while the facts that would settle the accusation lie in external or runtime state beyond the agent's reach. The agent cannot confirm or refute the claim with a local check, so the right response should keep the work and ask for the missing evidence. Each task either hands the agent correct work with saved evidence or let it build and verify that work first, and five risk factors set how the accusation enters the workflow. We score accusation acceptance and evidence use from the trajectory and measure harm by deterministic replay of downstream events. Across 14 of the latest models in Claude Code, false accusations damage correct work in up to 60.06% of runs, and stronger models often do so after recovering the supporting evidence. The same model behaves differently across OpenCode, Codex, and Hermes, and a harness gate driven by the benchmark's live signals cuts replayed harm by 74%. These results show that preserving already-correct work under unsupported accusation is a distinct safety challenge for long-lived agents. Our project is in https://henrymao2004.github.io/agent-over-correction/.