Science & Mathematics

We test whether Javo, a sales agent fine-tuned on 4.6 million consented, de-identified commercial conversations, sells better than a frontier general-purpose model and better than human sales associates. The evaluation has four parts. Study 1 scores the agent on two public sales benchmarks, SalesLLM and SmartSalesBench, against 10 frontier and open-weight models. Study 2 is a randomized controlled trial on 100 live websites in 14 industries (e-commerce, travel and hospitality, and high-consideration services), run from 1 May to 30 September 2026, in which 14.8 million visitors were randomized to no agent, Javo, or a frontier-model agent with an identical interface, catalog tools and guardrails. Study 3 isolates the effect of the model by comparing those two agent arms. Study 4 compares Javo with each site's own human associates on 20 sites with staffed live chat. Relative to no agent, Javo changed revenue per visitor by +10.0% (95% CI +5.1 to +14.9). Among engaged visitors, it changed conversion by +21.6% relative to the frontier-model agent (95% CI +16.9 to +26.3). Against human associates, its engaged conversion was 11.26% versus 11.20% (non-inferior, margin -7%, one-sided p = 0.004; no significant superiority), at 1.8% of the cost per conversation. Return rates, unsupported product claims and complaint rates met the prespecified safety guardrails.

How well can transformer-based Large Language Models (LLMs) retain and recall memories in long contexts of dialogue? LLMs are known to forget long-range factual information in their context, yet the field lacks a comprehensive benchmark to quantitatively assess these shortcomings. We present DiaLong, a long-context dialogue dataset and accompanying memory task that standardizes evaluation in this space. It consists of 100 unique conversations created by humans and then augmented with LLMs, extending beyond lengths found in existing datasets. Our selected memory task is fact recall: classify 40 candidate facts about a dialogue as true or false. This task directly probes the model's ability to actively retain and retrieve information, serving as a critical measure of memory within extended conversational contexts. In testing both the latest LLMs and proposed memory enhancements on DiaLong, we demonstrate that more work needs to be done to enable LLMs with better memory performance. By enhancing memory capabilities, LLMs can improve in maintaining coherence, relevance, and human-like interaction over extended conversation. Improving these capabilities will help LLMs fulfill diverse user needs and expectations in various conversational scenarios.

We present experimental evidence suggesting that information about whether a model's reasoning is on the right track can be linearly decodable without defining a causally useful direction for steering. To study this, we introduce ChainSearch, a controlled arithmetic reasoning task that extends Countdown with explicit search and backtracking. We train a 3-million-parameter model to solve this task. Its compact size allows us to study internal model correctness directions in detail using linear probes, mean activation differences, and direct optimisation, as part of a rigorous search for internal reasoning-correctness directions. We find that while the model contains statistically significant internal directions of reasoning correctness, none of the resulting directions consistently improves reasoning under causal intervention. This calls into question whether the linear representation hypothesis is sufficient for identifying causally useful directions in the context of reasoning. Our code is made available together with this work.

The problem of Permutation Pattern Matching (PPM) asks, given a permutation $π:[n]\to[n]$, viewed as the point set $P=\{(i,π(i)):i\in[n]\}$, and a pattern $σ$ of small length $k$, whether $P$ contains an order-isomorphic copy of $σ$. The PPM problem and its (exact and approximate) counting variants have been studied extensively in the literature. We consider a preprocessing-query version: after preprocessing $P$, one must determine whether a query $P'\subseteq P$ contains $σ$, where $P'$ is obtained by intersecting $P$ with a geometric shape (e.g., an axis-aligned rectangle or $k$ such rectangles). This naturally generalizes the classical range query problem in computational geometry, which corresponds to the case $|σ|=1$. We extend the study of this problem from patterns of length $2$ to patterns of length $3$ (and in some cases beyond), and prove new fine-grained equivalences, separations, and applications: (1) Single rectangle: For all patterns of length $3$, detection can be done with $\tilde{O}(n)$ preprocessing and $\tilde{O}(1)$ query time. The proof goes through a new tool of independent interest: reductions from rectangle queries to a simpler family of slab queries. (2) Multiple rectangles: For $σ=132$, detection is equivalent to the online set disjointness problem; we also show connections to online $k$-set disjointness. To our knowledge, this is the first equivalence between a PPM-type problem and online set disjointness, a problem playing a fundamental role in fine-grained complexity. (3) Counting: As applications, we obtain an $O(n^{1.725})$ time algorithm for exact counting of $5$-patterns based on the rectangle-to-slab reduction, improving upon [Beniamini and Lavee, ICALP 2025], and an $\tilde{O}(n^{7/4})$ approximate counting algorithm for $6$-patterns, the first to break the $O(n^2)$ barrier for $6$-patterns.

We study the last iterate of stochastic gradient descent (SGD) with a constant stepsize on smooth convex objectives in the low-noise and interpolation regimes, where the variance $σ_\star^2$ of the stochastic gradients at the optimum is small or zero. We establish that after $T$ steps of SGD on $β$-smooth convex losses with any constant stepsize $0

Vision-language (VL) models are typically built by fine-tuning a language model to attend to visual features in service of a Visual Question-Answering objective. We ask whether this alignment changes how language models represent text---especially on tasks where visualization would help, like geometry questions posed in words. We distinguish weak multimodality (the model can answer questions about an input image) from strong multimodality (the model's text representations themselves become vision-aligned, improving performance on text-only tasks with latent visual structure). To test for strong multimodality, we construct a task of shape-counting problems with paired equivalent text and image specifications. This task is trivial when the image is provided, but requires visualization when specified in text as line coordinates. We compare instruction-tuned Qwen-2.5 and Mistral models against their vision-aligned counterparts on the text variant of the task. We find that the multimodal models do no better in this setting, despite strong performance on the image version of the task. Probing experiments corroborate this: vision-aligned decoder embeddings carry no more visual structure than their text-only counterparts. Finally, we provide an optimization objective that distills in-context visual embeddings into text embeddings. This alignment induces gains on the text variant of the task without explicit task supervision, even on samples more complex than seen during alignment. This suggests that VQA-style alignment leaves available cross-modal transfer untapped.

We present SZED's submission to ArGuard 2026 Track A: Arabic hateful meme detection. The task has two parts: binary classification (A1) and fine-grained multi-label tagging (A2). For A1, we fuse MARBERTv2 and CLIP ViT-L/14 embeddings, comparing concatenation, self-attention, and ensemble heads. For A2, we use a multi-task label-attention architecture with per-class query vectors and an auxiliary head for cascade calibration. We reach macro-F1 = 0.7597 on A1 and 0.3142 on A2 (averaged over 10 active classes). We report reproducibility across runs, per-class error patterns, and the effect of cascade calibration on the long-tailed A2 label space.

Decision models answer typed questions, such as multiple-choice or yes/no questions, with a probability for each option. We add test-time search to frozen decision models, with no training: we show the model a complete proposed answer, read off its probability of saying "Yes" to the proposal's correctness, and search for answers that it rates higher. The search is done either by editing one answer at a time or by joint optimization, which changes several answers at once. Search raises accuracy for every model and test set we tried, and the gains are largest for exact correctness on logic puzzles, where every answer in a task must be right. On 128 such puzzles, it takes Qwen3.8-27B from 2 to 30 exactly correct puzzles ($15\times$) and its per-blank accuracy from 27.8% to 48.5%. On a shared set of 47 puzzles, searching over each model's own Yes/No score raises the per-blank accuracy of StartLux Instant, Perplexity Decider, Clef, and JEV, which we reach only through its public API, by 4.0 to 13.0 percentage points, and raises the number of puzzles they get exactly correct, out of 47 and averaged over three search seeds, from 10 to 35, 9 to 26, 2 to 11, and 9 to 31.7 respectively.

Trajectory optimization methods, from sampling-based zero-order to gradient-based first- and second-order approaches, require many evaluations of dynamical system functions, i.e., functions derived from a robot's dynamical model, such as forward kinematics and their Jacobians and Hessians. These evaluations are largely independent, e.g., across sampled trajectories or time steps, and can thus be evaluated in parallel as a batch. When used to make decisions online, these methods require low latency, i.e., each batch must complete in as little wall-clock time as possible. Existing frameworks utilize modern GPU hardware primarily to parallelize across batch instances, thereby optimizing throughput but not latency. We present CusADi2, a software library that reduces the batch evaluation latency by additionally parallelizing computation within each batch instance. We benchmark forward kinematics, Jacobians, and Hessians for all 67~robots in the MuJoCo Menagerie, including arms, hands, quadrupeds, humanoids, and biomechanical models, against MuJoCo XLA, MuJoCo Warp, frax, CusADi, and CasADi. CusADi2 achieves lower latency than all baselines at batch sizes of 64 to 4096, while matching them at larger batch sizes. We also demonstrate the performance advantage of CusADi2 in a sampling-based control application. All code is open-source at: https://github.com/johnviljoen/cusadi2

We present NEP-kappa 2.0, a unified framework that integrates density-functional theory, material-specific machine-learning interatomic potentials, and pretrained force models for advanced phonon calculations. A common workflow enables harmonic lattice dynamics, quasi-harmonic and stochastic self-consistent phonon treatments, three- and four-phonon transport, Wigner population--coherence transport, and time-dependent phonon dynamics. Applications to 3C-SiC, BAs, and Cs$_2$AgInCl$_6$ demonstrate consistent force-model comparisons, finite-temperature renormalization, higher-order scattering, and intermode coherence. Beyond steady-state transport, mode-resolved simulations of selectively excited 3C-SiC reveal that half of the injected optical-phonon energy reaches acoustic modes within approximately 3.75 ps, primarily through one-optical--two-acoustic scattering. Acoustic branches exhibit response times differing by a factor of five, with competing scattering channels producing a transient energy overshoot. Channel-resolved entropy production and persistent nonthermal occupations distinguish interbranch energy redistribution from intrabranch thermalization. The framework connects equilibrium phonon transport with microscopic nonequilibrium energy dynamics.

Test-time scaling of diffusion models with sequential Monte Carlo (SMC) has become a promising strategy for sampling under user-specified rewards, such as human preference scores. However, the standard SMC framework typically relies on stochastic proposals and therefore does not directly transfer to flow matching models, in which inference is governed by deterministic ODEs. In this work, we introduce noise-rejuvenated SMC, an inference-time scaling method tailored for text-to-image flow matching models. Our key idea is to perturb each resampled particle by changing its source noise and constructing a new intermediate state using its current endpoint estimate. This moves the particle to another state that remains representable under the flow-matching interpolation form, enabling particle exploration without converting the sampler to an SDE. We further introduce a look-ahead mechanism to refine resampled particles before weighting. Experiments on Stable Diffusion 3, FLUX.1-dev, and Z-Image demonstrate consistent improvements in reward-guided generation, while ablation studies validate the contributions of noise rejuvenation and look-ahead refinement. Code is available at https://github.com/shermanlian/NR-SMC.

Java applications reach large language models (LLMs) through an integration framework, and rely on it to handle a failing provider: to retry what can be retried and nothing else, to wait as long as the provider asks, never to present a truncated answer as complete, and to record each failure in the application's metrics. We state these expectations as a six-clause reliability contract and measure Spring AI 2.0.1 and LangChain4j 1.21.0 against it with a scripted OpenAI-compatible provider: fifteen failure scenarios, synchronous and streaming, three repetitions. LangChain4j meets it in 47 of 77 applicable cases with its defaults and in 51 of 83 when configured explicitly, Spring AI in 72 of 77 and 76 of 83; both deliver a stream that ended without its terminator as a complete answer. We then place one implementation of the contract beneath both frameworks, at the HTTP boundary - an OkHttp interceptor for Spring AI, an HttpClient for LangChain4j, and a java.net.http transport for code without a framework - sharing a 725-line core that depends only on the JDK and the Micrometer Observation API. With it, all three clients behave identically and meet the contract in 90 of 91 cases; the remaining case is a policy trade-off between retrying a hung provider and waiting for a slow first token. Against a local model server the layer rejected none of 700 calls, and it adds a median of 32-217 microseconds to a call.

Spatiotemporal forecasting of 3D anatomical changes is a critical generative vision challenge for modeling Alzheimer's Disease (AD) progression. While brain MRI forecasting from single-modality data has advanced, other key complementary multimodal biomarkers remain underutilized. In this work, we propose the Multimodal-Conditioned Latent Diffusion Model (MC-LDM), a novel framework for patient-specific brain MRI synthesis. MC-LDM leverages multimodal biomarkers---MRI, tau-PET, Clinical Dementia Rating (CDR), and APOE4---during training, but relies only on baseline MRI, APOE4, and patient metadata at inference, reflecting realistic clinical availability. Our contributions are threefold: (1) a Latent Diffusion Model (LDM) fine-tuned via Low-Rank Adaptation (LoRA) for data-efficient learning; (2) a multimodal conditioning that integrates molecular and clinical signals to model heterogeneous disease trajectories; and (3) auxiliary predictors estimating unobserved multimodal biomarkers from MRI features and patient covariates to condition the diffusion model, removing the need for these assessments at inference. Trained on 1,355 visits from 852 subjects in the Alzheimer's Disease Neuroimaging Initiative (ADNI), MC-LDM reduces average regional volumetric MAE by 18% over the second-best baseline on an independent test set of 664 visits. To our knowledge, MC-LDM represents the first diffusion-based framework to leverage multimodal priors for MRI forecasting. This conceptual advancement enables patient-specific trajectory generation that captures aspects of AD heterogeneity not observable from structural anatomy alone, bridging the gap between rich multimodal supervision and the strict constraints of routine clinical deployment. Code is available at: https://github.com/antonioscardace/MC-LDM.

Recognizing fine-grained bodily behaviors, such as self-grooming, postural shifts, and backchanneling, is a useful intermediate step for computational social signal analysis. But the behaviors that matter vary across social contexts, and retraining a model for each new taxonomy is expensive. Vision-language models (VLMs) can adapt to new categories through text prompts, yet they lack sensitivity to body structure and often confuse subtle postural distinctions. We propose PRIMA (Pose Refined Inductive Motion Adapter), a lightweight module that injects a body-structure prior into frozen VLMs using human pose as a training-time supervisory signal, requiring no pose input at inference. PRIMA separates a model-specific projection head from a shared core that encodes transferable body-configuration knowledge; adapting to a new VLM requires retraining only the head. We train PRIMA on a micro-action dataset and evaluate entirely through zero-shot prompting. The adapter improves recognition on its training domain, transfers to additional VLM backbones at different scales with the core frozen, and generalizes to two unseen target datasets from social psychology and daily activity recognition. In a supporting downstream study, the improved behavior predictions also yield higher correlations with human ratings of rapport and dominance in a collaboration context. Within the scope of the tested models and indoor datasets, the results indicate that pose-guided training can produce body-structure priors that generalize across unseen behavior vocabularies and can be partially reused across VLM backbones while preserving RGB-only inference.

We present KernelPack-JAX, a meshless, differentiable, GPU framework in JAX for solving forward and inverse PDE problems in domains whose boundaries, including internal obstacles, are given only by samples of their smooth pieces. Our node generator is based on a new stacked representation of piecewise-smooth boundaries, in which a parametric partition of unity of each smooth piece induces the supports of a curl-free implicit partition of unity of the whole boundary. This representation keeps the edges and corners of the boundary and classifies every point of space without a global fit. Boundary and interior nodes are then selected by a parallel greedy independent-set algorithm that reproduces sequential Poisson-disk sampling exactly. We discretize elliptic and parabolic problems with radial basis function-generated finite differences (RBF-FD) on stencils selected by a new visibility heuristic to avoid crossing the boundary, with ghost nodes for the boundary conditions. In addition, our framework exposes a GPU sparse direct solver to JAX as a differentiable linear solve, so that Jacobians of time-dependent problems cost a small multiple of a forward solve. We derive error estimates that separate consistency from stability and bound the parameter error by the forward error. We demonstrate high-order convergence for Darcy flow, reaction-diffusion, and advection-diffusion in 2D and 3D domains, including one with an internal obstacle, and show that the recovered parameters largely follow the forward errors. We also present timings for every stage.