Explore the vast range of titles available.

Extended reasoning in large language models (LLMs) creates severe KV cache memory bottlenecks. Leading KV cache compression methods estimate KV importance using attention scores from recent post-RoPE queries. However, queries rotate with position during RoPE, making representative queries very few, leading to poor top-key selection and unstable reasoning. To avoid this issue, we turn to the pre-RoPE space, where we observe that Q and K vectors are highly concentrated around fixed non-zero centers and remain stable across positions -- Q/K concentration. We show that this concentration causes queries to preferentially attend to keys at specific distances (e.g., nearest keys), with the centers determining which distances are preferred via a trigonometric series. Based on this, we propose TriAttention to estimate key importance by leveraging these centers. Via the trigonometric series, we use the distance preference characterized by these centers to score keys according to their positions, and also leverage Q/K norms as an additional signal for importance estimation. On AIME25 with 32K-token generation, TriAttention matches Full Attention reasoning accuracy while achieving 2.5x higher throughput or 10.7x KV memory reduction, whereas leading baselines achieve only about half the accuracy at the same efficiency. TriAttention enables OpenClaw deployment on a single consumer GPU, where long context would otherwise cause out-of-memory with Full Attention.

Autonomous LLM agents fail because long-horizon policy remains implicit in model weights and transcripts, while safety is retrofitted post hoc. We propose Traversal-as-Policy: distill sandboxed OpenHands execution logs into a single executable Gated Behavior Tree (GBT) and treat tree traversal -- rather than unconstrained generation -- as the control policy whenever a task is in coverage. Each node encodes a state-conditioned action macro mined and merge-checked from successful trajectories; macros implicated by unsafe traces attach deterministic pre-execution gates over structured tool context and bounded history, updated under experience-grounded monotonicity so previously rejected unsafe contexts cannot be re-admitted. At runtime, a lightweight traverser matches the base model's intent to child macros, executes one macro at a time under global and node-local gating, and when stalled performs risk-aware shortest-path recovery to a feasible success leaf; the visited path forms a compact spine memory that replaces transcript replay. Evaluated in a unified OpenHands sandbox on 15+ software, web, reasoning, and safety/security benchmarks, GBT improves success while driving violations toward zero and reducing cost. On SWE-bench Verified (Protocol A, 500 issues), GBT-SE raises success from 34.6% to 73.6%, reduces violations from 2.8% to 0.2%, and cuts token/character usage from 208k/820k to 126k/490k; with the same distilled tree, 8B executors more than double success on SWE-bench Verified (14.0%58.8%) and WebArena (9.1%37.3%).

Dimerized valence bond solids appear naturally in spin-1/2 systems on bipartite lattices, with the geometric frustrations playing a key role both in their stability and the eventual `melting' due to quantum fluctuations. Here, we ask the question of the stability of such dimerized solids in spin-1 systems, taking the anisotropic square lattice with bilinear and biquadratic spin-spin interactions as a paradigmatic model. The lattice can be viewed as a set of coupled spin-1 chains, which in the limit of vanishing inter-chain coupling are known to possess a stable dimer phase. We study this model using the density matrix renormalization group (DMRG) and infinite projected entangled-pair states (iPEPS) techniques, supplemented by the analytical mean-field and linear flavor wave theory calculations. While the latter predicts the dimer phase to remain stable up to a reasonably large interchain-to-intrachain coupling ratio \\(r 0.6\\), the DMRG and iPEPS find that the dimer solid melts for much weaker interchain coupling not exceeding \\(r 0.15\\). We find the transition into a magnetically ordered state to be first order, manifested by a hysteresis and order parameter jump, precluding the deconfined quantum critical scenario. The apparent lack of stability of dimerized phases in 2D spin-1 systems is indicative of strong quantum fluctuations that melt the dimer solid.

We study instruction-based image editing under professional workflows and identify three persistent challenges: (i) editors often over-edit, modifying content beyond the user's intent; (ii) existing models are largely single-turn, while multi-turn edits can alter object faithfulness; and (iii) evaluation at around 1K resolution is misaligned with real workflows that often operate on ultra high-definition images (e.g., 4K). We propose Agent Banana, a hierarchical agentic planner-executor framework for high-fidelity, object-aware, deliberative editing. Agent Banana introduces two key mechanisms: (1) Context Folding, which compresses long interaction histories into structured memory for stable long-horizon control; and (2) Image Layer Decomposition, which performs localized layer-based edits to preserve non-target regions while enabling native-resolution outputs. To support rigorous evaluation, we build HDD-Bench, a high-definition, dialogue-based benchmark featuring verifiable stepwise targets and native 4K images (11.8M pixels) for diagnosing long-horizon failures. On HDD-Bench, Agent Banana achieves the best multi-turn consistency and background fidelity (e.g., IC 0.871, SSIM-OM 0.84, LPIPS-OM 0.12) while remaining competitive on instruction following, and also attains strong performance on standard single-turn editing benchmarks. We hope this work advances reliable, professional-grade agentic image editing and its integration into real workflows.

Writing competitive programming problems is exacting. Authors must: set constraints, input distributions, and edge cases that rule out shortcuts; target specific algorithms (e.g., max-flow, dynamic programming, data structures); and calibrate complexity beyond the reach of most competitors. We argue that this makes for an ideal test of general large language model capabilities and study whether they can do this reliably. We introduce AutoCode, which uses multiple rounds of validation to yield competition-grade problem statements and test cases. On held-out problems, AutoCode test suites approach 99% consistency with official judgments, a significant improvement over current state-of-the-art methods like HardTests, which achieve less than 81%. Furthermore, starting with a random seed problem, AutoCode can create novel variants with reference and brute-force solutions. By cross-verifying these generated solutions against test cases, we can further filter out malformed problems. Our system ensures high correctness, as verified by human experts. AutoCode successfully produces novel problems judged by Grandmaster-level (top 0.3%) competitive programmers to be of contest quality.

Producing valuable chemicals like ethylene via catalytic carbon monoxide conversion is an important nonpetroleum route. Here we demonstrate an electrochemical route for highly efficient synthesis of multicarbon (C 2+ ) chemicals from CO. We achieve a C 2+ partial current density as high as 4.35 ± 0.07 A cm −2 at a low cell voltage of 2.78 ± 0.01 V over a grain boundary-rich Cu nanoparticle catalyst in an alkaline membrane electrode assembly (MEA) electrolyzer, with a C 2+ Faradaic efficiency of 87 ± 1% and a CO conversion of 85 ± 3%. Operando Raman spectroscopy and density functional theory calculations reveal that the grain boundaries of Cu nanoparticles facilitate CO adsorption and C − C coupling, thus rationalizing a qualitative trend between C 2+ production and grain boundary density. A scale-up demonstration using an electrolyzer stack with five 100 cm 2 MEAs achieves high C 2+ and ethylene formation rates of 118.9 mmol min −1 and 1.2 L min −1 , respectively, at a total current of 400 A (4 A cm −2 ) with a C 2+ Faradaic efficiency of 64%. Producing valuable chemicals like ethylene via catalytic CO conversion is an important nonpetroleum route. Here, authors demonstrate high-rate electrosynthesis of multicarbon chemicals via CO electrolysis, with a multicarbon product partial current density of 4.35 A cm −2 at a cell voltage of 2.78 V.

Background Evidence indicates that living arrangements significantly influence the well-being and health of older adults. However, the psychological needs of this population have increasingly emerged as important public health concerns, with preferences regarding living arrangements serving as a key determinant. Ignoring this issue can lead to incomplete or inaccurate assessments of how various living arrangements affect their health. To better understand this impact, this study investigates the relationship between living arrangements and survival outcomes among older adults, while considering the alignment between Actual Living Arrangements and Expectations (ALivE). Methods Data for this study were sourced from the China Longitudinal Healthy Longevity Survey (CLHLS) covering the period from 2005 to 2018. Living arrangements were assessed based on participants’ self-reported expect living arrangement (indicating preference) and current situations in the questionnaire, categorized into “living alone (or with a spouse only)”, “co-residence with children”, or “living in an institution”. Survival status and date of death were determined through interviews with close family members during each survey round. Cox proportional hazards regression models were used for longitudinal analysis, and strategies for heterogeneity analysis were employed to identify vulnerable subgroups. Results The study included 4,272 older adults, of whom 2,188 (51.22%) died during the follow-up period (2008 to 2018). The median survival time was 5 years. Compared to individuals with consistent living arrangements, those with discrepancies between actual and preferred arrangements had a hazard ratio (HR) of 1.36 ( 95% CI : 1.20–1.53), indicating a 36% increased risk of death. Sensitivity to these inconsistencies was greater among males, individuals with fixed occupations before age 60, those with average or difficult economic conditions, individuals with unfavorable dietary patterns, and those who are not currently exercising. Conclusions This study highlights the significance of aligning older adults’ living arrangements with their preferences. The association with health is more closely related to whether living arrangements match personal desires rather than the specific type of arrangement.

Hepatocellular carcinoma (HCC) is a prevalent and heterogeneous tumor with limited treatment options and unfavorable prognosis. The crucial role of a disintegrin and metalloprotease (ADAM) gene family in the tumor microenvironment of HCC remains unclear. This study employed a novel multi-omics integration strategy to investigate the potential roles of ADAM family signals in HCC. A series of single-cell and spatial omics algorithms were utilized to uncover the molecular characteristics of ADAM family genes within HCC. The GSVA package was utilized to compute the scores for ADAM family signals, subsequently stratified into three categories: high, medium, and low ADAM signal levels through unsupervised clustering. Furthermore, we developed and rigorously validated an innovative and robust clinical prognosis assessment model by employing 99 mainstream machine learning algorithms in conjunction with co-expression feature spectra of ADAM family genes. To validate our findings, we conducted PCR and IHC experiments to confirm differential expression patterns within the ADAM family genes. Gene signals from the ADAM family were notably abundant in endothelial cells, liver cells, and monocyte macrophages. Single-cell sequencing and spatial transcriptomics analyses have both revealed the molecular heterogeneity of the ADAM gene family, further emphasizing its significant impact on the development and progression of HCC. In HCC tissues, the expression levels of ADAM9, ADAM10, ADAM15, and ADAM17 were markedly elevated. Elevated ADAM family signal scores were linked to adverse clinical outcomes and disruptions in the immune microenvironment and metabolic reprogramming. An ADAM prognosis signal, developed through the utilization of 99 machine learning algorithms, could accurately forecast the survival duration of HCC, achieving an AUC value of approximately 0.9. This study represented the inaugural report on the deleterious impact and prognostic significance of ADAM family signals within the tumor microenvironment of HCC.

Cancer associated fibroblasts (CAFs) support tumors via multiple mechanisms, including maintaining the immunosuppressive tumor microenvironment and limiting infiltration of immune cells. The prolyl isomerase Pin1, whose overexpression in CAFs has not been fully profiled yet, plays critical roles in tumor initiation and progression. To decipher effects of selective Pin1 inhibition in CAFs on pancreatic cancer, here we formulate a DNA-barcoded micellular system (DMS) encapsulating the Pin1 inhibitor AG17724. DMS functionalized with CAF-targeting anti-FAP-α antibodies (antiCAFs-DMS) can selectively inhibit Pin1 in CAFs, leading to efficacious but transient tumor growth inhibition. We further integrate DNA aptamers (AptT), which can engage CD8+ T lymphocytes, to obtain a bispecific antiCAFs-DMS-AptT system. AntiCAFs-DMS-AptT inhibits tumor growth in subcutaneous and orthotopic pancreatic cancer models. Pharmacological inhibition of the prolyl isomerase PIN1, highly expressed in cancer cells and cancer associated fibroblasts (CAF), has been proposed for cancer therapy. Here the authors report the design of a DNA-barcoded micellular system functionalized with antibodies targeting CAFs and a T cell recruiting aptamer to deliver the PIN1 inhibitor AG17724, showing antitumor response in preclinical models of pancreatic cancer.

Speed breeding by artificial control of photothermal conditions facilitates generation advancement but was limited in scale and cost. In this study, we demonstrated a cost-saving off-site summer nursery pattern, taking full advantage of shorter daylength and higher temperature with lower latitude compared to the origin of the soybean cultivars used in the study. This substantially reduced the generation cycles under totally natural conditions. Using this approach, two generations of soybean cultivars from Northeastern Spring Planting Region (NE) and Yellow-Huai-Hai Valleys Summer Planting Region (YHH) were successfully obtained in Beijing and Hainan, respectively, compared to one generation in origin. Fresh-seeding method was also used to further shorten the generation duration by 7–10 days, thereby allowing at least four generations per year. Using DNA markers to define haplotypes of maturity genes E1–E4 , we proposed a model to predict the optimum adaptation region of the advanced generation lines. Taken together, we present a speed-breeding methodology combining off-site nursery, fresh-seeding method, and marker-assisted selection, aimed at accelerating soybean improvement.