AI agents may soon become capable of autonomously completing valuable, long-horizon tasks in diverse domains. Current benchmarks either do not measure real-world tasks, or are not sufficiently difficult to meaningfully measure frontier models. To this end, we present Terminal-Bench 2.0: a carefully curated hard benchmark composed of 89 tasks in computer terminal environments inspired by problems from real workflows. Each task features a unique environment, human-written solution, and comprehensive tests for verification. We show that frontier models and agents score less than 65% on the benchmark and conduct an error analysis to identify areas for model and agent improvement. We publish the dataset and evaluation harness to assist developers and researchers in future work at tbench.ai.
Agent Skills are structured packages of procedural knowledge that augment large language model (LLM) agents at inference time. Despite rapid adoption, there is no standard way to measure whether they actually help. We present SkillsBench, a benchmark whose current inventory contains 87 tasks across 8 domains paired with curated Skills and deterministic verifiers. Our latest aggregate evaluation runs the 87-task benchmark under matched no-Skills and curated-Skills conditions for 18 model-harness configurations. Curated Skills raise the average pass rate from 33.9
. The physical basis of grain size in polar ice crystals as a climate proxy has remained largely elusive. Here, this work investigates the influence of temperature, specifically between -30 and -3 degrees C, on grain evolution in firn samples from Summit, Greenland, simulating conditions encountered during the transport and storage of ice core samples. Utilizing 2-D optical micrographs, 3-D X-ray micro-computed tomography, and grain development models, the research reveals negligible variation in ice crystal grain size. Notably, the ability of these grains to return to their original size at a specific temperature, despite subsequent temperature changes, is referred to as the memorability of grain size on temperature (MoGSoT). This phenomenon can be attributed to the reduction in grain size caused by temperature gradient metamorphisms and the screw-step growth observed under isothermal conditions. In addition, sinusoidal signals incorporating six composite frequencies were developed to model temperature variations related to grain size, elucidating periodic climate cycles spanning multi-decadal, multi-centennial, and ten-millennial periods. Beyond establishing MoGSoT, which highlights the fundamental physics of polar ice grain size and its role in paleoclimate reconstruction, this work also emphasizes the intricate interactions between external factors, e.g., astronomical, orbital, solar, and planetary influences, and the internal dynamics of the climate system, including ocean-atmosphere oscillations, thereby offering new perspectives on the origins of climate change, whether anthropogenic or natural.
BackgroundLymphedema is a chronic disease characterized by swelling, inflammation, adipose deposition, and fibrosis. In the United States, lymphedema occurs most frequently as a sequela of oncologic lymphadenectomy. Axillary lymph node dissection (ALND) for breast cancer has been associated with a 5%-40% incidence of upper limb lymphedema. In immediate lymphatic reconstruction (ILR), surgeons perform lymphovenous anastomosis (LVA) with ALND to prevent the development of future lymphedema. Although early evidence supports a protective effect of ILR, there have been few large-scale studies on the topic. We leveraged a large claims database to better characterize the outcomes of ILR and national trends related to its adoption.MethodsAdult female patients with breast cancer who underwent axillary lymph node dissection (ALND) between 2007 and 2022 were identified within the Merative MarketScan Research Databases and stratified according to whether they underwent ILR. Adjusted odds of undergoing ILR, both overall and regionally, and of experiencing postmastectomy lymphedema syndrome (PMLS) were calculated.ResultsOf all patients undergoing mastectomy and ALND, 1.3% received ILR. ILR was associated with decreased odds of developing PMLS (OR, 0.81; P = 0.02). The frequency of ILR procedures increased in all regions between 2017 and 2022, most dramatically in the Midwest and East US. More recent surgery year, younger age, complete mastectomy, and delayed ALND elevated the odds of undergoing ILR (P <= 0.01).ConclusionsThe popularity of ILR has grown rapidly since 2017, with significant regional variability. Patients with breast cancer who undergo ILR at the time of ALND are significantly less likely to develop lymphedema.
The May 2024 geomagnetic superstorm provided the opportunity to explore how strong wave-particle interactions affect energetic electron precipitation under intense driving. Using coordinated measurements from a balloon-borne Timepix-based X-ray detector, ground-based riometers and magnetometers, and Arase satellite observations, we identified quasi-periodic bursts of energetic electron precipitation coincident with Pc5 ultra low frequency (ULF) wave oscillations. Arase satellite data revealed energy-dispersed trapped energetic electron flux modulations in the "seed" energy range, indicating that trapped electron flux was likely modulated by ULF waves. This letter reveals that these flux enhancements surpassed the Kennel-Petschek (K-P) limit, creating intense chorus waves and driving periodic electron precipitation. Drift-dispersion analysis traced these modulations back to a source in the post-noon magnetospheric sector, matching balloon and ground-based measurements. Here, we propose a novel indirect ULF wave-driven mechanism for modulated energetic electron precipitation, whereby periodic modulations of "seed" electron fluxes enhance electron losses.
