AgentSWE benchmark tests what coding agents actually deliver
The arXiv preprint defines 25 tasks across agent creation, editing existing code and optimization toward a target.
Technology, science, digital access and research data, with sources, scope and measurement periods. (Page 42 of 53)
The arXiv preprint defines 25 tasks across agent creation, editing existing code and optimization toward a target.
The 8th Science and Innovation Council announced a partnership linking UK Catapults and India’s Science and Technology Clusters for technology development, validation, industry collaboration, pilots and commercialisation.
The Ministry of Justice is expanding a system that connects judges and hearing participants digitally, building on work in London and the North-East; participants retain the right to request an in-person hearing.
NASA’s final solicitation invites companies to propose low-Earth-orbit destinations for human spaceflight, with proposals due Dec. 8 and awards expected in spring 2027.
NASA’s KREPE-3 experiment will use 12 small capsules carried by a Cygnus cargo spacecraft to collect re-entry data on heat-shield materials.
Researchers at ETH Zurich and Yonsei University developed a parallel FPGA system for Max-Cut optimization. For the K2000 case, the paper gives a 99% target success-probability threshold for a 12-engine portfolio at a 0.05 ms round time.
Researchers at JD.com and Beijing University of Posts and Telecommunications developed CompileMask, a compiler that connects flexible attention masks to efficient GPU kernels. Blackwell GPU experiments measured higher throughput than Flex-FA4 for several mask patterns.
Researchers at the Technical University of Munich developed Ember, a memory manager that tracks which data should remain on a GPU when its memory fills. In the paper’s data-intensive workloads, it reported higher speeds than standard Unified Virtual Memory on three NVIDIA GPUs.
Researchers at KTH Royal Institute of Technology introduced GraphTok, a system that tokenizes large language model requests on a GPU. In a Dynamo frontend experiment, a real H100 NVL performed tokenization while backend GPU workers were simulated.
Researchers affiliated with Hong Kong Polytechnic University and other institutions in China modeled optical circuit switching alongside network scheduling to shift AI data-center training toward renewable-energy windows. In a controlled five-center scenario, modeled grid-purchase cost was 12.50% lower and local renewable-energy use rose by 2.724 percentage points.
Researchers at UC Merced and the University of Texas at Austin introduced SlackSwap, a planner for deep-learning training. It recomputes some activations when the transfer budget between a GPU and host memory is exceeded; reported results come from simulation, not hardware measurements.
Microchip announced a product portfolio for AI infrastructure that includes a PCIe Gen6 switch and storage-controller updates.