The Graph-Augmented Evolution framework combines large language models with reinforcement optimization to improve automated scientific discovery by addressing limitations in evolutionary program search.
The authors introduce Who and When Pro, a large-scale benchmark designed to evaluate how effectively large language models can attribute failures in agentic workflows.
This study explores whether looped architectures and adaptive exit strategies can improve the computational depth and reasoning of state-space language models.
Researchers developed an automated tensor scheduling system that optimizes hybrid CPU-GPU memory offloading to run large language models more efficiently on consumer devices.
The ScaleCUA framework aims to improve computer use agents by combining verifiable task synthesis with efficient online reinforcement learning.
Researchers introduce DUNE, a training-free refinement method that reduces artifacts in diffusion models by analyzing and stabilizing early-stage latent fluctuations.
Chinese startup Novasilicon has secured multi-million yuan in seed funding to develop AI-driven chip design solutions.
Developers have successfully compiled the Firefox browser into WebAssembly, allowing it to run entirely within another web browser.
The paper introduces NameRank, a metric designed to measure how well large language models recognize specific researchers and tools within their parametric memory.
This paper critiques the prevailing optimization culture in AI alignment, arguing that evaluating models solely on predefined, measurable axes fails to capture true value.
Researchers analyzed the citation faithfulness and coverage of a four-billion parameter research agent running locally on a consumer laptop.
The paper introduces the Internet of Agentic Things, an architectural framework that connects autonomous AI agents across cloud, edge, and physical IoT layers.
Verrus, an Alphabet spin-out, is planning the construction of a new grid-reactive data center campus in Oregon to support infrastructure needs.
Chinese storage company Demingli projects a first-half net profit of up to 6.5 billion yuan, reversing a previous loss due to surging AI-driven storage demand.
The global artificial intelligence boom has driven up memory and solid-state drive prices in Shenzhen's Huaqiangbei electronics market, significantly increasing costs for PC builders.
FixItFlow is an automated system that leverages large language models to generate technical troubleshooting guides from historical cloud incident data.
A new decoupled training strategy for transfer learning aims to reduce computational and energy costs by optimizing classifier heads separately from feature extraction layers.
An analysis of Kaggle contest submissions examines whether the use of AI coding assistants is leading to increased homogenization of software development outputs.
The authors introduce STKAN, a spatio-temporal forecasting architecture that leverages Kolmogorov-Arnold Networks to better model complex traffic data.
The SteinGate framework introduces a new safety certification method for reinforcement learning to better mitigate rare but catastrophic risks.
A new method for test-time learning introduces learnable adaptation policies to help language agents improve their performance through iterative interaction.
This paper provides a theoretical analysis of the Muon optimizer to clarify the mechanisms behind its performance advantages in large-scale deep learning.
This paper proposes a framework for networked intelligence that uses shared context graphs to facilitate collaboration among multiple AI agents in scientific research.
A new benchmark based on cognitive psychology tests has been developed to evaluate how AI agents adapt when the reliability of their tools changes during operation.