Researchers present TimeWarp, a benchmark designed to reflect how websites change over time, addressing the concern that agents may not generalize from today's web to tomorrow's. It comprises three web environments that emulate an evolving web so agent performance can be measured under such shifts. The work appears as a replacement submission on arXiv.
A new arXiv preprint examines using neural networks to represent ground-state electron charge density in real space. The work is motivated by the Hohenberg-Kohn theorem, which holds that ground-state density encodes all ground-state information about a many-electron system. The authors also study how well such learned models generalize.
A new arXiv preprint introduces ModularRSI, a method aimed at making recursive self-improvement of agent harnesses both modular and generalizable. Prior work has shown agents can refine their execution mechanisms through experience on long-horizon coding and terminal tasks, but transferring those gains across settings remains difficult. The paper frames generalization as the main open obstacle for harness-level self-improvement.
A new arXiv preprint develops a theoretical account of why diffusion models sometimes reproduce training data verbatim rather than generating novel samples. The authors propose smoothing the score function as a way to reduce this memorization effect and improve generalization. The work is presented as an explanation of the phenomenon rather than a new model release.