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 paper contends that agent benchmarks assume each policy implies one correct action, an assumption that natural-language policies often break through silence, ambiguity, or contradiction. The authors describe these as policy loopholes, where multiple defensible readings exist but evaluations still count a single behavior as an agent error. The work suggests such ambiguous cases should be separated from genuine policy-compliance failures in benchmark scoring.
A revised arXiv preprint (2609.06396v2) introduces MetaRSI, a framework that applies recursive self-improvement at a meta level, treating the model-building machinery itself as the target of improvement so later generations inherit the gains. The authors argue that RSI research has so far been validated almost exclusively on coding and formal benchmarks such as science QA, and their approach seeks to broaden where such gains hold.