Loading the edition…
· Past edition · 9 stories
Retrospective edition reconstructed from later captures. Paper versions may include later updates; coverage labels describe the evidence used.
Research & academia
A paper distinguishes output review from checking an approved research contract.
Why it matters: Assessing whether a study followed its approved plan requires records of that plan.
Research & academia
Combines code-defined simulated worlds with a shared neural renderer, and studies agents carrying playbooks across interactive learning rounds.
Open each story in the reader. Summaries are shown when available; some sources provide an abstract only. With an account you can keep your own notes.
Why it matters: Reusable simulated environments may broaden the settings in which agents practice.
Research & academia
Presents a serving runtime for token-level routing among language models and reports throughput improvements under tested workloads.
Why it matters: Token-level routing requires serving systems that coordinate models efficiently.
Research & academia
Synthesizes runnable software environments to train computer-use agents and reports results on a held-out benchmark.
Why it matters: Synthesized applications may broaden computer-use training beyond a few fixed environments.
Research & academia
Describes agents maintaining revisable rulebooks compiled into executable world models, checked against observed transitions.
Why it matters: An explicit, testable memory can help an agent revise its understanding of an environment.
Research & academia
Introduces a benchmark for agents finding and fixing interaction-dependent bugs in web applications, reporting failure to discover bugs as a major bottleneck.
Why it matters: Interactive bugs require agents to discover failures before they can fix them.
Research & academia
Normalizing Trajectory Models (NTM) aims to make image generation work in four sampling steps while retaining exact likelihood over the generative trajectory.
Why it matters: NTM targets a practical trade-off: few-step image generation without giving up exact likelihood over the generative trajectory, a capability the abstract says the other few-step approaches sacrifice.