VOL. 039
8 research signalsDAILY EDITION / SAVED SNAPSHOT
A dated Agentic RL Daily snapshot with 8 verified primary-source signals across papers, official releases, deployment evidence, and safety or alignment findings.
EDITOR'S VIEW
FULL EDITION
Archived / 2026-08-25
01
Molecular science represents an important frontier for LLM-based agents.
Molecular LLM agents must perceive, reason about, and act upon chemical objects across symbolic strings, molecular graphs, 3D conformations, and other molecular representations.
02
Memory is becoming a default subsystem in deployed LLM agents to provide persistent personalization and continuity.
InjecMEM is a novel memory injection attack paradigm that requires only a small amount of memory to inject malicious data into an LLM agent's memory system.
03
Space robots operate in extreme environments where hardware degradation can critically compromise traditional control strategies.
This work proposes a reward-free continual adaptation framework for resilient space robots, which can adapt to changing environments without requiring access to a reward signal.
04
Agent skills are reusable procedural artifacts that extend language agents with specialized workflows, tool conventions, and domain behaviors at inference time.
This work proposes a framework for creating reliable skills for language agents, which can be used to extend their capabilities in various domains.
05
Open-world video understanding often requires a model to locate sparse visual evidence and acquire external knowledge that is absent from the video and its parametric memory.
This work proposes a unified framework for video reasoning and deep research, which can be used to develop open-world video agents that can reason and learn from video data.
06
Hint-based reinforcement learning addresses reward sparsity in long-horizon agentic tasks by retaining a prefix of an expert trajectory before each rollout, letting the policy explore from a state closer to success.
This work proposes a new guidance mechanism for agentic reinforcement learning, which can be used to improve the performance of language agents in various tasks.
07
Large language model (LLM) agents are increasingly attractive for automating network configuration, yet their reliability and failure patterns are poorly understood.
This work proposes a new benchmark for evaluating the performance of LLM agents in closed-loop network configuration tasks.
08
Language models are sequential processors, but long-horizon agency requires external information and computation beyond model weights and active context.
This work proposes a new harness for long-horizon evaluation and coding-agent workflows, which can be used to improve the performance of language agents in various tasks.