The Morning
From the arXiv
$R^3$: Training Robots to Reason in Natural Language via Reinforcement Learning
his paper introduces $R^3$, a method to train Vision-Language Models (VLMs) to reason in natural language for robotic manipulation. $R^3$ first fine-tunes a VLM on expert reasoning traces and then uses reinforcement learning with offline action data to improve its reasoning capabilities for guiding low-level robotic policies. The core contribution is enabling VLMs to perform free-form language reasoning for complex robotic tasks, unlike prior methods relying on structured supervision.
AsymSpec: Context-Asymmetric Speculative Decoding for Agentic LLMs
AsymSpec addresses the inference cost of agentic LLMs by breaking the symmetry in speculative decoding. It uses a lightweight drafter with full context to guide a larger verifier operating on compressed context, mitigating accuracy loss from compression. This …
ProgRouter: Online Progress-Guided Orchestration for Multi-Agent LLM Workflows under Quality-Cost Tradeoffs
ProgRouter addresses the high cost of multi-agent LLM workflows by introducing an online, progress-guided orchestration method. It adaptively selects LLM agents at each step based on a multi-view progress scorer that considers task completion, trends, and stat…
Repair or Resample? Rethinking Failure Debugging in LLM Multi-Agent Systems
This paper addresses the reliability bottleneck in LLM-based multi-agent systems (MASs) for complex tasks. It introduces SymTrace, a framework that enables controlled evaluation of repair methods by allowing selective regeneration of MAS execution trajectories…
SciMIF: Understanding Multimodal Instruction Following in Scientific Domains
This paper introduces SciMIF, a benchmark for evaluating multimodal instruction following in scientific domains. It proposes a taxonomy of 10 constraint groups derived from analyzing 22 tasks across 5 disciplines, and uses this to systematically augment scient…
SwarmWorld: Stigmergic technological evolution in societies of language-model agents
This paper introduces SwarmWorld, a novel framework where language-model agents collectively evolve technologies through indirect environmental interaction (stigmergy). The core me…
ToST: A Tree-of-Thought Socratic Teaching Framework for Multi-Path Guidance and Parallel Thinking
This paper introduces ToST, a novel framework for Socratic teaching using LLMs. ToST moves beyond single-solution guidance by employing "Parallel Sowing" to encourage students to e…
A Self-Evolving Multi-Agent Framework Defense against LLM Jailbreak Attacks
This paper introduces a novel, self-evolving defense framework for LLMs against jailbreak attacks. Its core method involves a persistent rule memory that learns from successful att…
VISA: Agentic Self-Evolving Data Synthesis for Multimodal Instruction Following
VISA is an agentic framework that synthesizes multimodal instruction-following data through a self-evolving loop. It iteratively generates, verifies, and refines instructions by an…
Candidate supply and answer selection shape the value of LLM judging in multi-agent systems
This paper investigates how LLM judges can improve multi-agent system (MAS) performance by providing selection pressure on candidate answers. The core method involves conceptualizi…
The Town Square
Zhipu AI has released GLM-5.3-Flash, a new large language model optimized for faster inference and improved performance on various tasks.
Workshops
This repository is an industrial-grade prompt engine and template library for GPT-Image2, offering over 530 reverse-engineered examples, 20+ industrial templates, and extracted "Skills" for prompt-as-code functionality, all continuously updated.
Archify is an AI agent skill that generates beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams as self-contained, interactive HTML files with crisp export options.
