The Morning
From the arXiv
A Scalable Multi-LLM Collaboration System with Retrieval-based Selection and Exploration-Exploitation-Driven Enhancement
his paper introduces SMCS, a scalable system for multi-LLM collaboration. It addresses scalability issues by using a retrieval module to select the best LLMs for a given task and an enhancement module to improve response diversity and quality. SMCS demonstrates superior performance compared to existing closed-source LLMs by effectively integrating multiple open-source models.
Ada-Diffuser: Latent-Aware Adaptive Diffusion for Decision-Making
Ada-Diffuser addresses decision-making by treating it as sequence modeling with diffusion models. Its core method is a unified framework that explicitly infers and models evolving latent dynamics alongside observed interactions. This allows for more precise en…
Agentic Discovery of Neural Architectures: AIRA-Compose and AIRA-Design
This paper introduces AIRA, a dual-framework approach where LLM agents autonomously discover novel neural architectures. AIRA-Compose searches for high-level primitives, while AIRA-Design handles low-level implementation, leading to new Transformer-based and h…
AstraFlow: Dataflow-Oriented Reinforcement Learning for Agentic LLMs
AstraFlow addresses the high cost of reinforcement learning for agentic LLMs by introducing a dataflow-oriented system. It decouples rollout, dataflow management, and training into autonomous components, enabling efficient support for complex, multi-policy tra…
Beyond Individual Intelligence: Surveying Collaboration, Failure Attribution, and Self-Evolution in LLM-based Multi-Agent Systems
This paper surveys LLM-based multi-agent systems by proposing a unified framework called the LIFE progression. It highlights how individual agent capabilities (Lay) enable collaboration (Integrate), which in turn necessitates fault attribution (Find) for effec…
CAP: Controllable Alignment Prompting for Unlearning in LLMs
This paper introduces CAP, a novel prompt-driven method for unlearning sensitive information in LLMs without modifying model weights. CAP uses reinforcement learning to optimize a …
Don't Retrieve, Navigate: Distilling Enterprise Knowledge into Navigable Agent Skills for QA and RAG
This paper introduces Corpus2Skill, a method that distills enterprise knowledge into a navigable, hierarchical skill directory. Instead of passively retrieving information, an LLM …
Frontier Large Language Models Rival State-of-the-Art Planners
This paper demonstrates that recent frontier Large Language Models (LLMs) can rival state-of-the-art classical planners on challenging planning tasks. Specifically, Gemini 3.1 Pro …
FutureWorld: A Live Reinforcement Learning Environment for Predictive Agents with Real-World Outcome Rewards
FutureWorld introduces a novel reinforcement learning environment for training predictive agents that learn from real-world outcomes. Its core method, verl-tool-future, addresses t…
How to Train Your Advisor: Steering Black-Box LLMs with Advisor Models
This paper introduces "Advisor Models," a novel method to enhance black-box large language models (LLMs) by training smaller, open-weight models to provide dynamic, instance-specif…
The Town Square
The article argues that AI-generated content is essentially large-scale, unauthorized plagiarism because it's trained on copyrighted material without permission or compensation to creators.
Workshops
This repository provides a pre-indexed code knowledge graph for AI models like Claude Code, Codex, and Cursor, enabling more efficient, local code understanding with fewer tokens and tool calls.
This repository's core function is to enhance Claude Code's programming capabilities by providing a single CLAUDE.md file, distilled from Andrej Karpathy's insights into common LLM coding errors.
