The Morning
From the arXiv
AutoResearchClaw: Self-Reinforcing Autonomous Research with Human-AI Collaboration
utoResearchClaw is a multi-agent autonomous research system that addresses the iterative nature of scientific discovery. Its core method involves structured multi-agent debate for hypothesis generation and analysis, coupled with a self-healing executor that learns from failures. The key contribution is a robust, collaborative framework that integrates human oversight and cross-run learning to enable more effective and resilient autonomous research.
PEEK: Context Map as an Orientation Cache for Long-Context LLM Agents
PEEK addresses the challenge of LLM agents repeatedly interacting with large contexts by introducing a "context map" as an orientation cache. This map, a small prompt artifact, stores reusable knowledge about the context's content, organization, and useful ent…
ThoughtTrace: Understanding User Thoughts in Real-World LLM Interactions
This paper introduces ThoughtTrace, the first large-scale dataset pairing real-world human-AI conversations with users' explicit thoughts. The core contribution is capturing users' underlying reasoning and reactions, which are semantically distinct from their …
Rethinking How to Remember: Beyond Atomic Facts in Lifelong LLM Agent Memory
This paper proposes TriMem, a novel memory system for LLM agents that moves beyond atomic facts. Instead of relying solely on extracted facts, TriMem maintains three coexisting representation granularities: raw dialogue segments, extracted facts, and synthesiz…
Rewarding Beliefs, Not Actions: Consistency-Guided Credit Assignment for Long-Horizon Agents
This paper proposes ReBel, a reinforcement learning method for long-horizon tasks where agents learn from verifiable rewards. ReBel addresses challenges in partially observable environments by explicitly modeling and updating agent beliefs, using belief consis…
TIDE: Efficient and Lossless MoE Diffusion LLM Inference with I/O-aware Expert Offload
TIDE addresses the challenge of efficiently inferring large Mixture-of-Experts (MoE) diffusion LLMs on resource-constrained devices. Its core method is an I/O-aware expert offload …
A Methodology for Selecting and Composing Runtime Architecture Patterns for Production LLM Agents
This paper introduces the "stochastic-deterministic boundary" (SDB) as a core architectural concept for production LLM agents, defining a four-part contract for integrating LLM out…
BalanceRAG: Joint Risk Calibration for Cascaded Retrieval-Augmented Generation
BalanceRAG addresses the challenge of calibrating cascaded Retrieval-Augmented Generation (RAG) systems. Its core method involves jointly calibrating uncertainty thresholds for bot…
CopT: Contrastive On-Policy Thinking with Continuous Spaces for General and Agentic Reasoning
CopT reverses the traditional Chain-of-Thought by first generating a draft answer and then using "on-policy thinking" to reflect and correct it. This approach leverages continuous …
Detecting Fluent Optimization-Based Adversarial Prompts via Sequential Entropy Changes
This paper proposes a novel method for detecting adversarial prompts by treating them as an online change-point detection problem. It analyzes the stream of next-token entropy, usi…
The Town Square
Google's Gemini 3.5 Flash is a new, faster, and more cost-effective AI model designed for large-scale applications, building on the capabilities of Gemini 3.5 Pro.
Workshops
CLI-Anything enables any software to be controlled by AI agents by translating natural language commands into executable CLI commands, acting as a universal interface for agent-native software interaction.
This repository provides a structured framework for academic research, guiding users through the iterative process of research, writing, review, revision, and finalization.
