The Morning
From the arXiv
EvoUndo: Recoverability-Constrained Self-Evolution for LLM Agent Harnesses
voUndo addresses the challenge of LLM agents self-modifying their systems, which can lead to irreversible changes. The framework enables the representation, synthesis, diagnosis, and verification of recoverability for these self-modifications across different states. EvoUndo significantly improves the ability to recover from failed mutations compared to conventional methods.
GRACE:Gradient-guided Coreset Selection for LLM Unlearning
GRACE addresses the challenge of unlearning unwanted knowledge from LLMs when only a few examples of undesired behavior are provided. Its core method is a gradient-guided coreset selection that constructs both forget and retain sets. GRACE first identifies a "…
Learning to Use Tools: Reinforcement Learning for Tool-Integrated Mathematical Reasoning
This paper introduces a method for improving mathematical reasoning in LLMs by integrating calculator tool use. The core approach involves supervised fine-tuning to teach tool usage patterns, followed by reinforcement learning with automated rewards to optimiz…
ContextPilot: Teaching Agents for Proactive Context Management via Fine-grained RL
ContextPilot addresses the challenge of managing extensive context in long-horizon agentic tasks by introducing a fine-grained Reinforcement Learning (RL) framework. Its core method involves teaching agents to proactively and adaptively manage their working co…
PersonaForge: Realistic Multi-Turn User Simulation for Agentic Systems
PersonaForge addresses the lack of realistic multi-turn user interaction data for training agentic systems. It synthesizes these interactions by combining a persona space, SOUL-driven behavioral control, and Reverse Deep Construction. This framework generates …
Finding Where the Buck Stops: An Automated Failure Attribution-Based Reflection Framework for Multi-Agent Collaboration
This paper introduces DoCtOR, an automated framework for multi-agent collaboration that addresses high failure rates. Its core method is to first identify the specific agent and st…
Layered LLM Defenses as an Ensemble: Access Tiers, Inference Cost, and the Measured Failure Correlation Between Defense Layers
This paper introduces a framework to analyze stacked LLM defenses as an ensemble. It proposes the Adversary Access-Tier Model (AATM) and an inference cost model to quantify defense…
LLM-Based Agents for Software and Systems Security: Approaches, Applications, and Assessment
This paper systematically reviews LLM-based agents for software and systems security. It categorizes existing approaches, applications, and assessment methods to provide a coherent…
LongPIBench: A Long-Context Benchmark for Prompt Injection
This paper introduces LongPIBench, a novel benchmark designed to evaluate prompt injection attacks and defenses in long-context scenarios, addressing a critical gap in existing res…
NL2AGBench: Benchmarking LLM Auto-Formalization for AlphaGeometry
This paper introduces NL2AGBench, a new benchmark designed to evaluate how well Large Language Models (LLMs) can convert informal English geometry problems into the formal language…
The Town Square
This tech news story, "No AI Fridays," likely discusses the latest developments and trends in artificial intelligence, possibly with a focus on weekly updates or analyses.
Workshops
This repository provides a comprehensive library of validated AI agent skills and scientific databases, enabling any AI agent to function as a scientist across various disciplines like biology and chemistry.
This repository provides a command-line interface (CLI) tool for interacting with a virtual phone, enabling users to manage calls, messages, and other phone functionalities from their terminal.
