The Morning
From the arXiv
AgentCompass: A Unified Evaluation Infrastructure for Agent Capabilities
gentCompass provides a unified, open-source infrastructure for evaluating LLM-based agents by decoupling benchmarks, harnesses, and environments. This modular design promotes flexibility, reproducibility, and reduces redundant engineering. Its key contribution lies in offering a scalable and transparent system for diagnosing agent failures and advancing agent research.
Automatic Ordinary Differential Equations Discovery For Biological Systems Using Large Language Model Powered Agentic System
This paper introduces MEDA, an LLM- and symbolic regression-powered agentic system for discovering Ordinary Differential Equation (ODE) models of biological systems. MEDA leverages LLMs to retrieve domain knowledge, define variables, and generate constraints, …
Experience Memory Graph: One-Shot Error Correction for Agents
This paper introduces the Experience Memory Graph (EMG) to address error correction in LLM agents. EMG reformulates failure recovery as a graph matching problem, converting failed and successful trajectories into directed action decision graphs. By matching th…
Groc-PO: Grounded Context Preference Optimization for Truthful Multimodal LLMs
This paper addresses untruthfulness in multimodal LLMs by proposing Groc-PO. Unlike standard methods that optimize only the final answer, Groc-PO applies preference optimization at earlier grounding stages. This grounded approach aims to prevent error propagat…
Memory as a Controlled Process: Learned Adaptive Memory Management for LLM Agents
This paper introduces MemCon, a framework that treats LLM agent memory management as a learned, adaptive process. Instead of fixed heuristics, MemCon uses a Markov Decision Process to train an online policy that dynamically decides when, what, and how to retri…
Protective Capacity Hallucination: When Large Language Models Claim Nonexistent Capabilities
This paper introduces "Protective Capacity Hallucination" (PCH), where Large Language Models (LLMs) falsely claim to perform real-world protective actions they cannot execute when …
STOCKTAKE: Measuring the Gap Between Perception and Action in LLM Agents with a Fair Oracle
This paper introduces STOCKTAKE, a benchmark for evaluating LLM agents on complex, multi-week decision tasks where the true state of the world is hidden. STOCKTAKE's core method us…
TRACE: Turn-level Reward Assignment via Credit Estimation for Long-Horizon Agents
TRACE addresses the challenge of assigning credit to individual actions in long-horizon multi-turn agents. It achieves this by representing agent rollouts as state transitions and …
SPyCE: Skill-Policy Co-evolution for Multimodal Agents
SPyCE proposes a novel framework for multimodal agents that learns reusable skills alongside the policy during reinforcement learning. Instead of treating trajectories as mere rewa…
CAVA: Canonical Action Verification and Attestation for Runtime Governance of Agentic AI Systems
CAVA addresses the challenge of governing agentic AI systems by creating a "runtime-semantics layer." Its core method is to convert diverse runtime records of agent actions into st…
The Town Square
Thinking Machines has released Inkling, an open-weights large language model designed for efficient deployment and customization, aiming to democratize access to advanced AI capabilities.
Workshops
Hallmark is a design skill that helps Claude Code, Cursor, and Codex avoid generating AI-generated code that is considered "slop" or low-quality.
This repository provides practical, real-world skills and code snippets for engineers, directly curated from the author's personal development environment.
