The Morning
From the arXiv
Co-ReAct: Rubrics as Step-Level Collaborators for ReAct Agents
o-ReAct introduces a novel framework where rubrics act as step-level collaborators for ReAct agents during inference. By injecting rubrics into the agent's context at each decision point, Co-ReAct provides explicit guidance on what to target in evidence seeking, reasoning, and self-evaluation, leading to more focused and effective multi-step reasoning trajectories. This approach moves beyond rubrics as mere evaluators to become active guides for agent actions.
DiLaDiff: Distilled Latent-Augmented Diffusion for Language Modeling
DiLaDiff addresses the token correlation issue in diffusion language models by introducing a continuous latent space. This latent space, learned via an auto-encoder and a latent diffusion model, captures semantic relationships. A consistency model then distill…
From Raw Experience to Skill Consumption: A Systematic Study of Model-Generated Agent Skills
This paper systematically studies the full lifecycle of model-generated skills for language agents, from experience generation to skill extraction and consumption. Their core contribution is a utility-grounded evaluation framework that reveals model-generated …
Goal-Conditioned Agents that Learn Everything All at Once
This paper introduces Learning Everything All at Once (LEO), a method for goal-conditioned reinforcement learning. LEO efficiently utilizes all transitions by performing off-policy updates for every possible goal simultaneously, overcoming the computational in…
It's the humans, not the data: Geopolitical bias in LLMs originates in post-training, amplified by the language of the prompt
This paper demonstrates that geopolitical bias in Large Language Models (LLMs) primarily emerges during the post-training (chat tuning) phase, not from the initial training data. The authors found that LLMs often develop a bias favoring the country of their de…
MemAudit: Post-hoc Auditing of Poisoned Agent Memory via Causal Attribution and Structural Anomaly Detection
MemAudit is a post-hoc framework for auditing poisoned memory in LLM agents. It uses causal attribution to identify memories that causally influence harmful outputs and structural …
Push Your Agent: Measuring and Enforcing Quantitative Goal Persistence in Long-Horizon LLM Agents
This paper introduces Quantitative Goal Persistence (QGP) to measure how well LLM agents complete tasks requiring a specific number of distinct items. Their benchmark, PushBench, u…
ARES: Automated Rubric Synthesis for Scalable LLM Reinforcement Learning
ARES automates the creation of question-answer pairs and corresponding weighted rubrics from raw text, enabling scalable reinforcement learning for LLMs on open-ended tasks. This f…
OpenSkillEval: Automatically Auditing the Open Skill Ecosystem for LLM Agents
This paper introduces OpenSkillEval, an automatic framework for evaluating LLM agents and the skills they use. It addresses the challenge of assessing the growing open-source skill…
AI Assurance: A Comprehensive Testing Strategy for Enterprise AI Systems
This paper proposes a new AI assurance strategy for enterprise AI systems, shifting from traditional correctness verification to continuous risk reduction. It emphasizes treating e…
The Town Square
Memory now accounts for almost two-thirds of the component costs in AI chips, significantly impacting overall expenses.
Workshops
This repository provides open-source plugins designed to enhance Claude Cowork for knowledge workers, enabling them to integrate external tools and data for improved productivity.
This repository provides a hands-on guide to building and deploying AI systems from the ground up, focusing on practical application and sharing knowledge.
