The Morning
From the arXiv
Agents in the Wild: Where Research Meets Deployment
his paper bridges the gap between LLM agent research and real-world deployment. Its core method involves analyzing practical challenges like robustness and safety encountered during deployment, drawing on case studies in pharmaceuticals and finance. The key contribution is identifying successful design patterns and mitigation strategies for failure modes, offering a roadmap for building reliable agentic systems.
Beyond Score Prediction: LLM-Based Essay Scoring and Feedback Generation via Reinforcement Learning with Rubric Rewards
This paper introduces RLAES, a novel framework that uses reinforcement learning to jointly optimize essay scoring and feedback generation in LLMs. Its core contribution is the Rubric-based Feedback Evaluation (RFE) system, which uses an LLM-as-judge and fine-g…
Copy Less, Ground More: Overcoming Repetitive Copying in Long-Context Reasoning via Evidence-Aware Reinforcement Learning
This paper addresses the problem of repetitive copying in long-context reasoning by Large Language Models. Their core method, GEAR, uses reinforcement learning to reward models not only for correct answers but also for grounding their reasoning in relevant evi…
Prompt Design at Scale: How Format, Instruction Count, and Context Length Shape Instruction Adherence and Hallucination in Large Language Models
This paper investigates how prompt design choices impact large language model performance. Through controlled experiments, it reveals that the number of instructions, their format, and context length significantly affect instruction adherence and hallucination…
AdaFlash: Adaptive Speculative Decoding via On-Policy Distilled Diffusion Drafters
AdaFlash accelerates large language model inference by improving speculative decoding with diffusion drafters. It addresses the high variance in diffusion drafters' draft quality by introducing an adaptive framework that uses on-policy distillation to create s…
ABot-World-0: Infinite Interactive World Rollout on a Single Desktop GPU
This paper introduces ABot-World-0, an action-conditioned video world model enabling real-time, long-horizon agent interaction. Its core method involves a novel training pipeline t…
Agentic Real2Sim: Physics-based World Modeling with Vision-Language Agents
This paper introduces Agentic Real2Sim, a framework that automates the creation of physics-based simulations from real-world recordings of object-robot interactions. It leverages v…
Athena-Brain Technical Report: An Efficient Robot Brain for General Intelligence and Embodied Interactio
This paper introduces Athena-Brain-8B, an 8-billion parameter LLM designed as an efficient on-device "brain" for embodied AI. Its core method involves a multi-stage post-training p…
AutoJourn: Multi-Perspective Summarisation, Bias Detection and Bias Neutralisation for LLM-Generated News in Automated Journalism
AutoJourn is a system for automated journalism that addresses bias in LLM-generated news. Its core method involves extracting diverse perspectives from social media, generating bal…
CodeRescue: Budget-Calibrated Recovery Routing for Coding Agents
This paper introduces CodeRescue, a method for coding agents to intelligently decide whether to retry with a cheaper model or escalate to a more expensive one after a failure. It t…
The Town Square
Google announced Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber, offering faster, more efficient AI models for developers with enhanced capabilities.
Workshops
This repository is the open-source hub for the book "Deep Understanding of AI Agents: Design Principles and Engineering Practice," providing the full text, compiled PDF, and chapter-specific code.
This repository builds a local-first code intelligence graph that creates a persistent map of your codebase, enabling AI coding tools to efficiently access relevant context for code reviews and large-repo workflows.
