The Morning
From the arXiv
Efficient Reasoning Training Does Not Always Harm CoT Faithfulness and Monitorability
his paper investigates whether efficient Chain-of-Thought (CoT) training harms faithfulness and monitorability in large language models. The authors fine-tune models using three distinct length-pressure methods and evaluate their impact on how well the CoT reflects the model's decision-making process. Their contribution lies in demonstrating that efficient reasoning training does not *always* negatively affect CoT faithfulness, depending on the specific efficiency method and task.
FrugalEvo: Towards Cost-Aware LLM-Guided Program Evolution
FrugalEvo introduces a cost-aware evolutionary framework for LLM-guided program optimization. It leverages a two-tiered LLM approach, with a powerful LLM guiding strategy and a cheaper LLM implementing and refining code, to maximize performance gain per unit c…
HazardWeaver: Scientific Route Selection for Hazard Analysis Agents
HazardWeaver addresses the challenge of AI agents selecting appropriate scientific methods for hazard analysis by formulating it as state-dependent route selection. Its core method involves compiling evidence-linked conditions for scientific applicability and …
HyperBrowseComp: A Multilingual and Multimodal Stress Test for Web-Browsing Agents
HyperBrowseComp is a new benchmark designed to rigorously test web-browsing agents. Its core method involves creating extremely challenging, multilingual, and multimodal questions that require agents to locate obscure information, follow complex clue chains, a…
MobiAgent: Dual-Loop Recursive Policy Self-Improvement for Long-Horizon Mobile Manipulation
MobiAgent tackles long-horizon mobile manipulation by introducing a dual-loop framework for robust deployment and continuous self-improvement. The Inner Loop uses a Vision-Language Model to dynamically compose atomic skills for receding-horizon planning and er…
NeutronGym: Physics-Graded Neutron Instrument Design for LLM Agents
NeutronGym is a novel, executable environment designed to test whether Large Language Models (LLMs) can truly perform physics-based design rather than just recall information. It a…
IDRF: Inverse-Distilled Reward Fine-tuning of Masked Discrete Diffusion Models
This paper introduces IDRF, a framework for efficiently fine-tuning masked discrete diffusion models with rewards. IDRF replaces computationally expensive sequence-level KL diverge…
UniIntervene++: An Adaptive Intervention Agent for Efficient Real-World Reinforcement Learning
UniIntervene++ introduces an adaptive intervention agent for online reinforcement learning in robots. It learns to dynamically switch between autonomous execution and various assis…
CLIMB: Confidence-Guided Complementary Evidence for Multimodal Retrieval-Augmented Generation
CLIMB is a training-free framework for multimodal retrieval-augmented generation (RAG) that improves knowledge-intensive visual question answering. It first creates a diverse pool …
Passing the Test You Trained On: Re-evaluating Prompt-Injection Detectors for LLM Agents
This paper re-evaluates prompt-injection detectors used by LLM agents, finding that their performance on public benchmarks doesn't reliably predict their effectiveness within actua…
The Town Square
Formula 1 drivers expressed frustration and a sense of powerlessness due to a software glitch during the Bahrain Grand Prix that impacted their car's performance.
Workshops
Impeccable is a design language that enhances AI's design capabilities by providing a structured and opinionated framework for building better AI-driven designs.
This repository provides a comprehensive guide to marketing skills, including CRO, copywriting, SEO, analytics, and growth engineering, specifically tailored for Claude Code and AI agents.
