All Articles
16256 articles total
Agent Team Work Zone: An Automated, Persistent Workspace for Long-Lived Coding Agent Teams
Proposes ATWZ (Agent Team Work Zone), a filesystem-based layer for Claude Code that preserves agent working states and reduces prompt overhead for long-lived coding teams.
SAGE: Safety-First Defense-in-Depth Guardrails for Verified Lifecycle Control of High-Impact Generative AI
Introduces SAGE, a safety-first architecture for high-impact generative AI that uses authorization-separated controls to prevent catastrophic misuse.
Design Theater: A Benchmark for Generative UI
Introduces 'Design Theater', a benchmark for Generative UI tools that reveals a disconnect between the design rationales these tools claim and their actual output.
Let AI Agents Translate Networks, Not Reason About Them
Proposes TypoNet, a system that uses LLMs to translate network configurations into formal symbolic models for reliable verification and fault localization.
Share No More Than the Request Requires: Federated Disclosure for Perspective-Aware AI
Presents Provenance Preserving Chronicles (PPC), a federated protocol for secure, minimum-necessary disclosure of personal data for perspective-aware AI.
LearnVector – Andrew Ng's AI company building one‑to‑one learning experiences
Andrew Ng's new AI company, LearnVector, focuses on creating personalized, one-to-one learning experiences powered by AI.
Truth is not a direction: a Tarski attack on LLM probes
A technical analysis exploring a 'Tarski attack' on LLM probes to demonstrate that truthfulness in LLMs is not a simple directional vector in latent space.
Test-Time Coverage: Test-Conditioned Data Curation for Deployment-Aware Learning
Introduces TTCov, a data curation method that optimizes training data to match the deployment distribution using a task Atlas, specifically applied to autonomous driving.
Similarity All The Way Up: Multilingual Generalization in LLMs Relies on Language-Level Similarity Structures
Research demonstrating that LLM multilingual generalization depends on how well the model's latent representations recover the hierarchical similarity structure of language families.
RoleMix: Unifying Sequential and Non-Sequential Features via Semantic Tokenization for Post-Click Conversion Rate Prediction
Proposes RoleMix, a unified interaction architecture for post-click conversion rate prediction that uses semantic tokenization to unify sequential and non-sequential features.
MPR-CiteG: Enhancing RAG with Multi-Portfolio Retrieval and Citation-Grounded Generation
Presents MPR-CiteG, a framework for RAG that improves retrieval diversity and ensures generated outputs are explicitly attributed to sources to reduce hallucinations.
SEGRA: Structured Experience-Guided Graph Reasoning Agent for Gremlin Based Question Answering
Introduces SEGRA, an experience-guided agent that transforms natural language questions into Gremlin graph queries for enterprise IT support knowledge graphs.
Spatial Reasoning in LLM Game Agents: Impact of Causal Context and Multi-Step Planning
Evaluates the impact of causal context and multi-step planning on the spatial reasoning capabilities of LLM-based game agents using the Qwen3 model family.
Commitment To Cooperation With Self-Negotiated Contracts
Explores the use of self-negotiated contracts to foster cooperation between autonomous AI agents in a spatial-temporal game environment.
Disentangling Multi-View Scanning in Mamba for Network Traffic Anomaly Detection
Proposes DisenMamba, a framework that disentangles view-invariant and view-specific components in Mamba-based network traffic anomaly detection to reduce redundancy.
Show HN: Learning Rust by writing a Markdown to HTML compiler
A project demonstrating how to learn the Rust programming language by building a Markdown to HTML compiler.
Industry Brief: Private 5G for Manufacturing and Industrial Sites [pdf]
An industry brief discussing the implementation and utility of private 5G networks for manufacturing and industrial environments.
Imprompt: A Language Framework for Prompt Programming
Introduction of Imprompt, a language framework designed to treat prompts as programmable control surfaces for complex LLM tasks.
Co-Harness: Co-Evolving Harnesses and Model Weights for LLM Agents
Co-Harness is a framework that jointly optimizes LLM agent parameters and their runtime harness (prompts, tools, memory) to improve performance.
Beyond Sequential Interaction: Benchmarking Parallel Execution and Coordination for GUI Agents
Introduction of ParaGUIBench for benchmarking parallel GUI agent coordination and ParaGUI, a planner-worker agent that improves efficiency over serial execution.