News
AI Chatbots Handled Foreign Propaganda Questions Better Than Search Engines, NPR Test Finds
NPR and NewsGuard tested how AI chatbots handle foreign propaganda by posing 30 questions built from false narratives that originated…
News
Anthropic Warns of Infostealer Malware Hijacking Claude Sessions
Anthropic has alerted affected Claude users that infostealer malware on their computers has compromised login sessions, enabling attackers to access…
Opinions
Unsafe at Any Speed: AI Optimists Are Turning Cautious as Safety Concerns Mount
Prominent AI industry figures including Sam Altman, Dario Amodei and Bill Gates have shifted from touting AI’s transformative potential to…
Opinions
We Still Don’t Know How People Are Really Using AI
MIT Technology Review reports that researchers at Stanford and MIT have launched the AI Observatory, an independent platform analyzing real…
Engineering
Sliding-window beats linear attention
This paper shows that Sliding Window Attention (SWA) with attention sinks delivers comparable or superior performance to post-trained Linear Attention…
Engineering
LMSM: LLM Security Framework Inspired by Linux Security Modules
Researchers propose Language Model Security Modules (LMSM), a framework that adapts the architecture of Linux Security Modules to protect large…
Engineering
ContextPilot: Teaching Agents for Proactive Context Management via Fine-grained RL
ContextPilot addresses context management in long-horizon agent reasoning by giving agents tools beyond simple search and deletion, including planning, long-term…
Engineering
EvoUndo: Recoverability-Constrained Self-Evolution for LLM Agent Harnesses
EvoUndo is a framework for evaluating whether self-modifying LLM agents can safely reverse their own runtime changes across different system…
Engineering
StarHarness: Evolving Harnesses with Stratified Search for Enterprise Environments
StarHarness is a framework that optimizes the executable environment surrounding a fixed language model by evolving prompts, tool interfaces, skills,…
Engineering
DART-SD: Diamond-topology Aware Retrieval and Tuning for Self-Distillation of Multi-Turn Tool-Calling Agents
DART-SD addresses limitations in training multi-turn tool-calling agents by modeling task execution as an Interaction-State Transition Graph that captures the…
Engineering
LoopArena: Benchmarking Models as Runtime Controllers for Loop Engineering
LoopArena is a benchmark that evaluates how effectively one AI model can direct a separate coding agent through extended, multi-round…
Engineering
VLANeXt: A Simple and Research-Oriented Codebase for Robotics Research
VLANeXt is a research-oriented codebase that systematically explores design choices for vision-language-action models across more than 500 experiments spanning foundational…