AI Agents
AI Agent technology and applications
Articles
Perplexity Launches Brain: A Self-Improving Memory System Thatβ8
Perplexity launches Brain, a self-improving context graph memory system for AI agents that learns and optimizes overnight without human input.
OpenAI Releases LifeSciBench: A 750-Task Benchmark forβ6
OpenAIβs LifeSciBench is a 750-task benchmark evaluating AI on real-life science research with expert rubrics, where even top models pass only ~1 in 3 tasks.
NVIDIA SkillSpector Guide: Scanning AI Skills for Security Risksβ7
Learn how to use NVIDIA SkillSpector for AI skill security scanning via static analysis, SARIF export, custom analyzers, and LLM-based validation.
Vercel Launches Eve: An Open-Source AI Agent Framework Thatβ6
Vercel launches Eve, an open-source AI agent framework that treats each agent as a directory of capability-mapped filesβsimplifying development, testing, and sc...
MiniMax Sparse Attention (MSA): A Two-Branch Block-Sparseβ8
MiniMax Sparse Attention (MSA) uses a two-branch block-sparse mechanism to cut compute 80-90%, trained on a 109B MoE with 3T tokens for efficient long-context L...
OpenAIβs Deployment Simulation Extends Pre-Deployment Riskβ8
OpenAI has introduced a new pre-deployment safety method called Deployment Simulation. The approach is straightforward: before shipping a model, simulat...
How to Build Memory-Efficient Transformers with xFormers: Aβ8
Build memory-efficient Transformers with xFormers: packed sequences, GQA, ALiBi, SwiGLU & causal attention. Reduce GPU memory 30-50% and speed up long-context m...
Meet Qwen-RobotSuite: Three Embodied AI Models for VLAβ7
Qwen-RobotSuite: Three AI models for VLA manipulation, video world modeling, and navigation, built on a Qwen vision-language backbone for embodied robotics.
Hermes Agent Adds Asynchronous Subagents, So Delegated Work Noβ6
Hermes Agent update adds asynchronous subagents, enabling non-blocking delegated tasks so parent chat sessions continue without interruption.
Google Cloud Introduces Open Knowledge Format (OKF): Aβ8
Google Cloud launches OKF, a vendor-neutral Markdown spec standardizing curated context for AI agents to improve interoperability and reduce lock-in.
