Tag
84 articles
Learn about SWE-2, a new AI coding model developed by Cognition that performs nearly as well as top models but at a much lower cost, thanks to post-training with Kimi K3.
Google's open-sourced Mantis toolkit enables AI coding agents to autonomously find, reproduce, and patch software vulnerabilities across different programming environments. This represents a significant advancement in AI-driven cybersecurity automation.
CUA-Lite is a new open platform from UC Berkeley that unifies sandboxes, data, evaluation, and reinforcement learning for computer-use agents, enabling more efficient and standardized AI agent development.
This article explains the technical concepts behind Hugging Face's Microduck robot, including reinforcement learning, sim-to-real pipelines, and the role of MuJoCo and ONNX in robotics AI.
Learn what IBM's new Granite 4.2 AI models are, how they work, and why they matter for real-world tasks like coding and problem-solving.
This explainer explores agentic AI - autonomous systems capable of planning, executing, and adapting to complex environments. Learn how these multi-agent systems work, why they matter for businesses, and why only 15% of organizations have achieved scaled adoption.
This explainer explores how Linkdaze's smart calendar uses advanced AI techniques like reinforcement learning, neural networks, and constraint satisfaction to manage household tasks beyond simple scheduling.
This article explores Richard Sutton's critique of synthetic data in AI, arguing that the infinite complexity of the real world cannot be adequately captured by artificial datasets. It examines the Big World Hypothesis and proposes continual learning agents as a more scalable alternative.
This article explains the advanced AI concept of in-situ learning, where robots can learn and adapt in real-time while operating in their environment, enabling them to use novel objects as tools without pre-programmed instructions.
This article explains CUDA Agent, a reinforcement learning system that uses large language models to generate optimized GPU kernels, outperforming traditional compilers in execution speed and efficiency.
This explainer examines the technical challenges behind Mark Zuckerberg's AI vision, focusing on AI alignment, interpretability, and trust mechanisms that affect market adoption.
World Labs introduces R2S2R, a simulation engine that trains robot controllers entirely in virtual environments, generating thousands of variations from a single real-world task for robust AI deployment.
