research.
73 writings found
Latest Archives
AgentHands Shows AI Needs Bodies, Not Just Brains
Google's new XR research demonstrates how synchronized hand gestures transform AI assistance from abstract to embodied, spatial, and genuinely useful for real-world tasks.
Transferring GPU Expertise Across Hardware with AI-Driven Kernel Search
How evolutionary search and CUDA-to-MLX translation layers enable automatic GPU kernel optimization for Apple Silicon, bridging decades of NVIDIA expertise.
Game Worlds as AI Laboratories: What DeepMind's SIMA Means for Developers
DeepMind's new partnership with game studios to develop general-purpose AI agents signals a major shift in how games and AI will coevolve.
Transferring GPU Expertise Across Hardware with AI-Driven Kernel Search
How K-Search uses LLMs and structured translation to port decades of CUDA optimization knowledge to Apple Silicon without rewriting from scratch.
Gemini 3.7 Flash: The Workhorse Model That Changes Agent Economics
Google's new Gemini 3.7 Flash delivers major coding improvements and 50% lower pricing. What it means for building production AI agents.
PhotoScan Shows How Smartphone AI Could Replace Clinical Imaging
Google researchers demonstrate deep learning can predict insulin resistance from phone photos with near-DXA accuracy, raising questions about the future of clinical diagnostics.
Gemini 3.7 Flash: The Coding Model That Actually Listens
Google's latest Flash model cuts costs in half while dramatically improving code generation, reasoning, and developer experience. What it means for your stack.
Transferring GPU Expertise Across Hardware: K-Search and the MLX Frontier
How AI-driven kernel search bridges CUDA expertise to Apple Silicon, enabling near-expert performance without rebuilding optimization knowledge from scratch.
Cross-Platform GPU Kernels: When AI Learns to Translate Optimization
How evolutionary kernel search with structured translation layers can port decades of CUDA expertise to Apple Silicon without rebuilding from scratch.
Gemini 3.7 Flash: The Coding Model That Actually Listens
Google's new Gemini 3.7 Flash brings 43% better code accuracy and cuts costs in half. What this means for your AI agent stack.
Gemini 3.7 Flash: The Coding Model That Actually Ships
Google's latest Gemini 3.7 Flash delivers major coding improvements and 50% cheaper pricing. Here's what it means for developers building production agents.
Transferring GPU Expertise Across Hardware with AI
How evolutionary kernel search bridges CUDA knowledge to Apple Silicon, reducing optimization work from months to minutes using structured translation.
Transferring GPU Expertise Across Hardware With AI-Driven Kernel Search
How evolutionary AI can translate CUDA kernel knowledge to Apple Silicon and beyond, breaking the cycle of rediscovering optimizations for each new chip.
LLMs Know More Than They Can Access: Why Recall Is the Real Bottleneck
Google Research reveals frontier LLMs encode 95-98% of facts but fail to recall 26-34% of them. The factuality problem isn't knowledge gaps, it's knowledge accessibility.
Transferring GPU Expertise Across Hardware with AI-Driven Kernel Search
How evolutionary search and structured translation layers enable CUDA optimization knowledge to transfer to Apple Silicon, unlocking near-expert performance without rebuilding from scratch.