Privacy-first AI training now runs 81% faster on edge devices
A breakthrough from MIT accelerates federated learning, making secure AI training feasible on everyday gadgets like smartwatches and sensors while cutting memory use by 80%.
A breakthrough from MIT accelerates federated learning, making secure AI training feasible on everyday gadgets like smartwatches and sensors while cutting memory use by 80%.
With shelves overflowing with smart glasses from Rokid, Meta, and others, the real question isn’t which model fits best—it’s what you actually do with them every day.
A 26-million-parameter AI model delivers real-time function calling at thousands of tokens per second on smartphones and wearables. Its creators argue that massive LLMs are unnecessary for basic tool use.
Google’s annual developer conference returns with a focus on AI-driven search, autonomous agents, and coding assistance. Here’s what analysts and leaks suggest we’ll see this year.