Research to Reality with Google DeepMind — Paige Bailey, Google DeepMind

תקציר מקורי באנגליתA model running entirely in the browser produces an emoji table ranking the Harry Potter books, and Paige Bailey immediately questions its taste. The response arrives almost instantly, without an API call or data leaving the machine. Bailey uses that demo to introduce Gemma 4, Google DeepMind's open model family, and show what local inference looks like in practice. She describes Apache 2 licensing, models ranging from two billion to 31 billion parameters, and the ability to download, adapt, and finetune them for a company's own projects. The opening also connects DeepMind's research mission to AlphaFold, medical models, robotics, and AI for science, before outlining computer use and managed agents that execute tasks in a Linux sandbox. The same local approach extends to phones through Goo