<?xml version="1.0" encoding="UTF-8"?><oembed><type>video</type><version>1.0</version><html>&lt;iframe src=&quot;https://www.loom.com/embed/9a8daee984694f1ea12f36a48b0725cd&quot; frameborder=&quot;0&quot; width=&quot;1110&quot; height=&quot;832&quot; webkitallowfullscreen mozallowfullscreen allowfullscreen&gt;&lt;/iframe&gt;</html><height>832</height><width>1110</width><provider_name>Loom</provider_name><provider_url>https://www.loom.com</provider_url><thumbnail_height>832</thumbnail_height><thumbnail_width>1110</thumbnail_width><thumbnail_url>https://cdn.loom.com/sessions/thumbnails/9a8daee984694f1ea12f36a48b0725cd-6b92c9474c0654ed.gif</thumbnail_url><duration>71.533</duration><title>WatchStandard Prevents Datacenter AI Chip Failures</title><description>This Loom presents WatchStandard, an AI system designed to prevent AI datacenter crashes. The team explains that without it, chip failures are only discovered after a chip dies, which can trigger costly damage and replacement orders too late. WatchStandard predicts risk from a real failing chip and shows that risk climbed 18 hours early. The system asks a person before acting and never acts on its own, ensuring timely human involvement before anything breaks.</description></oembed>