computer-using-world-model

Enable AI agents to safely explore action outcomes before execution by predicting UI state changes in desktop applications. Two-stage approach: first predict textual description of what changes, then synthesize visual representation of resulting screen. Allows agents to compare multiple candidate actions without risky trial-and-error, trained on Microsoft O…

Independent GitHub snapshot · Automated classification · Original documentation governs