Adept AI Labs discusses new foundation model at Oracle CloudWorld (venturebeat.com)

At Oracle CloudWorld 2022, Adept AI Labs unveiled ACT-1, a novel foundation model designed to enable human language-driven control over computer software. Unlike traditional foundation models that focus on language, images, or protein folding, ACT-1 targets the core challenge of computer interaction—translating human intent into software actions. Adept aims to create an intelligent overlay that allows users to bypass complex graphical interfaces and command their computers simply by describing tasks in natural language, vastly simplifying and accelerating how people use digital tools. This approach leverages transformers trained on vast unlabeled data to build a broadly capable model that can execute actions across diverse software environments with minimal task-specific fine-tuning. By focusing on actions rather than static outputs, Adept is pushing the foundation model paradigm into uncharted territory, potentially redefining human-computer interfaces. The startup’s ability to train ACT-1 at scale hinges on its collaboration with Oracle Cloud Infrastructure, which provides thousands of interconnected GPUs crucial for massive computation and data processing. The initiative highlights how the next wave of foundation models could revolutionize enterprise productivity and everyday computing by making interactions more intuitive, creative, and efficient.