Microsoft has announced the development of a new AI model called the Large Action Model (LAM), which is capable of independently running Windows programs and executing tasks autonomously, according to media reports.
The model enables the execution of commands in a practical, actionable manner. Unlike traditional language models, such as GPT-4, which are limited to processing and generating text, Microsoft’s new LAM model can transform user requests into tangible actions, such as launching programs or controlling devices.
While similar ideas have existed before, LAM is the first model specifically trained to work with Microsoft Office products and other Windows applications.
Among the tasks LAM can perform is directly completing actions, such navigating websites to finalize purchases—a step beyond the capabilities of traditional models, which are limited to providing text-based instructions for such tasks.
According to Microsoft, the development of LAM involved Task Planning Training, Learning from Advanced Models, Self-Exploration, and Reward-Based Training.
The company tested the model in a controlled environment using Microsoft Word, where it successfully completed tasks with a 71% success rate—outperforming GPT-4, which had achieved 63% without visual information. LAM was also faster, taking only 30 seconds to execute tasks.
Researchers consider LAM a major breakthrough in AI, suggesting it could pave the way for the development of Artificial General Intelligence (AGI). Instead of systems that only understand and generate text, companies may soon offer digital assistants capable of effectively carrying out daily tasks, transforming the role of AI in peoples’ everyday lives.