https://www.geeknetic.es/Noticia/36710/Microsoft-lanza-Fara-7B-un-modelo-de-IA-capaz-de-controlar-tu-raton-y-teclado-para-realizar-por-ti-las-tareas-mas-aburridas.html
Microsoft has taken a very relevant step beyond conversational chatbots with the presentation of Fara-7B. It is not a tool designed to generate text or answer questions, but rather a Small Language Model (SLM) created specifically to take physical control of the computer interface. Its main function is to act as an autonomous agent, similar to Kimi’s OK Computer or Google’s Computer Use, which manages the mouse and keyboard to complete tasks on behalf of the user.
What differentiates this technology is its way of working. Fara-7B analyzes the screen through screenshots, “seeing” the web page or application just as a human eye would, without relying on hidden accessibility codes. The system predicts coordinates for clicking, scrolling, or typing text. By having only 7 billion parametersthe model is light enough to run directly on the devicewhich reduces latency and ensures that user data remains local, improving privacy.
Efficient performance designed for web automation
Despite its compact size, performance tests indicate that Fara-7B is highly competitive. According to the technical data revealed, the model has managed to outperform much larger systems in specific benchmarks, including GPT-4owhen it comes to navigating interfaces. The training has been carried out using a synthetic data stream that imitates real human interactions, allowing the AI to learn to perform complex actions such as booking trips, filling out forms or comparing prices between different online stores.
However, Microsoft emphasizes that it is a experimental release intended for research and development. Aware of the risks of an AI controlling the PC, the developers have implemented a security system based on “Critical Points”. This feature automatically stops execution and requests the user’s explicit consent before performing any sensitive or irreversible action, such as sending an email or confirming a purchase.
Fara-7B is currently available under an MIT open license on platforms such as Hugging Face and Microsoft Foundry. In addition, the company has provided an optimized version for new Copilot+ PCs with Windows 11allowing the technology community to begin experimenting with the creation of agents capable of automating the daily digital routine. Now, it remains to be seen to what extent its implementation in real scenarios can be considered a real success.
