AI
Anthropic Released A Breaking-Through Products: Computer Use
By Timmy
Anthropic has announced the public beta release of a groundbreaking capability: computer use for its AI model Claude 3.5 Sonnet. This innovative feature allows the AI to interact with computers just as humans do—viewing screens, moving cursors, clicking buttons, and typing text. The release of the capabilities also puts Anthropic to be forefront of AI once again.
YouTube Videos demonstrations:
1. For Automating Operations:
The video demonstrates Claude is able to control a virtual computer, such as identifying elements on the screen and moving the mouse to switch tabs in the browser to complete tasks.
2. For coding:
This feature is still in beta and is accessible in Claude 3.5 Sonnet via API. Although the technology is still experimental and prone to occasional errors. Anthropic mentioned that some tech companies such as Asana, Canva, DoorDash, and Replit are already exploring its potential and applying it to tasks that require multiple complex steps.
What makes this development unique is Anthropic's approach to teaching Claude general computer skills, rather than writing specific tools for individual tasks. This enables AI to leverage standard software and tools designed for human users, opening up possibilities for automation, software testing and research applications.
However, Anthropic acknowledges that the current implementation has limitations. Basic actions like scrolling, dragging, and zooming can still present challenges for the AI. The company has also implemented safety measures, including new classifiers to identify computer use and potential harmful applications, addressing concerns about misuse for spam, misinformation, or fraud. They emphasizes that this release is part of an ongoing development process, with rapid improvements expected in the coming months.
This advance represents an incremental improvement in the ability of artificial intelligence to interact naturally with computer systems, it may change the way AI handle complex digital tasks.
