The artificial intelligence company Anthropic has just announced the release of a new version of the Claude AI model, along with a beta test for "computer use" on October 22.
This beta version will allow developers to work with the API that allows Claude to control the mouse cursor, click on buttons and fields, and automatically insert text within the user's computer environment.
A demonstration video of Claude's computer usage capabilities shows the machine's ability to navigate multiple windows, open new tabs to search for information, and transfer data from one window to another.
Computer Control
Under the current model, large language models controlled by artificial intelligence like Anthropic's Claude and ChatGPT of OpenAI interact with users through a dedicated interface window. Some models may perform limited web searches within the confines of their interface, but external interactions are usually limited to allowing users to upload documents for the AI to process.
With Anthropic's new beta version of computer use, the company hopes that developers will ultimately be able to automate any function that a human sitting at a workstation can perform.
According to a post on the company's blog:
"Developers can direct Claude to use computers the way humans do—by looking at the screen, moving the cursor, clicking buttons, and typing text. Claude 3.5 Sonnet is the first pioneering AI model to provide computer usage capabilities in a public beta stage. At this stage, it is still experimental—sometimes clunky and prone to errors."
The computer usage beta is only available to developers through the Claude API; however, Anthropic says they are releasing it early to get feedback from developers. The company expects this feature to "improve rapidly over time".
Privacy and Security
Anthropic addressed some privacy and security concerns in the blog post, noting that "because computer usage could provide a new attack vector for familiar threats like spam, misinformation, or fraud," the company is taking a "proactive approach to promoting its safe deployment" and may detect when computer usage is being deployed inappropriately.
