Anthropic, an artificial intelligence business backed by Alphabet and Amazon, unveiled significant changes to its AI models on Tuesday, including a new tool meant to automate hard computer jobs. This capability, known as “computer use,” enables AI to perform tasks such as mouse movement, clicking, and typing, all to free users from having to type manually.
Jared Kaplan, Anthropic’s Chief Science Officer, emphasized the potential of this feature. “The capability can tell AI where to move the mouse, where to click, what to type, to do quite complicated tasks,” he said during a recent interview. The capability is mainly targeted at software developers and marks a big step toward the development of AI agents—programs that can perform multi-step activities with minimum human involvement.
The inclusion of the “computer use” capability marks a departure from traditional chatbots, which are largely employed to generate text or code but cannot conduct actions. AI agents, on the other hand, can do more complex tasks without constant human supervision. Anthropic highlighted the feature’s capabilities by demonstrating how to create a rudimentary website and even arrange a morning expedition using Google Search and Apple Maps.
Anthropic’s AI models, collectively known as Claude, are available to developers at various price points according on performance level. The mid-tier Sonnet and entry-level Haiku models received improvements this week. The recently updated Haiku 3.5 model now produces computer code with the performance that Kaplan described. Anthropic CEO Dario Amodei stated that Opus, the company’s most advanced model, would be updated by the end of the year.
The new “computer use” capability is presently limited to the most recent version of Claude 3.5 Sonnet and comes with restrictions to avoid abuse in areas such as spam, fraud, and electoral tampering. Despite these protections, Kaplan noted that the AI component is still susceptible to error.The move follows Microsoft’s unveiling of a new tool that allows its clients to create custom AI agents for functions like as query resolution, sales lead identification, and inventory management.