Anthropic unveils updated AI models with new ‘computer use’ feature for automating complex tasks

Anthropic, the artificial intelligence startup backed by Alphabet and Amazon, announced significant updates to its AI models on Tuesday, unveiling a new feature designed to automate complex computer tasks. This feature, known as “computer use,” allows AI to autonomously perform actions such as moving the mouse, clicking, and typing, all aimed at saving users from manual keystrokes.

Jared Kaplan, Chief Science Officer at Anthropic, highlighted the potential of this feature. “The capability can tell AI where to move the mouse, where to click, what to type, in order to do quite complicated tasks,” he explained in an interview. The feature is especially targeted toward software developers and marks a significant step toward the development of AI agents—programs that can perform multi-step actions with minimal human intervention.

The introduction of the “computer use” feature represents a leap beyond traditional chatbots, which are primarily used to generate text or code but are not equipped to perform actions. AI agents, on the other hand, can handle more sophisticated operations without constant human oversight. Anthropic demonstrated the feature’s potential by showcasing its ability to code a basic website and even plan a sunrise outing using tools like Google Search and Apple Maps.

Anthropic’s lineup of AI models, collectively known as Claude, is offered to developers at varying price points depending on the performance level. This week’s updates were rolled out to Sonnet, the mid-tier model, and Haiku, the entry-level model. The newly updated Haiku 3.5 model now generates computer code with performance that Kaplan described as “almost comparable” to the previous version of Sonnet, which was released in June.

Anthropic CEO Dario Amodei hinted that the company’s most advanced model, Opus, would receive an update by the end of the year.

The new “computer use” capability is currently limited to the latest version of Claude 3.5 Sonnet and comes equipped with safeguards to prevent misuse in areas like spam, fraud, and election interference. Despite these safeguards, Kaplan acknowledged that the AI feature is still prone to errors.

The development comes on the heels of Microsoft’s announcement of a new application allowing its clients to build custom AI agents that handle tasks like query resolution, sales lead identification, and inventory management.