Anthropic Unveils Claude 3.5 Sonnet: Can It Control Users' Computers? This refined title maintains the original meaning while enhancing clarity and engagement.

Anthropic has officially launched its latest AI models, the upgraded Claude 3.5 Sonnet and the innovative Claude 3.5 Haiku. These enhancements not only significantly improve programming capabilities but also introduce a groundbreaking "computer use" feature, enabling the AI to mimic human computer operations and execute complex tasks.

The public can now access the beta version of the "computer use" feature in Claude 3.5 Sonnet. This model can follow user instructions to move the cursor on the computer screen, click relevant locations, and input information using a virtual keyboard, ensuring a seamless interaction that mimics human operations.

Jared Kaplan, Chief Scientist at Anthropic, stated that this innovation marks the beginning of a new era where AI models can use various tools to complete tasks like humans. The release of Claude 3.5 Sonnet signifies a significant advancement for Anthropic in the commercial AI sector, aiming to create a true "AI agent."

An "AI agent" refers to an AI model capable of using software and performing computer tasks as a human would. Anthropic envisions its AI agent as a versatile tool that can browse the web and utilize any website or application.

During demonstrations, Claude showcased its impressive capabilities. For instance, when planning a sunrise viewing trip at the Golden Gate Bridge, the AI effectively opened a web browser, searched for viewing locations, and added the itinerary to a calendar. However, while the performance was commendable, there are still areas needing refinement.

Despite the promising outlook for AI agent technology, challenges regarding reliability, particularly in coding, persist. Additionally, AI agents pose security risks, as users are understandably concerned about granting these technologies access to their personal computer files and web browsers.

In response, Anthropic aims to enhance AI agent safety by gradually rolling out limited, relatively secure AI models to address these potential issues.