Anthropic Unveils AI Models with Groundbreaking Computer Task Automation

Dario Amodei, Anthropic Co-Founder & CEO

Anthropic, an AI startup backed by Alphabet and Amazon, has unveiled two updated AI models, introducing a new feature designed to streamline complex computer tasks. The "computer use" feature enables the AI to autonomously control a computer, executing tasks such as moving the mouse, typing, and navigating applications. This innovation, aimed at software developers, marks a step toward creating AI agents that require minimal human input to complete multi-step operations.

Anthropic showcased the feature by coding a basic website and organizing an outing using programs like Google Search and Apple Maps. The company offers three versions of its Claude AI models, with recent updates focusing on the mid-tier Sonnet and entry-level Haiku models. The 3.5 version of Haiku is now nearly as adept at generating code as the more advanced Sonnet model, released in June. Anthropic plans to release an update for its top-tier Opus model by the year's end.

The new "computer use" capability is currently available in Claude 3.5 Sonnet and is designed with safeguards to prevent misuse, including spam, fraud, and election tampering. Despite the advancements, Chief Science Officer Jared Kaplan acknowledged the AI still makes occasional mistakes.

Source: Kenrick Cai and Jeffrey Dastin / Reuters