RE: LeoThread 2024-10-22 21:22

[taskmaster4450le]

Anthropic's new AI model can control your PC

Anthropic has released an updated version of its Claude 3.5 Sonnet model with a new Computer Use feature that can interact with apps on a PC.

In a pitch to investors last spring, Anthropic said it intended to build AI to power virtual assistants that could perform research, answer emails, and handle other back-office jobs on their own. The company referred to this as a “next-gen algorithm for AI self-teaching” — one it believed that could, if all goes according to plan, automate large portions of the economy someday.

#ai #technology #anthropic #technology

Comments

[taskmaster4450le]

Anthropic Unveils Major Update: Claude 3.5 Sonnet Gets Desktop Control Capabilities

Anthropic has taken a significant step toward its vision of AI-powered virtual assistants with the release of an upgraded Claude 3.5 Sonnet model that can nOW interact with desktop applications. This development, announced on Tuesday, introduces a new "Computer Use" API that allows the AI to emulate human-like computer interactions through keystrokes, mouse movements, and button clicks.

[taskmaster4450le]

Key Features and Capabilities

The new Computer Use API, currently in open beta, enables Claude to:

• Interpret screen contents through screenshots
• Calculate precise cursor movements for navigation
• Access and utilize any website or application
• Self-correct and retry tasks when encountering obstacles
• Handle complex, multi-step processes

[taskmaster4450le]

The feature is accessible through Anthropic's API, Amazon Bedrock, and Google Cloud's Vertex AI platform.

Performance and Limitations

While the update represents a significant advancement, Anthropic is transparent about its current limitations:

• Success rates below 50% for airline booking tasks
• Approximately 33% failure rate for basic consumer tasks like initiating returns
• Challenges with scrolling and zooming
• Difficulty capturing brief notifications or actions
• Overall slow and sometimes error-prone performance

[taskmaster4450le]

Safety Measures and Concerns

Anthropic has implemented several safety features to address potential risks:

• No training on user screenshots or prompts
• Restricted web access during training
• Built-in classifiers to prevent high-risk actions
• 30-day retention of screenshots for safety monitoring
• Ability to restrict access to specific websites and features
• Pre-deployment testing by U.S. and U.K. AI Safety Institutes

[taskmaster4450le]

Additional Updates: Claude 3.5 Haiku

Anthropic also announced an upcoming Claude 3.5 Haiku model, promising:

• Performance matching Claude 3 Opus on certain benchmarks
• Maintained efficiency and cost-effectiveness
• Initial text-only release, with multimodal capabilities planned
• Optimized for user-facing products and specialized tasks

[taskmaster4450le]

Industry Context

This release positions Anthropic in the growing AI agent market, competing with:

• Traditional RPA vendors
• Newer entrants like Relay and Induced AI
• Major tech companies including Salesforce, Microsoft, and OpenAi

[taskmaster4450le]

According to Capgemini, 10% of organizations currently use AI agents, with 82% planning to integrate them within three years.

[taskmaster4450le]

Looking Forward

While Anthropic recommends starting with low-risk tasks, this update represents a significant step toward their goal of automating back-office work. The company has confirmed that Claude 3.5 Opus is in development, suggesting continued evolution of their AI capabilities.

This development marks a crucial milestone in AI automation, though Anthropic emphasizes the importance of careful implementation and appropriate precautions when dealing with sensitive data.

[taskmaster4450le]

Article