TechyMag.com - is an online magazine where you can find news and updates on modern technologies


Back
Software

Anthropic's new AI software controls the computer instead of the user - how it works

Anthropic's new AI software controls the computer instead of the user - how it works
0 0 3 0

The company Anthropic has unveiled a new AI tool called Computer Use, alongside updated versions of its language models — Claude 3.5 Sonnet and Claude 3.5 Haiku. Computer Use can control the user's mouse cursor and perform basic tasks on the computer.

The updated version of Claude 3.5 Sonnet shows improvements across the board, particularly in programming, where the model was already an industry leader. Claude 3.5 Haiku matches the performance of Claude 3 Opus — the company's previous most powerful model — in many areas while maintaining the same cost and speed of operation.

Using Computer Use, the system executes multi-step instructions to interact with the computer: it observes the screen, moves the cursor, clicks buttons, and enters text. At this stage, the tool is only available through API and works with the Claude 3.5 Sonnet model.

For successful cursor management, Computer Use employs a special operational mechanism: the system takes sequential screenshots and calculates pixel counts for precise cursor movement. Its ability to accurately count pixels has become key for executing commands.

However, the current version of the tool has certain technical limitations. Because it operates with individual screenshots rather than a video stream, the system may miss brief notifications. Additionally, Computer Use has not yet mastered some common tasks, such as dragging objects.

Before the public launch, Anthropic engaged several powerful partners for testing. Companies like Amazon, Canva, Asana, and Notion have tried Computer Use, and their feedback has aided in refining the system before open beta testing.

There is intense competition in the development of such tools. Anthropic's main competitor, OpenAI, is also working on similar technology but is keeping it under wraps for now. Analysts predict that such tools will generate significant profits in the coming years.

Considering the potential risks, Anthropic has implemented safeguards against abuse. The company has integrated systems to detect suspicious activity and has set restrictions: Computer Use cannot interact with social media, register domains, or operate with government websites. Developers can join the public beta testing through Anthropic's official website.

Microsoft is also keeping pace with Anthropic, recently releasing a major update to its AI Copilot, adding voice communication and image recognition features.

Source: Arstechnika, Anthropic

Related tags:

Thanks, your opinion accepted.

Comments (0)

There are no comments for now

Leave a Comment:

To be able to leave a comment - you have to authorize on our website

Related Posts