This week, Anthropic announced a major update to its flagship language model: Claudius 3.5. This model, first presented last March, represents a big step forward in the capabilities of artificial intelligence, standing out for its unique balance between speed and efficiency, especially for businesses.

Innovative features for efficient AI

With the previous version, Claude 3.0, users already had access to solid performance at a competitive cost. However, Claude 3.5 takes a new step by introducing a revolutionary feature: “computer use”. This feature allows artificial intelligence to interact directly with a computer’s interface, performing tasks such as:

• Moving the cursor on the screen
• Click on specific locations
• Enter information using a virtual keyboard

The evolution of artificial intelligence

This development raises questions about the future of AI. Although the ability for AI to autonomously interact with software may seem daunting, the Anthropic team is convinced of the value of this innovation. They say this skill, currently in public beta, opens the door to new applications that were previously unimaginable.

Claude has become able to interpret what is happening on the screen and then use the software tools to accomplish various tasks. For example, when a developer asks him to use a certain piece of software, Claude can look at screenshots and count the pixels needed to move the cursor to the right place. This precision is essential to ensure that Claude can give commands to the mouse with the same ease as a human.

Increased attention to security

An important concern when training AI models is the safety. Anthropic ensured that Claude did not have access to the Internet during this training phase, thereby minimizing the risk of exposure to unsecured content.

Promising tests for AI

Preliminary beta testing demonstrates that Claude can self-correct and retry tasks when obstacles appear. This ability is explained by the “flipbook” nature of his vision: he can accumulate screenshots, which allows him to process condensed information rather than following a continuous video stream. However, this can also lead to gaps, such as omitting short-term actions or notifications.

Starting today, Claude 3.5 Sonnet is available to all users. Developers can now work on its beta version via the Anthropic API, as well as on platforms like Amazon Bedrock And Vertex AI of Google Cloud.

Claude Haiku: a welcome upgrade

Another innovation accompanying Claude 3.5 is the 3.5 Haiku model. This model is referred to as “the fastest”, promising exceptional performance in coding tasks, directly competing with models like GPT-4. While version 3.0 already stood out for its responsiveness, Haiku promises to be a tool for creating fluid and natural AI interactions.

Among its applications, Haiku is useful in:

• Content moderation, detecting risky behavior
• Economic management, such as logistics optimization
• Extracting information from unstructured data

The Claude 3.5 Haiku template is scheduled to launch later this month, first in text version, with image input expected soon.

In conclusion

Claude 3.5 embodies a significant advancement in the field of artificial intelligence, offering features that redefine our interaction with technology. The combination of direct interaction capabilities and improved performance positions this AI as an innovative solution for the professional world and beyond.