Anthropic is transforming its chatbot into an AI agent that can work independently with a computer. The only input required from users is a question.
The latest update for Claude lets the chatbot perform tasks independently on a computer. To answer a question, Claude will then use information available online and in local files on the computer. It also has the ability to click buttons, making it possible for the AI tool to open programs or files on a computer for the user.
Anthropic updates the Claude 3.5 Sonnet model, which was released in June. This multimode model is known for efficient performance. The updated version achieves better performance on LLM benchmarks. Both reasoning and programming skills are better. It achieves better results on tests for multimodal models and can solve math problems better. This while the previous model was already praised for better results than GPT-4o, Gemini 1.5 Pro, and Llama-400b.
Source: Anthropic
Limitations
The new tool ‘computer use‘ has several limitations. First, it is only available in API for now, allowing only developers access. The company says this is a way to gather feedback on the tool so that its capabilities improve quickly.
In addition, the tool cannot be used to create social media messages, and it is not possible to run Claude on government websites. In this way, Anthropic is trying to ensure the tool’s security. Anthropic is the first in the AI field to launch an AI agent that can work completely independently of all computing.
Claude 3.5 Haiku
The tests also present the results of Claude 3.5 Haiku. The company indicates that this model matches the performance of Claude 3 Opus in most areas. Haiku is more interesting in terms of price than the previous generation of the top model.
Tip! Belgian AI expert ML6: ‘Our AI agents give hands and feet to LLMs’