Chat AI 'Claude' will have the ability to automatically control PCs, and an improved version of 'Claude 3.5 Sonnet' and a lightweight model 'Claude 3.5 Haiku' will also be released



AI company Anthropic has announced an improved version of its AI model ' Claude 3.5 Sonnet ' and a new lightweight and high-performance model ' Claude 3.5 Haiku .' At the same time, public beta testing of a feature called ' computer use ' that allows Claude to operate a PC has also begun.

Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku \ Anthropic

https://www.anthropic.com/news/3-5-models-and-computer-use

Developing a computer use model \ Anthropic
https://www.anthropic.com/research/developing-computer-use

◆Claude 3.5 Improved version of Sonnet
The improved version of Claude 3.5 Sonnet has improved performance in all aspects compared to the previous version, and outperforms competing models such as GPT-4o and Gemini 1.5 Pro in most tests except for mathematical performance. In addition, although not listed in the table, it seems to have outperformed OpenAI o1 in coding ability.



◆Claude 3.5 Haiku
Claude 3.5 Haiku is a model that combines low system load and high performance, and has performance that far exceeds the previous generation model, Claude 3 Haiku. Claude 3.5 Haiku is also characterized by its high coding ability.



◆PC operation function 'computer use'
'Computer use' is a function that performs click operations and keyboard input according to the user's instructions. The user only needs to specify the content of the task, such as 'Make plans to observe the sunrise,' and does not need to specify the app to use.

For example, in the video below, Claude searches for a spot to watch the sunrise, finds the travel time to the spot, and enters the event into the calendar.

Claude | Computer use for orchestrating tasks - YouTube


I typed this into Claude: 'My friend is coming to San Francisco and I want to watch the sunrise from the Golden Gate Bridge. We'll start from Pacific Heights. Find a nice spot, find out the sunrise time and travel time to the spot, and add it to my calendar so that I can make it in time for sunrise.'



Then, Claude will launch Chrome and search for 'Golden Gate Bridge sunrise viewing spot' on Google.



In addition, we opened a map app and investigated the travel time to the observation spots found through Google search.



Finally, I opened the calendar app and scheduled an appointment to watch the sunrise. The memo field included the departure time, destination, and other details.



Claude can also access web apps within the PC and perform operations. In the example below, you can see 'Claude accesses Claude to generate website code.'

Claude | Computer use for coding - YouTube


At the time of writing, 'computer use' is a public beta version, and it often behaves unnaturally. Anthropic says, 'We expect the performance of' computer use 'to improve rapidly in the coming months.'

in Software,   Web Application,   Video, Posted by log1o_hf