Amazon Launches Nova Act, AI Body Outperforms Rivals by 6%

Amazon, a leading global technology giant, has made a significant stride into the AI intelligent body sector by launching Nova Act, an AI-powered intelligent body capable of performing web browser operations. The company also introduced the Nova Act SDK toolkit, designed to assist developers in building intelligent body prototypes. This move positions Amazon alongside other tech titans such as OpenAI, Google, Anthropic, and Manus AI, who have also been actively exploring the potential of AI intelligent bodies.

Nova Act is designed to allow AI agents to perform basic online operations, such as ordering food from a delivery platform or booking a table at a restaurant. Amazon claims that Nova Act outperforms competitors' products in several benchmark tests. For instance, in the ScreenSpot Web Text test, which measures an AI agent's ability to interact with screen text, Nova Act scored 94%, surpassing OpenAI's CUA model (88%) and Anthropic's Claude 3.7 Sonnet (90%). However, it is important to note that AI intelligent bodies are still in their early stages of development, and these test results should be considered as a reference for the actual utility of the product. Amazon has not disclosed Nova Act's performance in more common proxy evaluation standards, such as WebVoyager.

For a company with billions of global users, entering the intelligent body race is a significant move. Previous tests of products from OpenAI and Google have shown that current AI intelligent bodies have serious flaws, including slow responses, inability to operate independently for extended periods, and frequent low-level errors. These issues can significantly impact user experience.

As an internet giant, Amazon's entry into the AI race is relatively late. In June of last year, Amazon reached an agreement with AI startup Adept, gaining technology authorization and bringing on board Adept's co-founder and former OpenAI engineering vice president, David Luan, along with his team. In December, the company established the AGI Research Lab in San Francisco, co-led by David Luan and Pieter Abbeel, a professor of AI and robotics at the University of California, Berkeley. Amazon also released the Nova series of large models in December, fully entering the AI competition. Nova Act is the first product to be publicly released by the San Francisco lab.

When asked why the AGI lab is developing AI intelligent bodies, David Luan responded that he believes intelligent bodies are a key step in creating super-intelligent systems. He defined AGI as an AI system that can assist with all human operations on a computer. Luan also stated that the Nova Act SDK, designed by the San Francisco team, aims to reliably handle simple, short tasks and provide developers with tools to precisely define when human intervention is needed in the intelligent body workflow, thereby creating more reliable intelligent body applications.