Amazon’s AGI Lab Reveals Its First Work: Advanced AI Agents

Amazon is still seen as Legard in the competition of advanced development Artificial intellectHowever, it has quietly created a lab that is now setting a record for AI performance. Amazon AGI SF LabWhich is located in San Francisco and dedicated to the building General intellectOr AI that exceeds the capacity of the people, today published the first fruits of its work: a new AI model is capable of strengthening some advanced AI agents available anywhere.

The new model known as the Amazon Nova Act comes out of the Open AEE and ethnographic in several criteria designed to determine the intelligence and trends of AI agents. Benchmarks Grounds Web and Screenspot, Amazon Nova Act Clald 3.7 Sonnet and Openi Computer perform better than use agents. A major part of Amazon’s plan to compete in the AI ​​market is to concentrate on building agents and reflect the new model skills that reflect an attempt to create a generation equipment that can be best measured by the best available.

“I believe that the initial nuclear unit of computing in the future is about to be a monster call [AI] Agent, “David Luan said, who led Amazon’s AGI SF Lab.

Most of the top AI labs have been focused now AI agent capable of building the buildingThe In addition to the conversation of AI to independent activities, the technology promises to make the technology more useful and valuable. However, a lot of work is still underway to move from chat to action.

In the past six months OpenAE, anthropologists, Google and others have displayed Web browsing agent It took action as a response to a prompt. However, in most cases, these agents are still incredible, and they can easily be tried through open requests.

Luan says that Amazon’s goal is creating AI agents, which are more reliable. The thing that holds agents is “more cool demos of interesting skills that work per cent of the period, it is not a Wimo problem,” he said, “he says how self-driving cars need to be trained to deal with abnormal end cases before going down on an unexpected street.

Many so-called agents are created by combining large language models with multiple human-written rules that are designed to prevent them from stopping their courses, but make their behavior brittle. The Amazon Nova Act is a version of the company’s most powerful homegrown model Amazon Nova What steps it should be taken and at any time has received additional training to help decide. In general, Luan says AI models fight to decide when they should interfere with any work.

To improve Nova’s agential skills, using Amazon ReinforceA method that helped others AI models mimic better argumentThe