The Future of AI Agents Belongs to Amazon, Thanks to AGI SF Lab

Amazon has announced a new dedicated team, which forms what is called the ‘Amazon AGI SF Lab’, based in San Francisco, United States. The new team will work on developing highly capable AI agents that can ‘take actions in digital and physical worlds’. 

“Our work will build on that of Amazon’s broader AGI team, which recently introduced Amazon Nova, a new generation of state-of-the-art foundation models (FMs),” said Amazon in the official announcement. 

Interestingly, this new lab will be ‘seeded’ by the team hired by Adept, an AI startup whose founders were lured into Amazon earlier this year. 

Amazon also said that they’re keeping this team ‘lean’ to ensure each person receives a maximised amount of computing. 

“Our initial focus is on several key research bets that will enable AI agents to perform real-world actions, learn from human feedback, self-course-correct, and infer our goals,” read the announcement. 

Amazon also shared their excitement in ‘combining’ large language models with reinforcement learning to solve ‘solve reasoning and planning, learned world models, and generalising agents to physical environments’. 

“We’re entering an exciting new era where AI agents are the next playing field; the right research bets can reinvent what’s possible with AI,” added Amazon. 

Amazon also said they’re on the lookout for passionate and talented people to join their team, not just AI experts, ‘but also candidates from other disciplines who will bring fresh thinking to the field, such as physics, math, or quantitative finance, regardless of experience level’. 

Only a few days ago, Amazon announced a new family of foundational models, the Nova. At the AWS re:Invent conference in Las Vegas, Amazon announced the Nova Micro, Lite, Pro and Premier, available exclusively on Amazon Bedrock. 

The flagship Nova Pro modal offers performance parity with other industry-leading models like Claude 3.5 Sonnet and OpenAI’s GPT-4o. 

“They’re really cost-effective. They’re about 75% less expensive than the other leading models in Bedrock,” said Amazon’s CEO Andy Jassy.

The Nova models also possess multimodal read and write capabilities and can process text, image, and video inputs. 

“The models are integrated with the distillation feature, which allows you to infuse the intelligence of larger models into smaller models that are more cost-effective and offer lower latency,” Jassy added. 

Recently, Jeff Bezos announced his active return to Amazon and expressed his commitment to steering the company toward groundbreaking innovations. 

“My heart is in Amazon, my curiosity is Amazon, my fears are there, my love is there, I'm never going to forget about Amazon. I'll always be there to help,” Bezos stated in a recent interview. His involvement is expected to accelerate the AGI SF Lab’s ambitions, reinforcing Amazon’s position at the forefront of AI advancements. 

“We are literally working on a thousand applications internally. You have to remember modern AI is a horizontal enabling layer. It can be used to improve everything; it will be in everything. This is most like electricity,” said Bezos. 

Bezos also emphasised the importance of tackling large-scale challenges with a clear focus, stating, “Big leaders only have to do a few things. Big leaders have to identify the big ideas, they have to enforce tough execution, right against those big ideas, and they need to grow the next generation of leaders.”