Garman told WIRED ahead of the event that Amazon will also introduce a number of tools to help customers deal with generative AI models, which he says are often too expensive, unreliable and unpredictable.
These include a way to boost the capabilities of smaller models using larger models, a system to manage hundreds of different AI agents, and a tool that provides evidence that a chatbot's output is accurate. Amazon creates its own AI models to recommend products on its ecommerce platform and other functions, but it primarily serves as a platform to help other companies build their own AI programs.
Steven Dickens, CEO and principal analyst at Hyperframe Research, says that although Amazon does not have a ChatGPIT-type product to advertise its AI capabilities, the scope of its cloud services will give it an advantage in selling generative AI to others. “The extensibility of AWS—it's going to be an interesting thing,” he says.
Amazon's own line of chips will help make the AI software it sells more affordable. “Silicon is going to be a key part of any hyperscaler's strategy going forward,” Dickens says, referring to cloud providers that offer the hardware to build the largest, most capable AI. He also noted that Amazon has been developing its custom silicon for a longer time than competitors.
Garman says a growing number of AWS customers are now moving from demos to building commercially viable products and services incorporating generative AI. “One of the things we're really excited about is seeing customers move forward with their AI experiments and proof of concepts,” he told WIRED.
Garman says many customers are less interested in pushing the limits of generic AI than finding ways to make the technology cheaper and more reliable.
For example, a newly announced AWS service called Model Distillation can produce a smaller model that is faster and less expensive to run while still having the same capabilities as a larger model. “Let’s say you’re an insurance company,” says Garman. “You can take a whole set of questions, put them into a really advanced model, and then use that to train smaller models to become experts at those things.”
Another new cloud tool announced today, Bedrock Agents, can be used to create and manage so-called AI agents that automate useful tasks like customer support, order processing and analytics. It involves a master agent that will manage a team of AI subordinates, providing reports on how they are performing and coordinating changes. “You can basically create an agent that says you're the boss of all the other agents,” says Garman.