Amazon announces new slate of AI models

By Greg Bensinger

 

LAS VEGAS (Reuters) - Amazon announced a new slate of artificial intelligence platforms, known as foundation models, at its annual AWS conference, allowing for text, image and video generation, among other things.

 

The new offerings pit it against rivals like Adobe and Meta, which are racing to serve customers who want to automate more of their services.

 

Amazon CEO Andy Jassy announced the new "Nova" models at a conference in Las Vegas on Tuesday.

 

Developers, Jassy said, had a list of desires, prompting the new services.

 

"They want better latency. They want lower cost. They want the ability to do fine-tuning," Jassy said.

 

The announcements mark Amazon’s biggest step toward combating a reputation that it had been caught flat-footed in developing AI applications as competitors sped ahead.

 

Rohit Prasad, Amazon’s head of artificial general intelligence, said the company would compete with rivals on price and capabilities, highlighting what he said were faster speeds for the new models. "If I have something better to offer, then customers will come and use it," he said in an interview. 

 

"It is still very early" in the development of AI, and Amazon has an opportunity to take a lead as a result, he said.

 

Video generation from a single image or text prompt has been particularly hot, with Adobe, Meta, OpenAI and TikTok parent ByteDance, among others, all announcing new such AI applications. Amazon on Tuesday said its Nova Reel software allows users to make six-second videos that can be useful for, say, displaying products on the Amazon website. Videos of up to two minutes will be available in the coming months, Amazon said.

 

Entertainment industry technologists are eager to get their hands on such tools to more efficiently enhance and expedite filmmaking capabilities. Still, others worry that such systems could infringe on copyright works.

 

Also on Tuesday, Amazon said it had developed Canvas for generating images from short text prompts. Jassy emphasized that Amazon would include watermarking to ensure the software is used responsibly to prevent harmful content from being spread.

 

Other offerings announced Tuesday are meant to speed the time it takes to process and analyze text. And in the near future, Amazon plans to introduce an AI model that can take in text, images, speech and video and produce any of those.

 

In his comments on Tuesday, Jassy said Amazon would release a version of its Alexa speech assistant revamped with AI in the coming months. The project, known internally as Banyan, has been plagued by delays amid concerns over the accuracy and speed of its answers, Reuters has reported. 

 

Prasad, who previously oversaw Banyan, declined to offer additional specifics beyond saying he is a huge optimist over the future of voice assistants.

 

(Reporting by Greg Bensinger in Las Vegas; Editing by Will Dunham and Jonathan Oatis)