AWS showed its hand in an ever-intensifying generative AI race, unveiling a new cloud-based service for developers to build applications on systems developed by a number of partnering startups, as well as the company’s own in-house language model.

AWS, Amazon’s cloud unit, which alongside the new service also rolled out a generative AI whitepaper and a startup accelerator programme, indicated a major jump into the AI arms race, but is taking somewhat of a different approach to rival early movers Microsoft which is backing OpenAI’s ChatGPT and Google.

The stand-out from AWS was the introduction of Bedrock, a generative AI platform that will open access to its language system dubbed Titan, as well as other models from third party startups A121 Labs, Google-backed Anthropic and Stability AI.

AWS explained Bedrock lets businesses and developers customise their own foundation models (FMs) to generate an image from a prompt or respond to queries with a human-like response, while helping to build ‘enterprise-scale’ AI apps.

Those clients adopting the Titan model will be able to input their own data, but that data will not be able to train the system.

Not unlike other recent announcements around AI systems, AWS stated Titan has the capability to generate text from blog posts, emails and other documents, while the image-based option helps with search and personalisation.

Rapid race
Speaking in an interview with CNBC, Bratin Saha, AWS VP, added Amazon is focused on ensuring its Titan model was developed to produce high quality responses, after ChatGPT and Microsoft’s Bing chatbot had generated inaccurate information in some cases.

Despite an open letter backed by technology heavyweights (including Elon Musk) calling for a six month pause in the development of AI systems last month, AWS becomes the second company this week alone to make noises around new products based on the technology, joining Chinese heavyweight Alibaba.

Amazon CEO Steve Jassey further highlighted the company’s generative AI strategy in a letter to shareholders.

He noted Amazon has been using machine learning extensively across various applications for 25 years. He stated the new large language models would “transform and improve virtually every customer experience”.

He also said Amazon would continue to invest Project Kuiper, which is the company’s satellite-based broadband service. It expects to begin mass-producing satellites by the end of this year, with initial launches and commercial services planned for 2024.

*Additional reporting by Mike Robuck