Amazon Zeroes In On Diverse, Safe AI


By: R. Scott Raynovich

LAS VEGAS -- If you expected anything other than AI, AI, and more AI at Amazon's re:Invent conference, you expected wrong. Amazon Web Services featured a theme of pursuing a diverse and safe AI strategy here at its annual conference.

AWS executives were quick to pounce on the setup from the OpenAI drama of the preceding weeks, when Sam Altman, the CEO of Microsoft's partner, OpenAI, was fired and then later restored in a reconfiguration of the board led by OpenAI investors including Microsoft.

AWS took the opportunity to quickly position itself as the more diverse AI provider, with more AI services and more AI partners. It also emphasized that all of its AI products will focus on data security, privacy, and safety "guardrails."

"You don't want a cloud provider that is beholden to one model provider," said AWS CEO Adam Selipsky, in a thinly veiled reference to Microsoft's tight partnership with OpenAI. "The events of the past ten days have made that very clear."

Diversifying AI Models

Selipsky pointed to a strategy of diversifying AI training and foundation models. For example, Amazon's Bedrock AI service uses models including Anthropic's Claude 2.1, Stability AI's Diffusion XL, Cohere, Amazon's Titan, and Meta's Llama2 -- among others.

In addition to launching a cornerstone new gen AI service, Amazon Q, as well as enhancements to its AI service Amazon Bedrock, speakers here emphasized that any use of AI in Amazon's services has to be safe and must protect user and customer data.

Amazon VP of Technology Matt Wood argued that generative AI is still in the early stages of the technology "S curve" and needs a lot of work in verification and accuracy.

"I would argue we are at the lowest point in the S curve," said Wood in a keynote with analysts. "Huge amount of verification that needs to be done going forward." He said Amazon's focus will be on responsible AI. "Amazon was built with privacy and security from the ground up," said Wood.

AWS Launches the Q Business Assistant

These themes ran through many of Amazon's announcements, including one of the cornerstone launches: Amazon Q, a custom business assistant with safeguards for business context, data, and permissions.

"Q understands and respects your own identity, roles, and permissions," said Selipsky. "We will never use business customers' content to train Q."

The idea behind Q is that the assistant can be customized and fine-tuned for specific businesses and context, so that customers can also control which data is used on their own terms.

Customer presentations by Pfizer, BMW, and Delta demonstrated how AWS customers are using AWS services to build custom AI apps and data services.

Fine-tuning Drives Business Customization

In other announcements, AWS emphasized customization. For example, AWS announced fine-tuning of AI models including Meta Llama, Titan, and Cohere foundation models (FMs), which means that organizations can label specific datasets for use in AI models to provide context for their businesses.

Some additional announcements at AWS (you can see all of them here):

Futuriom Take: The AWS team did a masterful job of answering recent debates on AI and taking control of the message by focusing on a controlled and diverse AI strategy emphasizing data security and safety.