AWS dives into generative AI with new offerings

Coinciding with its latest letter to shareholders written by chief executive Andy Jassy, Amazon's cloud computing division has officially launched a set of generative artificial intelligence offerings.

Jassy, who sounded an upbeat note on Amazon's overall prospects despite what he called "this challenging macroeconomic time" and large-scale redundancies across the online retailer and cloud giant's entire business, said he believed AI will transform virtually every customer experience.

"I could write an entire letter on LLMs (large language models) and Generative AI as I think they will be that transformative, but I’ll leave that for a future letter," Jassy said.

"Let’s just say that LLMs and Generative AI are going to be a big deal for customers, our shareholders, and Amazon," he added.

AWS vice president of databases, analytics, and machine learning, Swami Sivasubramaniam, said "we are truly at an exciting inflection point in the widespread adoption of ML" and outlined the new offerings being launched.

Starting with the AWS Bedrock service, which in a similar fashion to OpenAI's ChatGPT provides access to "foundation models" (FMs) such as LLMs for building generative AI applications that generate text, images, audio and synthetic data in response to input prompts, Sivasubramaniam the cloud giant's customers had asked it how they can quickly take advantage of what's out there today.

Bedrock provides access to FMs from AI21 Labs, Anthropic, Stability AI and Amazon itself, via an application programming interface (API).

The FMs include Amazon Titan, two new LLMs; one can be used for tasks such as summarising and text generation to create for example blog posts, the company said.

It can also be used for classification, open-ended quality and assurance, and information extraction.

The second Titan LLM translates text into numeric representations or embeddings, that contain the semantic meaning of the written material.

"While this LLM will not generate text, it is useful for applications like personalisation and search because by comparing embeddings the model will produce more relevant and contextual responses than word matching," Sivasubramaniam said.

AWS also introduced general availability of its Elastic Compute Cloud (EC2) Trn1n instances with in-house developed Trainium processors.

Trn1n can deliver up to 50 percent savings on training costs compared to other EC2 instances, AWS claimed; they are optimised to distribute training across multiple servers over the second-generation Elastic Fabric Adapter (EFA) networking, with 800 gigabits per second performance.

The instances can be scale up to 30,000 Trainium chips, offering more than 6 exaflops of compute, using Trn1n instances deployed in UltraClusters in the same AWS Availability Zone with petabit scale networking.

AWS also made generally available network-optimised Trn1n instances with 1600 Gbps bandwidth, designed to deliver 20 percent higher performance over standard instances.

For inferencing, AWS made its Inf2 instances generally available.

Inf2 "are optimised specifically for large-scale generative AI applications with models containing hundreds of billions of parameters," Sivasubramaniam said.

AWS claimed Inf2, based on the second-generation Inferentia2 accelerator, deliver up to four times higher throughput and up to ten times lower latency compared to earlier versions of its inferencing instances.

Similar to Microsoft-owned Github's Copilot, Amazon launched generative AI tool for developers named CodeWhisperer.

CodeWhisperer is free for individual developers and uses an FM to improve coder productivity.

It can generate suggestions in real-time based on developers' comments in natural language and earlier code in the integrated developmenet environment (IDE).

Currently, CodeWhisperer supports Python, Java, Javascript, TypeScript and C#, with more languages being accessed via the IDE and extensions.

CodeWhisperer also comes with built-in security scanning with automated reasoning to detect and remediate vulnerabilities, and more.

It can also help developers code responsibly, filtering out biased and unfair programming suggestions, and those resembling open source code that customers may want to reference or license to use, Sivasubramaniam said.

With the new services, hardware and developer tools, Amazon is seeking to position itself as the cloud-based ML and AI workload provider of choice, offering the lowest cost of training and running inference.

The cloud giant faces stiff competition however from Microsoft which is backing ChatGPT creator OpenAI which has grabbed headlines lately, and Google which is accelerating its AI efforts currently.