Amazon Web Services (AWS) at the Cannes Lions International Festival of Creativity last year debuted its multibillion-dollar advertising business. This year it’s telling clients and agencies what AWS’s focus on generative artificial intelligence (GAI) and first-party data collaboration can do for them.
Some discussions include forthcoming identity resolution capabilities to help businesses match and link customer records stored across disparate channels without the need to build and maintain any any workloads. The idea is to let its customers preconfigure identity resolution workflows with rule based techniques and machine learning models.
Others will focus on GAI as a subset of machine learning (ML) powered by ultra-large models, including large language models (LLMs) and multi-modal models for text, images, video, and audio.
Applications like ChatGPT have captured everyone’s attention, but there are other companies like AWS working on GAI applications for advertising and development.
AWS designs silicon in-house. Discussion also will focus on the general availability of Trn1n instances, powered by its Trainium chips, which doubles the network bandwidth to deliver higher performance for training GAI models.
Amazon Bedrock, for example, launched in April, is a new service for building and scaling GAI applications that can generate text, images, audio, and synthetic data in response to prompts. It gives customers access to foundation models (FMs) — ultra-large ML models that GAI relies on — from the top AI startup-model providers, including AI21, Anthropic, and Stability AI, and access to the AWS developed Titan family of foundation models.
Jon Williams, global head of Agency BD and Solutions at AWS, says the idea is to help clients save cost and time, and easily navigate through emerging hurdles based on technology they might not fully understand or know how to use. They will do it with access to foundation models.
“If you develop it by yourself, it’s quite an expensive and time-consuming task,” he said. “We're finding that many customers don't have the technical capability to do that by themselves.”
At the top of the stack, AWS provides CodeWhisperer, a GAI coding companion that generates whole lines and full function code suggestions, prompts, in an Interactive Disassembler (IDA) or code debugger.
It is used by developers to improve productivity by generating code recommendations for AWS based on natural-language comments and prior code to integrate into a developer environment.
Williams said the company sees code that can “complete tasks 57% faster than those that didn't use CodeWhisperer, and also more successfully complete tasks 27% more frequently.”
The company sees CodeWhisperer as a tool to free up valuable time for developers, allowing companies to innovate faster. These improvements for all tools are wrapped in techniques that keep data private.
“We want to make sure a customer's private data and their custom models do not benefit other companies or, especially, their competitors,” he said.
Companies running GAI models on AWS use the company’s Inferential, chips with low-energy efficiency and low cost for running these types of workloads.
Williams said AWS has a GAI team. Industry teams, advertising and marketing, support the technology. Both work together to inform clients of the services.
“We work backwards from common challenges heard from customers,” he said. “Clean rooms, for example, is one way I mentioned earlier. We heard from many of our customers they wanted to be able to collaborate on data on AWS, but without any data movement.”
Amazon entered the semiconductor business in 2016 to build its own chips, purchasing Annapurna Labs, an Israeli company based in Silicon Valley. It initially targeted customers equipment like Wi-Fi routers, data storage gear and media-streaming devices. At the time, Annapurna said its products included chips based on technology licensed by ARM Holdings.