After Amazon Cloud Technology launched the "Generative AI Technology Stack" on the 29th, it announced a series of key services and tools that support this new stack at the re:Invent 2023 conference on the 30th.
In his keynote speech, Swami Sivasubramanian, Vice President of Data and Artificial Intelligence at Amazon Cloud Technology, compared generative artificial intelligence with a "supernova explosion" to describe the innovative potential that current artificial intelligence brings to the public. “The powerful relationship between humans and technology is unfolding in front of all of us,” Sivasubramanian said. “I like to think of what’s happening today as a beneficial relationship between data, generative AI, and humans.”
But to make good use of the capabilities of generative AI, you must have corresponding tools for building and deploying applications and machine learning infrastructure. In the speech, Amazon Cloud Technology mainly talked about their artificial intelligence tools and services from three aspects: the machine learning platform SageMaker, the newly launched Amazon Q, and the Amazon Bedrock platform that provides basic model services.
Amazon Cloud Technologies has launched a new solution for SageMaker called HyperPod that aims to eliminate the tedious machine learning infrastructure work involved in optimizing model training. The company claims that this new product can reduce model training time by 40%. “This is a major event and customers will clearly feel the improvement.” Sivasubramanian emphasized.
HyperPod solves the limitations of distributed infrastructure in processing big data and training model complexity. Typically, this involves spreading model training tasks across hundreds or thousands of accelerators and running trillions of data computations in parallel over weeks or months. This means writing a lot of code and constantly optimizing it. SageMaker HyperPod allows automatic distribution of training workloads among computing resources and saves checkpoints regularly to avoid lost time if a component fails.
In addition, based on Amazon Q released by Amazon Cloud Technology, generative SQL has been added to Redshift to simplify query writing, and a data integration function has been added to the serverless Amazon Glue platform, which can be called through natural language (Amazon Q data integration).
"Q supports every area of your business by connecting your data," said Sivasubramanian. "We can use this technology to solve some of the big challenges in data management."
Amazon Bedrock was the highlight of the conference, showcased during keynotes by CEO Adam Selipsky and Sivasubramanian. Amazon Bedrock is used by more than 10,000 customers, and the company has been steadily accumulating new features and making a series of enhancements to Amazon Bedrock, including the fine-tuning capabilities and new security services mentioned in Selipsky's presentation yesterday.
In order to meet the needs of a wider range of model selection, Amazon Cloud Technology announced that it has included the latest version of Anthropic's language model, Claude 2.1, and will also add Meta's Llama2 70B to Bedrock.
"We are still in the early stages of generative AI, and these models will continue to accelerate," Sivasubramanian said.
In addition, vector databases are also important for generative AI. Amazon Cloud Technology has added vector search functions to many mature databases, including Amazon Aurora PostgreSQL, Amazon RDS for PostgreSQL, Amazon OpenSearch Service, Amazon OpenSearch Serverless, etc. Amazon Document DB and Amazon Dynamo DB have also added new vector retrieval functions.
In addition, Sivasubramanian also specifically introduced the vector search function in Amazon MemoryDB for Redis, emphasizing that it can provide millisecond-level response under the conditions of the highest recall rate and highest throughput.
“The powerful symbiotic relationship between data, generative AI and humans is accelerating our ability to innovate,” said Sivasubramanian. "Amazon Cloud Technologies has everything it takes to unlock this powerful technology."