Amazon Web Services (AWS) is getting into the generational artificial intelligence (AI) game with Bedrock, a cloud-based service that the company says will make it easier and quicker for developers to build generative-AI apps.
In a blog post, Swami Sivasubramanian, VP of data, analytics and AI at AWS, said that Bedrock builds on the company’s decades of work “democratizing” AI and machine learning (ML) to make it more accessible to its customers. He added that Amazon incorporated feedback from customers who said they needed AI to be secure and private and that they wanted to control how their data was shared and used by Bedrock.
Basically, Bedrock makes foundational models (FMs) from startups like A121 Labs, Anthropic, Stability AI and Amazon accessible via an API. In addition, Bedrock will offer access to other FMs, including Amazon’s Titan FMs, which consists of two new large language models (LLMs) that Amazon also revealed today. Those new Titan FMs include a generative LLM for tasks like summarization, text generation (for creating a blog post), classification, open-ended Q&A and information extraction. Sivasubramanian said that the company has been previewing the Titan FMs with customers and they will be widely available in the coming months. Bedrock is available for limited preview now.
Amazon is the latest company to launch a generational AI service. In March, Open AI launched GPT-4, an updated version of the platform that powers the chatbot ChatGPT. In addition, Microsoft recently launched a new version of its Bing search engine that includes a chatbot powered by GPT-4 and Google also debuted its own AI chatbot, Bard.
Arun Chandrasekaran, distinguished VP analyst with Gartner, said that Amazon’s Bedrock announcement is notable. “Generative AI is clearly the new battleground for Enterprise Cloud,” he stated.
He added that Amazon is betting on customer demand for fine-tuned models that directly align with their businesses needs. “The rise in more generative AI services along with more fine-tuned and open-source models is a positive step for clients as it can reduce the time to market and provide more deployment flexibility,” Chandrasekaran said.
Besides Bedrock, AWS also announced general availability of Amazon CodeWhisperer, a real-time AI coding companion that helps developers write code with built-in security scanning features. AWS is making CodeWhisperer available for free for individual developers.
Plus, the company said its Amazon EC2 Trn1n Instances powered by AWS Trainium and Amazon EC2 Inf2 Instances powered by AWS Inferentia2 are now generally available.
For the past five years, AWS has been investing in its own silicon because it was anticipating the need for better performance due to the demanding workloads produced by ML and AI.
Sivasubramanian said that the Trn1n instances, powered by Trainium, can help reduce costs and deliver as much as 50% savings on training costs compared to other EC2 instances.
This is important because right now customers are in the early stages of deploying FMs into production and spending much of their time on training FMs. However, when FMs are deployed at scale, customers can rack up a lot of costs running the models and generating hundreds of thousands to millions of inferences per hour. By using custom silicon such as Trainium and Inferencia2, the process can be optimized at the silicon level and the software stack level, making it more cost effective, Sivasubramanian concluded.