Artificial intelligence and machine learning have been the focus of Amazon for more than 20 years. Many of the features Amazon provides to customers are driven by machine learning, such as e-commerce recommendation engines, routing for picking robots in fulfillment centers, and supply chain, forecasting, and capacity planning. The computer vision technology in the Prime Air (the Amazon drone) and the Amazon Go both use deep learning. Alexa responds to billions of customer requests each week to manage smart homes, shop, and get information and entertainment, also powered by more than 30 different machine learning systems. Amazon has thousands of engineers focused on machine learning research, which is the most concerned idea now and where its future-oriented strength lies.
Introducing Amazon Bedrock and Amazon Titan models: the easiest way to build and scale generative AI applications with foundational models
After listening to all the feedback from customers, Amazon Cloud Technology announced the launch of Amazon Bedrock. The new service allows users to access underlying models from AI21 Labs, Anthropic, Stability AI, and Amazon via an API. Bedrock is the easiest way for customers to build and extend generative AI applications using base models, lowering the barrier to entry for all developers. On Bedrock, users can access a range of powerful base models from text to images, as well as the Amazon Titan base model we released today, through scalable, reliable, and secure Amazon cloud technology hosting services. The Amazon Titan base model currently includes two new large language models. With the serverless experience brought by Bedrock, customers can easily find a model suitable for their own business, get started quickly, use their own data to customize based on the basic model, and use the model they are already familiar with while ensuring data security and privacy protection. Amazon cloud technology tools and capabilities to integrate and deploy custom models into their applications without having to manage any infrastructure. For example, customers can integrate the basic model with Amazon SageMaker machine learning functions, use Experiments to test different models, and use Pipelines to manage the basic model at scale.
Customers can also use Bedrock to access some of the most advanced base models available. This will include the Jurassic-2 multilingual large language model series developed by AI21 Labs, which can generate text content based on natural language instructions and currently supports Spanish, French, German, Portuguese, Italian and Dutch. There is also the large language model Claude developed by Anthropic, which is based on Anthropic's extensive research on training honest and responsible AI (responsible AI) systems, capable of performing a variety of dialogue and text processing tasks. Customers can also easily access Stable Diffusion, the basic Vincentian graph model developed by Stability AI through Bedrock, which is currently the most popular model in the Vincentian graph field, capable of generating unique, realistic, high-definition images, artwork, logos, and other design drawings.
One of Bedrock's most important capabilities is the extreme ease of customizing the model. Customers only need to show Bedrock a few labeled data examples in Amazon S3, and Bedrock can fine-tune the model for a specific task, with as few as 20 examples at least, without labeling a large amount of data. Let’s say a content marketing manager in the fashion retail industry wants to develop a new, targeted ad creative for an upcoming line of handbags. He provided Bedrock with some annotated examples of top-performing past marketing ads, as well as descriptions of new products, for which Bedrock would automatically generate effective social media tweet content, display ads, and product pages. No customer data was used to train the underlying model. All data is encrypted and never leaves the customer's virtual private network (VPC), so customers can be confident of data security and privacy protection.
Some customers have already previewed Amazon's new Titan base model, and Amazon Cloud Technology will further expand its availability in the coming months. First, two Titan models will be released. The first is a generative large language model for tasks such as summarization, text generation (such as original blogging), classification, open-ended question answering, and information extraction. The second is a text embedding (embeddings) large language model, which can translate text input (words, phrases, and even large articles) into digital representations that contain semantics (ie, embeddings embedded codes). While such large language models do not generate text, they are beneficial for applications such as personalized recommendation and search, because contrastive encodings can help the model return more relevant and contextual results than matching text. In fact, Amazon.com's product search capability uses a similar text embedding model to help customers better find what they're looking for. In order to continue to promote best practices in the use of responsible AI, the Titan base model can identify and remove harmful content in the data submitted by customers to custom models, reject user input of inappropriate content, and filter the output results of the model containing inappropriate content, such as hate speech , swearing and verbal violence.
Enterprises of any size can access the underlying models through Bedrock, accelerate the application of machine learning within the organization, and with its easy-to-use features, build their own generative AI applications. The Bedrock will be a big step in the generalization of the base model. Partners such as Accenture, Deloitte, Infosys, and Slalom are all building best practices to help businesses move fast with generative AI. Independent Software Vendors (ISVs) such as C3AI and Pega are looking forward to using Bedrock to easily access a large number of basic models, combining security, privacy and reliability.