Amazon Web Services (AWS) today announced new Amazon Bedrock innovations that provide customers with the easiest, fastest, and most secure way to develop advanced generative artificial intelligence (AI) applications and experiences.
Let's take a closer look at them with illustrations created with one of these innovations, the Amazon Titan Image Generator (from the surface of Saturn's largest moon Titan, of course).
New custom model import feature helps organizations bring their own models to Amazon Bedrock
Amazon Bedrock customers are increasingly leveraging their own data by customizing publishing models for domain-specific use cases. This is because when a customer combines smarts from the various Foundation Models (FM) and Large-Scale Language Models (LLM) available in her Bedrock with their own data, a compound intelligence effect occurs. Think of this as the generative AI version of “Two heads (or more) are better than one.” The added intelligence also means that the resulting applications can support a wider variety of use cases. So what does a customer want? An easy and secure way to add his own custom models to his Bedrock.
Amazon Bedrock Custom Model Import allows organizations to import and access their own custom models as Bedrock's fully managed application programming interfaces (APIs), giving organizations unprecedented choice when building generative AI applications. You can First, organizations can easily add models to Amazon Bedrock that they customize with Amazon SageMaker or via another third-party tool or cloud provider. Once you complete the automated validation process, you can seamlessly access your custom model just like any other model on Amazon Bedrock.
With this new feature, AWS makes it easy for organizations to choose a combination of Amazon Bedrock models and their own custom models through the same API. Amazon Bedrock Custom Model Import is currently available in preview and supports three of the most popular open model architectures (Flan-T5, Llama, and Mistral), with more architectures planned for the future. is.
Model evaluation helps customers evaluate, compare, and select the best models to build and deploy generative AI applications
Of course, customers want to understand more precisely which models are best suited for their specific applications before combining models to increase intelligence. Choosing the best model for a particular use case requires a delicate balance between model accuracy and model performance. Historically, organizations have had to perform this often tedious and time-consuming balancing act for each new model or use case. As a result, the development and delivery of generative AI experiences to customers was delayed.
Model evaluation, now generally available, is the fastest way for organizations to analyze and compare models in Amazon Bedrock, reducing the time it takes to evaluate models and bringing new applications and experiences to market faster. It will look like this. Customers can get started quickly by selecting predefined evaluation criteria (such as accuracy and robustness), uploading their own prompt library, or choosing from built-in, publicly available datasets. For content that requires subjective criteria or nuanced judgment, customers can easily set up human-based rating workflows. After you complete the setup process, Amazon Bedrock runs evaluations and generates reports so you understand how your models performed against key criteria and quickly select the best model for your use case. You can select
Two more AI models in the Amazon Titan family
AWS is announcing the general availability of Amazon Titan Image Generator with invisible watermarks and the latest version of Amazon Titan Text Embeddings, both exclusively on Amazon Bedrock.
People in industries such as advertising, e-commerce, and media and entertainment now have access to Amazon Titan Image Generator to create low-cost, high-quality images from scratch, or to enhance and edit existing images. became. Simply enter a text description in the prompt field, and Amazon Titan will convert that text into any images and styles you describe. For example, the prompt in the image shown above reads, “From the surface of the moon Titan with Saturn in the background, the text “Flowers'' in a modern font emerges from the mouth of a very friendly robot, and the text reads “Sunny.'' It will be an image of a flower. on the surface of the moon. ”
Amazon Titan applies an invisible watermark to every image it generates, identifying AI-generated images to foster the development of safe, secure, and transparent AI technologies and reduce the spread of misinformation. Reduces This model can also check for the presence of watermarks, allowing customers to see if an image was generated by Amazon Titan Image Generator.
Amazon Titan Image Generator Demo | Amazon Web Services
The second member of the family announced today is Amazon Titan Text Embeddings V2. It is optimized for working with search augmentation generation (RAG) use cases and is suitable for a variety of use cases such as information retrieval, question and answer, etc. Chatbots and personalized recommendations. RAG is a common model customization technique that connects the FM to additional knowledge sources that it can consult to curate more accurate responses. Although the results are ideal, performing these operations can be computationally and storage intensive. With the release of Amazon Titan Text Embeddings V2 later this month, customers will have the option to take advantage of flexible embedding sizes to meet the needs of a variety of applications, from low-latency mobile deployments to high-precision asynchronous workflows. , overall storage is reduced by up to 4x. While maintaining 97% accuracy for the RAG use case.
Guardrails for Amazon Bedrock allows customers to easily implement safety measures to remove personal and sensitive information, profanity, specific words, and block harmful content.
For generative AI to become pervasive across all industries, organizations must ensure that generative AI is implemented in a safe, reliable, and responsible manner. Many models use built-in controls to filter unwanted and harmful content, but organizations can also use built-in controls to help ensure responses remain relevant, align with corporate policies, and adhere to responsible AI principles. We would like to further select models. Now generally available, Guardrails for Bedrock provides industry-leading safety protections that help customers block up to 85% of harmful content.
To create guardrails, customers simply provide natural language descriptions that define topics that are denied within the context of their application. You can also set thresholds to filter out content across hate speech, insults, sexual content, violence, and more. This is in addition to filters that remove profanity and certain blocked words. Guardrails is the only solution from a top cloud provider that allows customers to take advantage of built-in and custom protection features in one of his products, all large-scale language models and fine-tuning within Amazon Bedrock. Works with models that are
Amazon Bedrock gives you access to a wide selection of first-party and third-party large-scale language models (LLMs) and other foundational models (FMs) for leading AI, so tens of thousands of customers are already creating the foundation for their AI strategies. Amazon Bedrock is selected as the Along with companies like AI21 Labs, Anthropic, Cohere, Meta, Mistral AI, Stability AI, and Amazon, we offer great ease-of-use features to quickly build and deploy generative AI applications. Amazon Bedrock's powerful model is delivered as a fully managed service, so customers don't have to worry about provisioning compute instances, ensuring seamless deployment, scalability, and continuous optimization.
let's start Amazon bedrock and learn about Amazon's AI innovation.
