Bedrock on AWS Provides Foundational Models for Building Generative AI Apps

Amazon Web Services (AWS) has released a new service called Amazon Bedrock that offers multiple foundational models designed to allow businesses to customize and create their own models. Generative AI Applications – Contains programs for general commercial use.

Amazon Bedrock provides users with underlying models from AI21 Labs, Anthropic, Stability AI, and Amazon, accessible via APIs. Announced on Thursday and currently in private preview, the service will be available in just one day. Databricks announced Dolly 2.0, its own open-source-based Large Language Model (LLM). It has a similar strategy: helping companies get around the constraints of the closed-loop model ( Chat GPT) prevents you from creating your own customized generative AI applications.

A closed-loop, pre-trained underlying model prevents companies from creating any form of generative AI that competes with the original.

This constraint accelerates research to create open-source models and other alternatives to generative AI targeted for commercial use, as companies demand customizable models for targeted use cases. I’m here.

Fundamental Models for Amazon AI

Amazon Foundation models available through the new service include: Titan FM on Amazonconsists of two new LLMs. Generally, AI models trained on large amounts of text data to produce human-like responses. These will be previewed on Thursday and will be available to the general public in the coming months.

According to the company, the first Titan foundation model is LLM, which generates tasks such as summarization, text generation, classification, open-ended Q&A, and information extraction.

The second, the company says, is an LLM that converts text input (words, phrases, or large units of text) into numerical representations (called embeddings) that contain the semantic meaning of the text.

“Although this LLM does not generate text, it is useful for applications such as personalization and search, because by comparing embeddings the model produces more relevant and contextual responses than word matching. and blog post.

The company says these underlying models are tuned to detect and remove harmful content within corporate data provided for customization, and AWS says these models remove harmful content from the output. It adds that you can also remove the .

According to Sivasubramanian, Amazon Bedrock can be used to find suitable models for different use cases and customize them accordingly for enterprise data sets before integrating and deploying them into applications with AWS tools. increase.

“Customers simply point Bedrock to a few labeled examples from Amazon S3, and the service refines the model for a given task without having to annotate a large amount of data (20 examples is enough). We can tune it,” Sivasubramanian said.

Train AI models without enterprise data

“Customer data is never used to train the underlying models, and all data is encrypted and never leaves the customer’s virtual private cloud (VPC), so customers can keep their data private. And you can trust it will remain confidential,” Sivasubramanian added.

In addition to the Titan foundational model, Bedrock includes the Jurassic-2 family of multilingual LLMs from AI21 Labs, which dictates text in Spanish, French, German, Portuguese, Italian, and Dutch according to natural language instructions. to generate

Anthropic’s LLM, Claude, is also included in Bedrock, capable of performing a variety of speech and text processing tasks, and is based on Anthropic’s extensive research into training honest and responsible AI systems, AWS said. .

In addition, Bedrock gets a text-to-imaging underlying model in the form of Stability AI’s text-to-image suite.

Other updates to our generative AI tools include making Amazon CodeWhisperer generally available in Python, Java, JavaScript, TypeScript, and C#.

CodeWhisperer, an AI-based code generator, CoPilot on GitHub owned by Microsoft, We recently added AI features to help developers.

AWS is Inf2 Cloud instance powered by AWS Inferentia2. According to the company, these instances are “specially optimized for large-scale generative AI applications with models containing hundreds of billions of parameters.”