AWS Launches New Q Apps and AI App Studio

Applications of AI

On Wednesday, July 10, Amazon's AWS Summit in New York featured many announcements. The company is actively adopting AI applications in various products. One of the most important announcements was the public preview of App Studio, a platform for creating applications without code. Developers can build applications using AI by providing instructions in natural language.

By simply specifying the app's functionality and data sources, you can generate an enterprise-grade application within minutes, a task that typically takes a developer days.

In a live demo, Amazon showed how App Studio builds an invoice tracking app based on a simple request, with AI providing detailed suggestions for additional features. Once the app structure is approved, users can further customize it using intuitive drag-and-drop tools.

App Studio integrates with third-party services and AWS itself via connectors, and Adam Seligman, vice president of developer experience at AWS, said there are more integrations to come based on customer requirements.

Q Developer to Sagemaker

Another announcement concerns Amazon Q, the company's AI assistant. The developer version will be available for Sagemaker, Amazon's machine learning environment. Q was previously only available in the AWS console and in developer environments such as IntelliJ IDEA, Visual Studio and VS Code, where the assistant already provides advice, generates code and helps with troubleshooting.

According to Amazon, the integration into Sagemaker simplifies machine learning workloads, including providing advice on preparing your data (important since this is essentially what is used to train the AI) and showing which approach (code, no-code, what data format to use) is recommended for each use case.

Amazon Q Developer now allows you to use your organization's internal code libraries as well as APIs, packages, classes, and other methods that enable the AI ​​assistant to generate appropriate code suggestions.

Additionally, Amazon Q apps are becoming generally available as a feature within Q Business, allowing non-developers to build their own apps on a limited scale with their own data. Think applications for consolidating feedback, creating onboarding plans, taking notes, etc. The idea is that such capabilities will free up IT staff and developers to focus on more important or difficult tasks, while allowing other employees to somewhat build their own tools.

Bedrock Updates

Bedrock, Amazon's enterprise platform for building generative AI based on one or more existing models, is also receiving a series of updates: Anthropic's compact Claude 3 Haiku model running on Amazon can now be tweaked in Bedrock, marking the first time that a Claude 3 model has been suitable for this purpose.

With proper fine-tuning, customers can deploy models for very specific use cases, taking into account their organizational idiosyncrasies, interests, compliance requirements, and workflows. This feature is currently in preview.

Additionally, Bedrock's knowledge base feature can now take advantage of more data sources, including connectors for Confluence, SharePoint, Salesforce, and user-specified internet sources. The tool also knows how to more accurately handle data from CSV and PDF files. The idea is that trained models can better navigate your own data maze and extract data specifically from it.

Agents built with Bedrock have improved their ability to interpret code and retain memory, allowing them to keep summaries of previous conversations with a user and take them into account when generating advice for multi-step tasks like booking a flight or processing a claim. Such agents can now analyze and visualize data, perform complex calculations, and process text.

Guardrails to combat hallucinations

Capping off a series of announcements, features, and enhancements, AWS is releasing important updates around security and compliance: With Guardrails for Bedrock, AWS provides another defense against AI hallucinations by grounding models, especially when customers use RAGs and their own source data.

This way answers are always generated with the right unique data and always take into account the user's original query. This feature, added on top of existing filters, prevents 75% of hallucinations. Additionally, Guardrails' API has been released, making the feature available on non-Bedrock underlying models.

AWS goes further The company announced updates on its responsible AI policy efforts, including visual watermarks, enhanced guidelines, and more. We also provide tips on responsible AI training.

Read also: AWS invests $7.8 billion to expand European sovereign cloud

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *