Hybrid AI Era... "Smartly Choose and Use with Amazon Bedrock"

Amazon Bedrock New Features Equipped
Enhancing Agent Execution Power and Safety

Amazon Web Services (AWS)'s generative artificial intelligence (AI) platform, Amazon Bedrock, has evolved to the next level. It has enhanced execution capabilities and safety to accelerate its penetration into the enterprise AI market.

AWS Korea held the 'AWS 2024 Generative AI Media Briefing' on the 13th at Centerfield East in Yeoksam-dong, sharing new features of Amazon Bedrock and domestic enterprise use cases.

Kim Seonsu, Senior Specialist in AI/ML Business Development at AWS Korea, is introducing new features of Amazon Bedrock at the 'AWS 2024 Generative AI Media Briefing' held on the 3rd at Centerfield East in Yeoksam-dong.
[Photo by AWS Korea]

Amazon Bedrock offers various foundation models through a single application programming interface (API). Customers can use not only Amazon's own model, Titan, but also Anthropic's Claude, Meta's LLaMA, Mistral AI's Mistral, and others. This allows customers to build generative AI applications utilizing diverse models.

Through the Amazon Bedrock update, AWS has achieved ▲ optimization of generative AI models ▲ enhancement of data connectivity performance ▲ strengthening of responsible AI features ▲ and improvement of execution capabilities.

First, to increase the model's execution capability, a memory retention feature was added to the Amazon Bedrock Agent. The agent is an AI capable of executing complex multi-step tasks. The Bedrock Agent remembers multiple user interactions over time, providing a personalized experience. Each user's conversation history is stored with a unique memory identifier (ID), ensuring the security of user data.

Reliability has also been strengthened through the 'Guardrail API' feature and 'Contextual Grounding Checks.' The Guardrail API applies standardized and consistent safeguards to all AI applications regardless of the underlying infrastructure. Contextual Grounding Checks detect and block hallucinations by verifying whether the generative AI's responses are relevant to the query and based on corporate data.

Additionally, data connectivity performance has been improved. Enterprises use Retrieval-Augmented Generation (RAG) techniques to enhance the expertise of generative AI models. RAG links up-to-date information or internal corporate data to models trained on large-scale data to improve answer accuracy. AWS connects external data sources such as Salesforce and SharePoint for use in RAG.

Kim Seonsu, Senior Specialist for AI/Machine Learning (ML) Business Development at AWS Korea, said, "Among companies using generative AI models, 41% use three or more models, showing a high demand for multiple models. The key point of Bedrock is providing multiple latest models through a single API."

Generative AI tools for developers have also been equipped with new features. The developer tool 'Amazon Q Developer' has newly launched customization and code transformation features. 'Amazon Q Developer Customization' suggests customized code tailored to developer needs based on internal codes or best practices. 'Amazon Q Developer Code Transformation' automates upgrade tasks such as code updates, testing, and deployment readiness checks, reducing the time required for these tasks.

For non-developers, 'AWS App Studio' was also introduced, enabling application creation through natural language. Users simply input the desired application functions and data sources to integrate, and App Studio generates the application within minutes. It is explained that this can reduce costs by up to 80% compared to other low-code solutions.