Incorporate offline and online human – machine workflows into your generative AI applications on AWS

This article discusses how to incorporate offline and online human workflows into generative AI applications on AWS.

Specifically, the article covers:

• An offline human feedback workflow using Amazon SageMaker Ground Truth to collect human evaluations on chatbot responses, which can be used for reinforcement learning to fine-tune the chatbot's language model (RLHF).
• An online human workflow using Amazon Bedrock, LangChain, and AWS Step Functions to invoke real-time human intervention based on sentiment analysis of chatbot responses, allowing human experts to take over or join conversations when the AI reaches its limits.
• Code snippets, architecture diagrams, and an example use case demonstrating the implementation of these workflows.
• The conclusion emphasizes how these human-in-the-loop techniques enable the development of responsible and robust generative AI applications.