AIF-C01 — Frequently Asked Questions

To use a custom trained model in Amazon Bedrock, you must first import the model by registering it from Amazon SageMaker or importing it from Amazon S3. Once imported, the model can be provisioned for inference and used within the Bedrock environment.

You can increase image specificity by using more detailed and descriptive prompts that specify attributes like style, lighting, and composition. Additionally, using negative prompts to explicitly exclude unwanted elements helps guide the model towards a more specific output.

The most effective way is to use Amazon SageMaker Serverless Inference or Real-time Inference endpoints configured to run inside your VPC, which ensures that all traffic between your applications and the model endpoint remains securely within the AWS network.

Amazon Q Developer increases productivity by providing AI-powered assistance for generating code, writing unit tests, debugging applications, and upgrading Java versions, while understanding the context of your entire codebase.

Amazon EventBridge is used to send notifications by configuring a rule to listen for specific events from AWS Artifact, which then triggers an Amazon Simple Notification Service (SNS) topic to deliver the email.

Data access can be restricted using AWS Identity and Access Management (IAM) by creating policies that grant or deny permissions to specific Bedrock actions and attaching these policies to IAM roles or groups corresponding to each team.

Best practices include identifying and mitigating bias, ensuring transparency, implementing robust security measures against prompt injection, establishing human-in-the-loop review processes, and continuously monitoring model performance.

Compliance and regulatory requirements are determined during the initial Problem Formulation and Planning phase to ensure the entire project is designed to be compliant from the ground up.

The best technique is to use a clear, structured prompt with specific instructions and examples (few-shot prompting) that explicitly state the desired format, length, and key features to highlight.

You can prevent this by implementing Amazon Bedrock Guardrails to screen prompts and responses for sensitive information, or by using a pre-processing step to redact confidential data before it reaches the model.

A Retrieval-Augmented Generation (RAG) architecture with Amazon Bedrock is best, as it allows the LLM to access and reason over specific information from a vector database containing the legal documents.

SageMaker Batch Transform is the most suitable option for running inference on large volumes of data asynchronously without requiring a persistent endpoint.

The BLEU (Bilingual Evaluation Understudy) score is a common metric that evaluates machine translation by comparing the generated output to human reference translations based on n-gram overlap.

SageMaker Asynchronous Inference is designed for requests with large payload sizes and long processing times, queuing requests and notifying via Amazon SNS when results are ready.

Return on Investment (ROI) is a key metric used to measure financial effect by comparing the cost savings or revenue generated by the chatbot against its total development and operational costs.

Clustering algorithms, such as K-Means, are used for this unsupervised learning task to group data points into clusters based on feature similarity without pre-existing labels.

Prompt Probing or Model Probing attacks are designed to systematically test the model to understand its underlying instructions, guardrails, and limitations.

Content creation and summarization, including generating text, creating images from descriptions, summarizing long documents, and translating languages, are primary use cases.

Societal Bias or Prejudicial Bias occurs when an ML model learns and amplifies existing stereotypes and inequalities present in the real world or the training data.

The On-Demand pricing model offers flexibility by allowing you to pay only for the number of input and output tokens processed, with no minimum fees or upfront commitments.