Practice Quiz

If you access Amazon Bedrock models by using API calls, the models don't recall prior prompts and previous requests, unless the previous interaction is included within the current prompt.

When using Anthropic Claude models, it's a good practice to use <example></example> tags to include demonstration examples, and to use different delimiters such as H: and A: in the examples to avoid confusion with the delimiters Human: and Assistant: for the whole prompt.

To reduce hallucinations, you can refine your prompt using prompt optimization techniques; use techniques like Retrieval Augmented Generation (RAG) to provide the model access to more relevant data; or use a different model that might produce improved results.

Few-shot prompting (in-context learning) involves providing paired example inputs and desired outputs (shots) to help LLMs better calibrate their output, whereas zero-shot prompting provides no example input-output pair in the prompt text.

If you are accessing Anthropic Claude models using API calls instead of the Amazon Bedrock text playground, you wrap your prompts in a conversational style to get the desired response, whereas this conversational style formatting is not necessary for the Titan Text and AI21 Labs Jurassic model families.

To use conversational mode on Titan, you can use the format of User: {{}} \n Bot: when prompting the model.

Amazon Bedrock models do not recall prior prompts or previous requests across API calls unless the previous context is explicitly included within the current prompt text.

Double curly braces {{ and }} are used in prompt templates to indicate placeholders for data-specific information and should be excluded from the final prompt text.

To reduce hallucinations in LLMs, users can refine prompts using prompt optimization techniques, utilize Retrieval Augmented Generation (RAG) to provide access to relevant data, or select a different model.

For Anthropic Claude models, it is recommended to wrap demonstration examples within <example></example> tags and use delimiters like H: and A: inside them to avoid confusion with the main prompt's Human: and Assistant: delimiters.

Few-shot prompting, also known as in-context learning, provides paired example inputs and desired outputs (shots) to calibrate model expectations, whereas zero-shot prompting contains no examples.

A single prompt includes several components to guide a model, including a task or instruction, context of the task, demonstration examples, and the input text.

Prompts sent to Anthropic Claude models via API must use a conversational style containing the specific delimiters '\\n\\nHuman:' and '\\n\\nAssistant:'.