Amazon Web Services GenAI

Flashcards covering key terminology and concepts from the provided lecture notes on Amazon Web Services (AWS) GenAI services.

Manus.im

A multi-modal intelligent agent product with over 50 agent application scenarios that has ignited industry demand for customized enterprise intelligent agents.

Claude 3.7 Sonnet

Preferred for its coding, agentic, reasoning, and multi-modal capabilities, making it ideal for intelligent agent applications.

MCP+Agent+Computer Use+Coding

Delivers results that depends mainly on the model capabilities.

Nova

Sensitive to low latency and low cost, making it suitable for real-time translation.

Nova

Good for its consistent style, making it appropriate for image generation.

Claude

Good for role-playing with lively characters.

DeepSeek

Expert at reasoning and planning.

Claude

Provides a strong text understanding, and has creation and generative capabilities. Good for content analysis and creation.

DeepSeek R1/Claude 3.5

Used as reasoning in complex jobs until there is a continuous iteration

Claude Sonnet

Preferred for its language understanding depth, retention of language style, adaptation to context, and consistent language.

Claude Sonnet

Preferred for literary control ability, artistic conception restoration ability, language diversity and stylistic consistence, and depth of cultural conversion.

Claude Sonnet

Preferred for maintaining the integrity of format and structure, technical translation accuracy, context understanding, and long text processing ability.

Amazon Nova (Micro and Lite)

Most preferred model because of its low latency and cost.

Translation Agent

Employs a unique reflection workflow that simulates the thinking process of human translation experts, breaking down the translation task into three main steps.

Three main steps to the Translation Agent workflow

Includes initial translation, reflection and improvement, and optimized output.

Initial Translation

The action of using LLM to make a preliminary translation of the input text, for an initial translation.

Reflection and Improvement

The process of helping LLM reflect on its own translation and to offer suggestions for improvement.

Optimized Output

The process of optimizing the initial translation based on the improvements recommended by LLM, and generating a more accurate rendering.

Sonnet

Has the capacity of language style retention, contextual adaptation ability, and language simple and accurate.

Translating for GenAI

A quick, high-quality, and low-cost way to realize content localization.

AWS

A cloud vendor who can also assist customers in promoting AI agent-related projects.

Claude Sonnet

Preferred for its high-precision text understanding ability, multi-dimensional analysis ability, structured output ability and multilingual processing ability.

Amazon Bedrock

Allows a user prompt to be translated into english.

Voice of the Customer (VOC)

Provides sentiment analysis and insight reports, allowing for an understanding of end-customer emotions.

Sonnet

High accuracy of the Spanish language in the translation of terms.

Sonnet

For sentiment analysis and comment analysis, the main model is __.

Claude Sonnet 3.7

Preferred for its multimodal understanding, rich language expression, strong multilingual ability, stable output format/style, and enhanced plot driving ability.

Claude Sonnet 3.7

Used for code development, supports 200K contexts and up to 128K output, fast generation.

Claude Sonnet 3.7

Preferred for its strong reasoning ability, higher quality and more diverse generated titles and a stronger ability to call tools.

Nova Pro/Lite

The model with the highest performance price ratio and fast reasoning speed.

Claude Sonnet 3.7

Aided by Text2SQL, reasoning and planning capabilities and higher report quality.

Claude Sonnet

Good at extracting complex information and has a strong multilingual capabilities.

Claude Sonnet

Good at understanding complex terminology and professional expressions.

Claude Sonnet

Can understand the content of the document more accurately and process complex document summaries and abstracts.

Nova Model

It can be used in simple meeting information extraction and other tasks.

Amazon Transcribe

Used to parse multi-person audio data, uses the detectron2 model to analyze the document layout, and implements MapReduce summarization of different document modalities.

Nova series

The model recommended for writing assistants, based on scenario simplicity and fast response speed.

Amazon Bedrock

Used for semantic checking of documents. The model's fast response can effectively improve modification speed and reduce user waiting time.

LLM

A way for the model to extract a presentation outline+details.

Small and Medium size Language Models

Model that does not have as much technical training as others.

Dialogue Robot

A combination of Sonnet and Nova.

Claude Sonnet

Choose this model for more accurate fine-grained analysis.

Nova Pro & Lite

A model that is often used to evaluate a customer case and is low cost.

Claude Sonnet

Preferred Model for marketing content review.

Claude Sonnet

A strong model for reviewing long documents.

Claude Sonnet

Can understand Chinese fonts and complex charts more accurately.

Amazon Q Developer

Provide key functions such as, Generating Dockerfile, Generating CDK project,

MCP

A specification open sourced by Anthropic to solve the connection problem between LLM and various datasources.