4.6 KiB
title, type, source-type, category, tags, date-added, video-source, audio-source, status
| title | type | source-type | category | tags | date-added | video-source | audio-source | status | |||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Public Cloud Learning Sessions (OpenText)- Generative AI & Prompt Engineering - 20241112 160112-Meeting Recording | cloud-learning | video | DevOps & SRE/09_Serverless_AI |
|
2026-04-14 | nas:///volume2/work/Public Cloud Learning Sessions/Public Cloud Learning Sessions (OpenText)- Generative AI & Prompt Engineering - 20241112_160112-Meeting Recording.mp4 | summarized (Gemini 摘要) |
Public Cloud Learning Sessions (OpenText)- Generative AI & Prompt Engineering - 20241112 160112-Meeting Recording
Source: NAS /volume2/work/Public Cloud Learning Sessions/Public Cloud Learning Sessions (OpenText)- Generative AI & Prompt Engineering - 20241112_160112-Meeting Recording.mp4
Type: VIDEO | Category: 09_Serverless_AI
Status: 🟡 Awaiting Whisper transcription → Summary
Generative A.I. and Front-Engineering
The learning session covers Generative A.I. on AWS, common use cases, AWS services like Amazon Q, Amazon Better or Amazon SageMaker, and the basics of Front-Engineering. Basic familiarity with Generative A.I. concepts and terminology is required.
Shikad Holtzman, a technical account manager based in Israel, discusses innovation opportunities, common use cases across industries, and how to make Generative AI applications valuable for business using data. The presentation includes AWS services and prompt engineering concepts.
Generative AI can create value by creating new experiences, boosting employee productivity, extracting insights, and fostering creativity. Use cases span customer experience (chatbots, virtual assistants), employee activities (code generation, summarization), business operations (document processing), and creative tasks (image generation). Amazon uses AI/ML, including Generative AI, for innovation, such as summarizing customer reviews on product pages.
Your data is your differentiator and it is what makes the difference between generic application to a specific application that can actually bring business to your value, to your business, sorry.
To create specific generative applications, techniques include retrieval augmented generation, fine-tuning, and continued retraining. Retrieval augmented generation is the cheapest and easiest, connecting multiple data sources without retraining the model. Fine-tuning involves retraining the model with labeled examples. Continued retraining adapts the model to a specific domain using unlabeled data.
AWS allows users to move quickly, use their own data, and scale using its global infrastructure. The AWS Generative AI stack has three layers: infrastructure, services (Amazon Bedrock), and applications.
Amazon SageMaker is a managed service for the entire life cycle of building, training, and deploying foundation models. SageMaker Jumpstart provides access to publicly available foundation models and third-party models. AWS also offers dedicated chips like AWS trainium and AWS inferencia for training and inference.
Amazon Bedrock is a fully managed service providing access to a wide range of foundation models from Anthropic, Meta, and Amazon (Titan models), including multi-modal and image models. Bedrock includes customization options, a managed RAG solution (knowledge bases), fine-tuning, continued training, agents, and responsible AI capabilities.
None of your data nor not the prompts, not the data that you are using for customizing the model is being shared with the model providers.
Guardrails for Amazon Bedrock allow users to filter harmful content based on their own policies.
Amazon Q is an AI-powered assistant with flavors like Amazon Q for business and Amazon Q developer. Amazon Q for business connects to multiple data sources for search, summarization, and insight extraction, maintaining existing permissions. Amazon Q for developer focuses on code generation, unit testing, and code migration.
Prompt engineering involves creating, designing, and optimizing prompts to guide the LLM's response, ensuring accuracy and relevancy. LLMs are trained on data created by humans, so prompts should consider human responses. The process is iterative, involving testing and refining prompts against use cases. Instructions should be clear, accurate, and specific.
Components of a prompt include instructions, context, user input, and an output indicator. Basic techniques include one-shot prompting and few-shot prompting, where examples are provided to the model. Chain of thoughts involves providing the model with step-by-step thinking to solve complex tasks.