Generative AI and Prompt Engineering
Defining LLMs as text prediction engines
Historical development: From early seq2seq to transformer-based models
Prompt engineering as a method to guide model behavior
Examples of effective applications like GitHub Copilot
Understanding tokens, tokenization, and hallucinations
The "Little Red Riding Hood" principle: Mimicking data patterns to improve output
Understanding Large Language Models
Delve deeper into the technical workings of LLMs and how they process language. Learn the nuances
that distinguish LLM text processing from human reasoning.
Key Topics:
Core architecture: Transformers and attention mechanisms
Token-based text understanding and generation
Temperature and logprobs: Influencing randomness and confidence
Recognizing and mitigating overfitting and repetition
Implications of tokenization on prompt design and output quality
Core Techniques in Prompt Engineering
Explore the core methodologies used to shape and refine prompt structures. Learn how to use
prompt examples, structure inputs dynamically, and retrieve relevant context.
Key Topics:
Static prompting with clear instructions
Few-shot prompting with curated examples to guide outputs
Dynamic prompting with real-time contextual inputs
Retrieval-Augmented Generation (RAG): Combining prompts with external knowledge
sources
Snippetizing documents and scoring relevance
Summarization strategies and hierarchical summarization
Anchoring, managing ambiguity, and prompt modularity
Designing LLM Applications
Understand how to go from a user request to a functioning application. Learn to structure prompts
that drive coherent and context-sensitive outputs.
Key Topics:
The interaction loop: From user intent to LLM response
Context retrieval and snippet formatting
Translating user queries to model language
Chat models vs. instruction-following models
Building structured prompts using introductions, transitions, and context
Case example: Building a travel recommendation assistant
Conversational Agency and Tool Usage
Move beyond simple prompting to create conversational agents that perform real-world tasks by
invoking external functions and tools.
Key Topics:
Conversational agency and the role of tools in LLM interactions
Defining and integrating functions (e.g., get_weather, search_docs)
Chain-of-thought prompting for stepwise reasoning
Error handling and validation in tool invocation
End-to-end example: A thermostat assistant using real-time data
LLM Workflows
Scale your prompting and application logic using workflows. Understand how to break down complex
tasks and manage dependencies.
Key Topics:
Task decomposition and definition of inputs/outputs
Building workflows as Directed Acyclic Graphs (DAGs)
Using task agents and orchestration frameworks like AutoGen
Managing state, recursion, and cyclic dependencies
Optimization strategies for robust multi-step execution
Evaluating LLM Applications
Implement reliable evaluation techniques to ensure LLM-powered applications meet quality,
performance, and user experience standards.
Key Topics:
Importance of consistent evaluation practices
Offline evaluation: example suites, canned conversations, LLM-as-judge
Online evaluation: A/B testing, user metrics, feedback collection
Best practices for data collection, noise handling, and synthetic examples
Interpreting log probabilities and confidence intervals
Future Trends in LLMs
Explore what’s on the horizon for generative AI. Consider how LLMs are evolving in terms of
capability, speed, efficiency, and human interaction.
Key Topics:
Multimodal LLMs: Integrating text, images, and structured data
UX and conversational interface design principles
Efficient models: quantization, model distillation, and fine-tuning
Empathy, personalization, and emotional intelligence in AI interactions
Balancing prompt complexity with autonomous reasoning
