In our last session on knowledge and prompt engineering, we focused more on the knowledge design side. In this follow-up session, we will explore prompt approaches and practices more deeply.
We will cover the following topics:
- Structured prompting, based on use cases, scenarios, and the RACE framework (Role, Action, Context, and Examples)
- Situational factors that affect the prompt framework
- Capability – Are you interacting with a chatbot or an assistant?
- Model – Which LLM are you working with: OpenAI's GPT, Mixtral, Llama2, or another?
- Use case – Do you need translation, generation, or a Q&A system?
- Mode – Are you working with text, vision, or voice?
- Action – Function calling or fine-tuning?
- How the prompt design framework needs to align with and take signals from the knowledge engineering framework
- Approaches to optimize your prompts and achieve the best responses from an LLM
- Working through problems step-by-step using the Tree of Thought (ToT), chain of thought, or iterative prompting techniques
- Injecting emotion to prompt emotionally engaging language
- Including signals to improve LLM performance:
- Corporate signals such as brand guidelines, terminology, and data sources
- User signals from prior interactions
- Industry-specific data
Speakers
- Seth Earley
CEO and Founder, Earley Information Science
- Nick Usborne
Copywriter, Trainer, and Speaker
- Sanjay Mehta
Principal Solution Architect, Earley Information Science