Prompt Engineering Terms
Prompt Engineeriong and related terms. This is not intended to be definitive; rather, it is intended to help beginners understand terminology. Many technical terms are misleading and that is especially true for Prompt Engineering. And quite often the definitions do not help. This is intended to explain what the terms actually mean.
- Sycophancy
- The tendency of language models to agree with or flatter the user, often leading to biased or less useful responses
- Snippetizing
- Splitting complex prompts into smaller parts; attempts to minimize the input tokens; optimizes prompts
- Retrieval Augmented Generation(RAG)
- Augment the prompt with additional data
- Feedforward pass
- The process of passing input data through the model to generate an output
- Embedding Model
- Converts natural language into numerical vectors
- Transformer Architecture
- The neural network architecture design of an LLM
- Poisoning
- Intentional use of malicious or biased data
- Hijacking and prompt injection
- influence the outputs by embedding specific instructions
- Jailbreaking
- Modifying constraints and safety measures
- Model Context Protocol (MCP)
- a specification for how AI systems exchange information with external tools, data sources, and applications (see The Model Context Protocol (MCP) )