AI/ML terminology for founders—prompts, tokens, embeddings, fine-tuning, RAG. 19 patterns for understanding and leveraging AI in products.
AI vocabulary patterns help founders understand and implement AI features. These 19 patterns cover prompts (instructions to AI), tokens (text units), temperature (creativity control), embeddings (semantic meaning), fine-tuning (custom models), RAG (retrieval-augmented generation), context windows, hallucinations, and more. You don't need to be an ML engineer to build AI products—but you need the vocabulary to make smart decisions and communicate with technical teams.
AI hallucination is when AI generates false or nonsensical information confidently. Learn how to detect and reduce hallucinations.
Zero-shot learning means AI performs tasks without training examples. Learn when zero-shot works and when you need few-shot.
AI persona assigns a role to the AI (expert, teacher, critic). Learn how personas improve output quality and tone.
JSON Mode forces AI to output valid JSON. Learn how to use JSON mode for reliable structured data extraction.
Top-K sampling limits AI to the K most likely next words. Learn how Top-K controls randomness and output quality.
Stop sequences tell the AI when to stop generating text. Learn how to use stop sequences for precise output control.
Temperature controls AI randomness. Learn when to use low vs high temperature for factual or creative outputs.
Context window is how much text AI can process at once. Learn context limits and how to work within them.
Few-shot learning gives AI 2-5 examples to learn from. Learn when to use few-shot vs zero-shot prompting.
Fine-tuning trains AI on your data to customize behavior. Learn when fine-tuning beats prompting and how to do it.
Prompt engineering is designing prompts to get better AI outputs. Learn techniques for writing effective prompts.
RAG combines AI with search to ground responses in your data. Learn how RAG reduces hallucinations and adds knowledge.
Embeddings convert text to numbers that capture meaning. Learn how embeddings power search, recommendations, and RAG.
Chain of Thought is a prompting technique that asks AI to think step-by-step. Learn when to use it, common mistakes, and how it improves AI accuracy for complex tasks.
Verbose prompting tells AI to give detailed, thorough explanations instead of brief answers. Learn when to use it, common mistakes, and how to balance detail vs. clarity.
Concise prompting tells AI to give short, direct answers without fluff. Learn when to use it, how to avoid missing critical details, and when to use verbose instead.
A system prompt sets the AI's role, behavior, and rules before the conversation starts. Learn how to write effective system prompts, common mistakes, and real examples.
Prompt chaining breaks complex tasks into smaller AI prompts that feed into each other. Learn when to use it, common mistakes, and how to build reliable AI workflows.
Token limit is the maximum text an AI can process at once (input + output). Learn how tokens work, how to stay under limits, and why it matters for cost and performance.
Structured output forces AI to return data in a specific format (JSON, XML, CSV). Learn when to use it, how to enforce schemas, and common mistakes to avoid.