Question 1

Token

Accepted Answer

A token is a chunk of text that AI processes. In English, 1 token ≈ ¾ of a word. "Hello world" is 2 tokens. API pricing is based on token count — the more tokens, the higher the cost.

Question 2

Context Length (Window)

Accepted Answer

The maximum amount of text an AI can handle in one conversation. 128K tokens ≈ 300 pages, 1M tokens ≈ 3,000 pages. Longer documents need larger context windows.

Question 3

Hallucination

Accepted Answer

When AI produces information that sounds plausible but is factually wrong. Mitigation strategies include verifying sources, using search-based AIs like Perplexity, or source-grounded tools like NotebookLM.

Question 4

Prompt

Accepted Answer

The text you type to tell AI what you want. Better prompts = better results. Being specific and clear dramatically improves output quality. This has led to 'prompt engineering' as a skill.

Question 5

Fine-tuning

Accepted Answer

Taking a general-purpose AI and training it further with specialized data to improve performance on specific tasks. For example, fine-tuning on medical data creates a healthcare-specialized AI.

Question 6

RAG (Retrieval-Augmented Generation)

Accepted Answer

A technique where AI searches external databases for relevant information before generating an answer. Perplexity's search and NotebookLM's source referencing use this. Highly effective against hallucinations.

Question 7

Multimodal

Accepted Answer

AI that can process multiple data types (modalities) — not just text, but also images, audio, and video. GPT-4o and Gemini are leading examples that can see images and hold voice conversations.

Question 8

Temperature

Accepted Answer

A setting from 0 to 1. Lower values (0.1-0.3) give consistent, factual outputs — good for analysis. Higher values (0.7-0.9) give creative, varied outputs — good for writing and brainstorming.

Question 9

API

Accepted Answer

Application Programming Interface — lets developers integrate AI into apps and services via code. Unlike using ChatGPT in a browser, APIs are for building products. Pay-per-use pricing based on tokens consumed.

Question 10

Input / Output Tokens

Accepted Answer

API billing is split: input tokens (your text) and output tokens (AI's response). Output is typically 2-5x more expensive. Prices are shown per 1 million tokens.

Question 11

Context Window (Billing)

Accepted Answer

With APIs, the entire conversation history is sent each time, and all of it counts as input tokens. Longer conversations cost more. Trimming history or summarizing old messages saves money.

Question 12

Open Source Model

Accepted Answer

Models with publicly available code and weights that can be self-hosted. LLaMA (Meta), DeepSeek, and Mistral are examples. Great for privacy since data stays on your servers, but requires GPU hardware.

AI Glossary

Basics(4)

Token

Context Length (Window)

Hallucination

Prompt

Technical(4)

Fine-tuning

RAG (Retrieval-Augmented Generation)

Multimodal

Temperature

Business(2)

API

Open Source Model

Pricing(2)

Input / Output Tokens

Context Window (Billing)

AI Glossary

Basics(4)

Token

Context Length (Window)

Hallucination

Prompt

Technical(4)

Fine-tuning

RAG (Retrieval-Augmented Generation)

Multimodal

Temperature

Business(2)

API

Open Source Model

Pricing(2)

Input / Output Tokens

Context Window (Billing)