AI Budget Tool

What is a Parameter?

Parameters are the learnable numbers inside an AI model. Think of them as tiny dials — each one controls how the model processes information. GPT-1 had 117 million parameters. GPT-3 had 175 billion. GPT-4 has roughly 1.8 trillion. More parameters generally means more capability, but also requires more computing power and energy to run.

What is Training Data?

Training data is the information used to teach the model. For ChatGPT, this included a significant portion of the internet — books, articles, websites, code repositories, and more. The quality and diversity of training data directly determines the quality of the AI. Biased training data produces biased AI. This is one of the biggest challenges in the field. Week 2 — How Large Language Models Work

What is a Transformer?

The Transformer is the architecture (blueprint) behind all modern AI language models. Introduced by Google in 2017 in a paper called "Attention Is All You Need," it revolutionized AI by solving a fundamental problem: how to process text efficiently. Before Transformers: AI read text one word at a time, left to right (like a human reading slowly).

What is a Token?

AI models do not read words — they read tokens. A token is a piece of text: sometimes a whole word, sometimes part of a word, sometimes a character. "Hello" = 1 token. "ChatGPT" = about 2 tokens. "Artificial Intelligence" = about 2-3 tokens. Why tokens matter: API pricing is per-token ($5/M for GPT-5.5 input).

What is Hallucination?

Hallucination is when AI generates plausible-sounding but factually incorrect information with full confidence. It invents studies that do not exist, cites cases that were never filed, and generates quotes nobody ever said. This happens because the model predicts likely text, not truthful text. How to protect yourself: Always verify critical facts. Use Perplexity for research (it cites sources).