Tokens, Context Windows & Other Nerd Shit You Need to Know

Okay, there's some technical vocabulary you need to understand. Not because you're going to be building AI models, but because you'll hear these words constantly and you need to not look like an idiot.

TOKENS: This is how AI measures text. Roughly 1 token = 3/4 of a word. When someone says a model has a '200K context window,' they mean it can handle about 150,000 words at once. Why do you care? Because it determines how much you can feed the AI in one shot. Claude has a massive context window - you can upload entire contracts, spec documents, RFPs.

CONTEXT WINDOW: Think of it as the AI's working memory. Everything in the conversation - your prompts, the AI's responses, any documents you uploaded - all of that takes up space in the context window. When you max it out, the AI starts 'forgetting' the beginning of your conversation.

HALLUCINATION: When the AI confidently states something that's completely wrong. This is the #1 risk. The AI doesn't say 'I'm not sure' - it just makes shit up with the same confidence as when it's right. ALWAYS verify important facts, numbers, and claims.

RAG (Retrieval-Augmented Generation): Don't let the fancy name scare you. It just means giving the AI specific documents to reference before it answers. Instead of relying on its general training, it pulls from YOUR files. This is how you make AI actually useful for your specific business.

FINE-TUNING: Training an existing AI model with your own data to make it better at your specific tasks. Think of it as hiring a general contractor and then training them specifically on your type of projects. You probably won't do this yourself, but you should know what it means.

API (Application Programming Interface): A way for software to talk to other software. When apps connect to AI in the background, they use an API. You'll hear this word a lot when people talk about automation.