// Free Guide

AI Without the Magic

A plain-English breakdown you can actually use — and share.

By Alex · Working Knowledge AI
// Part 1

What an LLM actually is

The anxiety around AI usually comes from one idea: that it's magical or human-like. It isn't. At a high level, an LLM is a function — the same kind you saw in high school algebra.

// The idea
f(x) = x + 2  — pass in 1, get 3. always.
"text in"LLM"text out"
Pass in "The cat…" → get "The cat sat". Bigger function, same idea.

What makes it feel magical is the sheer size — billions of numbers doing simple math at enormous speed. But it's still just a function.

// Part 2

ChatGPT is not an LLM

ChatGPT is a software application (an AI chatbot) wrapped around an LLM. Before your message ever reaches the model, ChatGPT intercepts it and loads it up with extra context.

ChatGPTLLM

ChatGPT adds: your conversation history, your account info, a timestamp, tone instructions, and more — before anything hits the LLM.

This is why the same question gives different answers. Under correct circumstances, the LLM is deterministic — the output doesn't change for a given input. ChatGPT changes the results.

What you typed
"Why do I have purple spots on my arm?"
What actually hit the LLM
The user asked: "Why do I have purple spots on my arm?" This appears to be a medical question. Respond in a concerned tone. Add medical boilerplate. Known facts about user: age 40, previous questions about hiking in Berkeley…
// Part 3

The randomness dial

LLMs have a setting called temperature. When picking the next word, the model considers several options. Temperature controls how adventurous it gets.

0MAXTEMPERATURE
Temp = 0

Always picks the most likely word. "The cat sat." Every time.

Temp = high

Rolls the dice more fairly. "The cat philosophized." More creative, less reliable.

Temperature is just a knob. It's not the AI "being creative" — it's a setting a human chose.

// Reference

Buzzword decoder

What they say vs. what it actually means.

Agent
The hype

Autonomous AI making its own decisions

The reality

A program that calls an LLM in a loop and runs functions a human wrote. The LLM produces text; the program pushes the buttons.

Hallucination
The hype

AI going rogue and seeing things

The reality

The LLM outputs something that sounds confident and coherent but is factually wrong. No perception involved — just a bad statistical guess.

Thinking
The hype

AI reasoning like a human brain

The reality

Billions of simple math operations running very fast. High school algebra, repeated at insane scale.

Training / Learning
The hype

AI teaching itself and evolving

The reality

Feeding data through the model, measuring how wrong the output is, and nudging billions of numbers slightly in the right direction. Mathematical optimization.

Bias
The hype

AI with prejudiced opinions or motives

The reality

Lopsided patterns in the training data carry through to the output. A statistical skew, not an emotional opinion.

Neural / Neuron
The hype

Digital brain modeled on the human mind

The reality

A math model loosely inspired by neurons — but barely comparable to actual biology. Don't let the name mislead you.

Temperature
The hype

How "hot" or emotional the AI is

The reality

A dial that controls randomness. At 0, it always picks the most likely next word. Higher values make it pick less obvious words more often.

AI is taking jobs
The hype

Unstoppable machine making decisions

The reality

Companies choosing to automate work. The AI didn't decide — a human did.

// Want more?

Come build something.

Upcoming in-person East Bay workshops where we go from zero to working AI tools. No hype. Bring your laptop.

Tell us about yourself →