Outcomes

What You Will Be Able to Do

Explain in plain words what an LLM computes, and why "it predicts the next token" changes how you should use it.
Train a small neural network by hand: adjust weights, read a loss function, and run gradient descent until it learns.
Use attention and context windows to your advantage when you feed documents to a model.
Control randomness with temperature, and know when you want a boring answer.
Write prompts that reliably produce useful work output: summaries, drafts, analysis, and code.
Catch hallucinations before they cost you money or credibility, using a simple verification rule.
Decide what is safe to paste into which tool, and avoid the mistakes that made global headlines.

The Evidence

What the Research Says

40%

less time and 18% higher quality on professional writing tasks for workers using an LLM assistant, in a randomized experiment.

Noy & Zhang, Science 381 (2023)

55.8%

faster completion of a programming task by developers using an LLM coding assistant in a controlled trial.

Peng et al., "The Impact of AI on Developer Productivity" (2023)

+34%

productivity gain for the least experienced customer support agents given an LLM assistant. Experts gained far less. Skill transfers down.

Brynjolfsson, Li & Raymond, NBER 31161 (2023)

$5,000

sanction against lawyers who filed a brief with six fake cases invented by ChatGPT. One skipped step, checking that the cases existed, turned a routine filing into a global story.

Mata v. Avianca, S.D.N.Y. (2023)

Curriculum

7 Lessons, 10 Minutes Each

Sign up free and every lesson unlocks. Each lesson opens as its own full page in the classroom, with its animations, labs, and games, and your progress is saved as you go. Finish with the 20-question exam to earn your certificate.

1 What an LLM Actually Is Tokens, parameters, and the one sentence that explains everything • Tokenizer lab • Next-word game 10 min › 2 The Engine Room: Train a Network by Hand Weights, loss, and gradient descent, learned by moving them yourself • Neural network playground 10 min › 3 Attention and the Context Window How the model decides what matters • Attention visualizer 10 min › 4 Temperature: Why the Same Question Gets Different Answers Probability, sampling, and controlling randomness • Temperature lab 10 min › 5 Prompting That Works Role, context, task, format, examples • Prompt duel game 10 min › 6 Hallucinations: When Plausible Beats True Why models invent facts and how to catch them • Spot-the-fabrication game 10 min › 7 Safe and Secure: LLMs With Your Company's Data Data leakage, tool tiers, prompt injection • Safe-to-paste game 10 min › ★ Final Exam & Certificate 20 questions • 80% to pass • your name on the certificate 20 min ›

Demystified

Glossary: Every Term, Plain Words

LLM: Large language model. A neural network trained on huge amounts of text to predict the next token. ChatGPT, Claude, and Gemini are products built around LLMs.

Token: The chunk of text a model reads and writes, roughly three-quarters of a word in English on average.

Parameter / Weight: A learned number inside the model. Billions of them together encode everything the model "knows".

Training: Adjusting the weights, using gradient descent, so the model gets better at predicting text. Done before you ever meet the model.

Inference: Using the trained model to generate answers. What happens when you chat. The weights do not change.

Loss function: The single number that measures how wrong the model currently is. Training pushes it down.

Gradient descent: Repeatedly nudging every weight a small step in the direction that reduces the loss. The engine of all modern AI training.

Attention: The mechanism that lets the model score how relevant every other token is when processing each token.

Transformer: The neural network architecture built on attention (2017). The T in GPT.

Context window: The maximum tokens the model can consider at once. Text outside it is invisible.

Temperature: The knob controlling randomness in word choice. Low = consistent, high = varied.

Few-shot prompting: Including one or more examples of the output you want inside your prompt.

Hallucination: Confident, fluent output that is factually wrong. A structural side effect of predicting plausible text.

Fine-tuning: Additional training on a smaller, specialized dataset to adapt a general model to a specific task or style.

RAG: Retrieval-augmented generation: the system searches your documents first and pastes the relevant passages into the context window so answers cite real sources.

Prompt injection: Hidden instructions inside content the model reads, designed to hijack its behavior.