In 1826, Nicéphore Niépce took the first photograph. The painters of his era were dismissive — mere mechanical reproduction, they said, no vision, no skill. Within two generations painting had survived but the business of painting had transformed. Portrait studios closed. Wedding artists retrained. Realism, as a goal, largely moved from oil to film.
The same thing happened when recording arrived for musicians, when typesetting arrived for scribes, when CAD arrived for drafting, when DAWs arrived for producers. Every creative tool resets the economic floor for what counts as hand-made and what counts as composed from existing work. The question is never whether the tool destroys creativity. It's who the tool moves from being a creator to being a craftsman, and who it lets through the door for the first time.
This course is about what AI does to creativity — to writing, music, visual art, photography, and the question of what we mean when we say a work is original. It covers consent and copyright, the mechanics of how generative AI actually makes images and words, and the practical skill of using these tools as a partner rather than a replacement.
If you finish every module, here's who you become:
On November 30, 2022, OpenAI released ChatGPT to the public. Within five days it had one million users. Within two months — one hundred million. It became the fastest-growing consumer application in history. Students around the world suddenly found themselves asking: What just happened? And how does this thing actually work?
Generative AI is a type of artificial intelligence that can create new content — text, images, music, code, and more — by learning patterns from enormous amounts of existing human-made content. The word "generative" simply means it generates, or produces, something new.
But here's the key thing to understand early: generative AI does not think the way you do. It does not have experiences, opinions, or feelings. It is an extremely sophisticated pattern-matching and prediction machine — and those predictions can look remarkably like creativity.
Imagine reading ten thousand novels and then being asked to write one more sentence that "sounds like" all of them combined. That's roughly what a language model does — except it has read hundreds of billions of sentences, and it does this prediction millions of times per second.
You've probably already used generative AI without realizing it. Here are three real systems you may have encountered:
| System | Company | What It Generates | Released |
|---|---|---|---|
| ChatGPT | OpenAI | Text, code, answers, essays | Nov 2022 |
| DALL·E 3 | OpenAI | Images from text descriptions | Oct 2023 |
| Suno | Suno AI | Full songs with lyrics & vocals | Dec 2023 |
Older AI systems were built with rules. A programmer would write: "If the user asks about weather, look up the forecast." These systems could only do exactly what they were told. They could not improvise.
Generative AI learns from examples, not rules. Instead of being told what to do in every situation, it figures out patterns from data. This makes it far more flexible — and far more surprising — than anything that came before it.
Traditional AI: programmed with specific rules. Generative AI: trained on examples to learn patterns. The shift from rules to learning is what made modern AI so powerful — and so unpredictable.
Understanding how AI works doesn't mean you need to build one. Knowing the basics makes you a smarter user — someone who can spot when AI gets things wrong, use it more effectively, and make better creative decisions alongside it.
You've just learned what generative AI is and how it's different from older AI. Now it's time to explore. Use this AI assistant to dig deeper — ask it questions, challenge it, or try to stump it.
In May 2020, OpenAI released GPT-3 — a language model trained on approximately 570 gigabytes of text scraped from the internet, including Wikipedia, books, and millions of websites. That's roughly equal to reading 570,000 novels. The AI was never explicitly taught grammar rules or facts. It discovered them by processing that mountain of text and learning what patterns appeared together.
Training a generative AI happens in stages. Here's how a language model like GPT goes from nothing to something that can hold a conversation:
Parameters are the numbers inside an AI model that get adjusted during training. Think of them like the settings on a very complex mixing board with billions of dials. Each dial gets tuned based on what the AI gets right or wrong.
GPT-3 had 175 billion parameters. GPT-4's exact count has not been officially released, but estimates suggest it is significantly larger. More parameters generally means more capacity to learn — but also vastly more computing power and energy required.
Imagine you're learning to bake cookies by trying thousands of recipes. Each time a batch burns, you adjust something: less sugar, lower temperature, shorter time. After enough tries, you develop an intuition for baking — not because someone told you the rules, but because your experience "trained" you. AI training works the same way, just with billions of examples and tiny numerical adjustments instead of cookie trays.
Training large AI models is expensive in real, concrete ways. In 2023, researchers estimated that training GPT-4 cost over $100 million in computing resources alone. The energy used was equivalent to powering thousands of homes for a year.
This is why only a small number of companies — OpenAI, Google DeepMind, Anthropic, Meta — can afford to train frontier models from scratch. Once trained, however, the model can be used by millions of people at a fraction of that cost.
Because AI learned from human-created content, its "creativity" is fundamentally shaped by what humans have already made. It can remix, recombine, and extrapolate — but it cannot step entirely outside the patterns it was trained on. This is both its power and its limitation.
You learned how AI trains on massive data, adjusts billions of parameters, and gets fine-tuned by humans. Now explore the messy details. What happens when training data is biased? How does fine-tuning actually change behavior?
In June 2017, eight Google researchers published a paper titled "Attention Is All You Need." It introduced the Transformer architecture — a new way to process sequences of text. At the time, few outside academia noticed. Within three years, this paper had become the foundation of GPT, BERT, Claude, Gemini, and virtually every major AI system in the world. It is now one of the most cited research papers in history.
A Transformer is a type of neural network architecture — basically, a design pattern for how information flows through an AI model. Before Transformers, AI systems processed text word-by-word in order, like reading a sentence left to right one letter at a time. This was slow and caused the AI to "forget" the beginning of long sentences by the time it got to the end.
Transformers solved this by processing all the words in a sentence at the same time, and — crucially — figuring out which words should pay attention to which other words.
Imagine reading the sentence "The trophy didn't fit in the suitcase because it was too big." What does "it" refer to — the trophy or the suitcase? You know instantly because you paid attention to the right word. Transformers learn to do exactly this: figure out which words are connected to which, even across long distances in a text.
The "attention mechanism" lets a Transformer model assign different levels of importance to different words when processing any given word. When the model is generating the word "she" in a story, it looks back at the entire context and figures out which earlier words — a character's name, a pronoun used before — are most relevant.
This happens in what are called attention heads. A large model like GPT-4 has many layers, each with multiple attention heads — and each head can learn to "attend to" different kinds of relationships. One head might track grammar. Another might track topic. Another might track named entities.
Every Transformer-based AI has a context window — the maximum amount of text it can "see" at once. Early models like GPT-2 (2019) had a context window of about 1,000 tokens. GPT-4 Turbo (2023) expanded this to 128,000 tokens — roughly the length of a full novel.
This is why AI sometimes seems to forget things from earlier in a very long conversation: if the conversation exceeds its context window, earlier messages literally fall out of what the model can see. It's not forgetting — it's that the information is no longer visible to it.
| Model | Year | Context Window | Roughly Equivalent To |
|---|---|---|---|
| GPT-2 | 2019 | ~1,024 tokens | A few pages of text |
| GPT-3 | 2020 | ~4,096 tokens | A short story |
| GPT-4 Turbo | 2023 | 128,000 tokens | A full novel |
| Claude 3.5 (Anthropic) | 2024 | 200,000 tokens | A very long novel |
The Transformer architecture is what enables AI to maintain coherent style across a long piece of writing, understand nuance in a poem, or generate music that has a consistent theme. Attention allows the model to "remember" (within its context window) that a story started with a particular character's voice and maintain it throughout.
It's also why AI-generated text sometimes still goes off the rails: no matter how large the context window, the model is always predicting the next most likely token — not planning ahead like a human writer would.
You don't need to understand the math behind attention mechanisms to use AI creatively. But knowing that AI generates text by predicting one piece at a time — without a plan — helps you understand why giving it a clear, detailed prompt produces better results. You're essentially giving it a stronger starting context to "attend to."
You've learned about Transformers, attention, and context windows. Now test these concepts. Try to see the attention mechanism in action — or find the edges of the context window. Ask technical questions or creative ones.
In June 2023, two New York lawyers filed a court brief containing six entirely fictitious case citations generated by ChatGPT. The cases did not exist. The judge names were invented. The legal reasoning was plausible-sounding nonsense. When confronted, ChatGPT insisted the cases were real. The lawyers faced sanctions. The incident became one of the most widely reported demonstrations of AI hallucination in history.
AI "hallucination" is when a generative AI confidently produces information that is completely fabricated. The term is borrowed from psychology but means something specific here: the model generates text that seems coherent and authoritative but is simply wrong — sometimes wildly wrong.
This happens because AI language models are not retrieving facts from a database. They are predicting plausible sequences of tokens. A plausible-sounding legal case name is statistically similar to real legal case names — so the model produces it. It has no way to check whether it's real.
Treat AI-generated factual claims like you'd treat a very confident classmate who hasn't studied. They might be right. They might be completely wrong. Always verify important facts from primary sources before using them in your work.
Despite its limitations, generative AI is genuinely powerful for certain creative tasks. Here is what it excels at — based on documented uses as of 2024:
Understanding the limits is just as important as knowing the capabilities — especially for creative work:
In February 2023, the U.S. Copyright Office ruled that AI-generated images cannot be copyrighted because copyright requires human authorship. In August 2023, a federal judge upheld this ruling in Thaler v. Perlmutter. However, images where AI was used as a tool by a human author may still qualify.
This is an active legal and philosophical debate — and it matters enormously for creative industries. For now, the practical answer is: you are the author when you use AI as a creative tool. The AI does not own or claim the work.
A photographer who uses a camera is the author of the photograph, not the camera manufacturer. When you use AI as a creative tool — providing the prompt, selecting the output, editing, curating — you are the creative agent. The AI is the very sophisticated brush.
AI is not your replacement. It is your collaborator — a powerful one that works best when you bring the intention, judgment, taste, and purpose. The students who learn to direct AI effectively will have a significant advantage in any creative field they choose to enter.
You now know AI can hallucinate, can't verify facts, and can't truly originate. But it can also do genuinely impressive creative work. In this lab, probe both sides. Try to spot the difference between AI confidence and AI accuracy. Discuss what it means for your own creative work.