A generative language model predicts what word most likely comes next, then does it again, one word at a time, until a whole answer has poured out.
If you remember one thing: generative AI writes one word at a time, by prediction. Genuinely powerful, not a mind. Both halves are true.
| Fact | Detail |
|---|
| Date | November 30, 2022 |
| What happened | OpenAI released ChatGPT to the public as a free website |
| Scale | A reported one million users within five days |
| What actually changed | Who could touch the technology, not the technology itself |
| How to treat it | A historical marker, like the launch of the web browser; the tools keep changing |
| Older AI | Generative AI |
|---|
| Job | Sorts into categories (spam or not, fraud or fine) | Makes new things: memos, poems, plans, images, code |
| Scope | One job each | Attempts almost any writing or thinking task |
| Interface | Built and tuned by specialists | Plain English |
Framing from the Harvard course, borrowed from a 2023 Goldman Sachs analysis: general, generative, approachable.
| Behavior you will see | Why it happens |
|---|
| Confident answers that are wrong | It predicts likely words; it does not look up verified facts |
| Different answers to the same question | It predicts rather than retrieves; the prediction insight explains the wobble |
| Replies that sound like a person | Trained on trillions of words of human writing |
| Answers pouring out word by word | That is the machine actually working, not a typing effect |
| Pitfall | Correction |
|---|
| Treating it like a search engine | Search retrieves pages that exist; a model composes new text that sounds right |
| Assuming it thinks like a person | One word at a time, by prediction; do not imagine feelings or intent |
| Deciding it is all hype after one wrong answer | A tool can be flawed and still be a generational change; both at once |
| Waiting until you feel ready | The interface is plain language; you are already qualified |
| Arc | Lessons | Question |
|---|
| Use it well | 2 through 5 | How do I use this well? |
| Judge what it means | 6 through 9 | What does it mean for my world? |
| Line | Meaning |
|---|
| ”It’s a complicated prediction problem but it’s still really just prediction.” | Sharad Goel’s one-sentence core of this lesson |
| ”humans mimicking machines mimicking humans” | Goel on the classroom exercise where students voted the next word |
| The admissions office is every office | Every use question eventually becomes a fairness, privacy, and work question |
