Skip to content
Clawdemy

Summary: Asking well, the anatomy of a good prompt

You paste a job posting, type “write a cover letter for this job,” and get three tidy paragraphs that could have been written about anyone, by no one in particular. The machine did exactly what you asked; one sentence of guidance buys bland mush. This lesson is the fix: the small set of parts a good prompt is made of, and why the same question can get different answers twice.

Core ideas

Section titled “Core ideas”
  • A strong prompt has three parts, TIC for short. The task is “what you want the AI to do.” The instructions are “how you want the AI to do it.” The context is “what you want the AI to know.”
  • Context is the part almost everyone skips and the part that helps most. The machine is a prediction machine, and context is the material it predicts from. Starve it, and it predicts the most average document on the internet; feed it, and it predicts something shaped like you.
  • Keep the course’s habits on a sticky note: start simple and build through conversation; be specific; say what to do rather than what not to do, because naming the thing you want avoided puts those very words into play; ask for many options; fence off pasted material, for example between triple quote marks; start fresh when a thread turns to sludge; and when writing the prompt feels hard, ask the machine to draft the prompt for you.
  • The fresh-thread trick: the machine rereads the entire conversation every time it replies, so old dead ends steer new answers. Ask a tired thread to summarize its useful parts as a single prompt, and carry that into a clean one.
  • Why answers vary: at each word the machine holds a ranked menu of plausible next words, and most chat tools deliberately let it pick among the strong candidates. Think of weighted dice rolled at every word. Thread history and frequent tool updates stack on top.
  • One answer is a sample, not a verdict. One of the course’s teachers compares judging the technology by one weak reply to “having the first result of your first Google search color what you think of the internet.” Ask again, rephrase, request a second version.
  • Variability is fuel. In December 2023, Google DeepMind researchers reported using a language model to make a genuinely new discovery on a long unsolved math problem: huge numbers of candidate programs, a separate program scoring them, until solutions emerged that beat the best previously known. Generate plenty and pick the best works for naming a workshop too.
  • When you need consistency instead of surprise, tight instructions and concrete examples narrow the menu at every step. Your prompt cannot switch the dice off. It decides which game the dice are playing.
  • Upgrade one, ask for steps. Researchers call it chain-of-thought prompting. The logic is human: a hard problem worked step by step on paper goes right more often, and a 2022 research paper showed the same pattern in language models. In the course’s worked example, rating forty reader comments on a Boston Globe article with a reason for every rating revealed the machine’s hidden yardstick: it was scoring extreme language, so a quietly seething complaint got rated moderate.
  • Upgrade two, show examples. Researchers call it few-shot prompting. Two labeled samples, one moderate and one intense, moved the seething comment to intense on the rerun. Two or three examples can outperform a paragraph of description.
  • How this has aged: when the course was filmed in 2024, telling a model to think step by step measurably improved answers. Since then, many newer systems, often described as reasoning models, do much of that stepwise work on their own. What did not age: visible steps let the work be checked, though a written explanation is itself output to check, not a certified window into the machine.
  • The persona move, which the course calls “giving an AI personality”: assign a role (mentor, critic, tutor, student, interviewer, simulator) and talk with the character. Four tips: open with the words “you are”; say you will speak turn by turn; give the character a spine so it does not cave instantly; and end by asking the machine to step out of the role and critique your performance.
  • A persona changes the machine’s style and stance, not its knowledge. Rehearse with the character. Verify with sources.

What changes for you

Section titled “What changes for you”

The first-answer trap now has an explanation: one weak reply is a dice roll, not a verdict on the technology or on you. And prompting turns out to be clear thinking made visible. Task, instructions, context is also how you brief a colleague, so practicing on a machine that never gets bored quietly makes you better at delegating to people. The next lesson steps beyond the chat window: why some assistants arrive already knowing their job, how an AI can work from your own documents, and what changes when these systems can see images and speak out loud.