05 — Work Through a Problem with AI

You can ask the assistant to help with the work itself, not just write a prompt. Bring an unfinished problem and use the conversation to find a useful way forward.

Try a guided conversation

Copy this block, replace the one bracketed field with your situation, and send it. If you have a real task, use it. Otherwise use: “I want to organize a fictional neighborhood cleanup, but I do not know where to start.”

Copy this prompt

I am trying to [WHAT I WANT TO DO OR WHERE I AM STUCK]. Help me think it through.

Ask one useful question at a time and wait for my answer. Help me clarify the result I want, the facts we have and what is uncertain. Explain unfamiliar terms with an example. Suggest options and a small next step when there is enough information; let me choose. Do not invent facts or carry out external actions.

Answer in your own words. You can say “I do not know,” “Show me an example,” “What are my options?” or “That suggestion does not fit because…” For the fictional cleanup, use Saturday, 10 a.m., park entrance and gloves if relevant; a park name and end time are unknown.

When a next step seems useful, choose it and ask the assistant to help do that part in the chat. For example: “Let us start with an invitation. Draft it using only those facts.” Read the result, compare facts and request the changes you want. If the conversation circles without helping, ask for a simpler next step or a summary of what is still missing.

Turn the approach into a prompt when useful

You can begin with a rough goal. The assistant can help turn it into a request, but you decide whether its questions and proposed instructions fit.

Start with your unfinished idea

Use a new chat for a different task, or stay in your current chat if you are developing the same task. Copy this block, replace the one bracketed field with what you want help doing, and send it.

Help me describe the task

Copy this prompt

I want help with [MY ROUGH GOAL], but I do not know how to ask for it well.

Help me make a useful prompt. Ask one question at a time about the result I need and any important facts or constraints. Wait for my answer. Once you have enough, draft a short prompt I can review. Put the proposed prompt by itself so I can copy it. Do not invent my situation or claim a long prompt is necessarily better.

No goal in mind? Replace the field with: “writing a fictional neighborhood-cleanup invitation people can act on.” For this example, use these facts if asked: Saturday, 10 a.m., park entrance, bring gloves. No park name or end time is known. Answer each question in the same chat. You can say “I do not know” or “That detail does not matter.” Do not invent facts just to fill an answer.

Choose the instructions, then try them

Read the proposed prompt. Does it describe your goal? Does it add a requirement you did not want? Ask it to remove or change that part. When you accept the prompt, copy only the accepted prompt text, not the surrounding explanation, into a new chat, supply any material it asks for, and send it. A new chat lets you test whether the prompt carries enough context on its own.

Check the prompt using its actual result

Copy this prompt

Review the prompt and the result it produced against what I wanted. Which missing fact, vague instruction or unnecessary demand affected the result? If the prompt worked well enough, say so. Suggest a specific change for my decision; do not rewrite automatically.

In the chat containing your prompt and result, send that check. If your goal is missing from the new chat, start your check message with a sentence describing it. Choose any useful change, revise the prompt and try it again. A polished prompt is only a proposal until you see what it produces.

Next: lesson 06 shows how to supply notes or a file instead of making the assistant guess.

This is alpha material. Report unclear instructions or errors through a public issue or email help@bigleverinstitute.org. Share only redacted or fictional material.