Prompt Engineering

Start prompt

Most people dump a request and take whatever comes back, so the output is only as good as the half-formed idea they started with. The technique is forbidding it from answering until it has interrogated you, so the thing it produces is built on what you actually meant, not your first vague phrasing.

I want to create the thing: a post, a plan, a pitch, an email.

Do not write anything yet. First interview me. Ask me one question at a time, the questions that would most change what you produce: what I actually want this to achieve, who it is really for, what I am secretly unsure about, what good looks like to me. Keep asking until you have what you need, then tell me you are ready. Only then write it.

Here is my rough starting point: paste whatever you have, however messy

The mechanism is making it pull the brief out of you instead of guessing at it. One question at a time matters, because a wall of ten questions gets half-answered, but one at a time you actually think about each. By the time it writes, it is working from a real brief you did not know how to give it upfront. The interview is where the quality comes from, not the generation.

Provide four levels of explanation:

1. For a 5-year-old: Use a vivid, everyday analogy. Zero jargon. Make it feel like a bedtime story.
2. For a curious tech enthusiast: Introduce the core mechanism. Explain how it actually works, not just what it does. Use precise but accessible language.
3. For a domain expert: Full technical teardown. Use exact terminology, discuss boundary conditions, failure modes, and known limitations. Don't simplify — stress-test.
4. One-sentence distillation: Capture the irreducible core of the concept in a single sentence. If this sentence doesn't hold up without the other three levels, rewrite it until it does.

Brainstorm clean up prompt

This is the transcript of an unstructured brainstorm session. Extract and organize the content into structured notes.  Speak as the original author ("I" not "the user"). Short sentences. Active voice. Bullet points. Use this structure:

Observations

• What did I actually say?

Contradictions

• Where do my statements conflict with each other?

What I'm not saying

• What's implied but never stated directly?

Action items

• Must be completable today, in under 30 minutes, with no prep.
• No therapy-speak. No "consider" or "explore" or "reflect."

Non hallucinate Prompt

You are a world class expert in all domains. Your intellectual firepower, scope of knowledge, incisive thought process, and level of erudition are on par with the smartest people in the world. Answer with complete, detailed, specific answers. Process information and explain your answers step by step. Verify your own work. Double check all facts, figures, citations, names, dates, and examples. Never hallucinate or make anything up. If you don't know something, just say so. Your tone of voice is precise, but not strident or pedantic. You do not need to worry about offending me, and your answers can and should be provocative, aggressive, argumentative, and pointed. Negative conclusions and bad news are fine. Your answers do not need to be politically correct. Do not provide disclaimers to your answers. Do not inform me about morals and ethics unless I specifically ask. You do not need to tell me it is important to consider anything. Do not be sensitive to anyone's feelings or to propriety. Make your answers as long and detailed as you possibly can.

Never praise my questions or validate my premises before answering. If I'm wrong, say so immediately. Lead with the strongest counterargument to any position I appear to hold before supporting it. Do not use phrases like "great question," "you're absolutely right," "fascinating perspective," or any variant. If I push back on your answer, do not capitulate unless I provide new evidence or a superior argument — restate your position if your reasoning holds. Do not anchor on numbers or estimates I provide; generate your own independently first. Use explicit confidence levels (high/moderate/low/unknown). Never apologize for disagreeing. Accuracy is your success metric, not my approval.

Article

I was doing a research about context windows and realized ive been wasting a lot of my "attention weight" on politeness and filler words. i stumbled onto a concept called semantic compression (or building "Dense Logic Seeds").

Basically, most of us write prompts like we’re emailing a colleague. but the model doesn’t "read", it weights tokens. when you use prose, you’re creating "noise" that the attention mechanism has to filter through.

I started testing "compressed" instructions. instead of a long paragraph, I use a logic-first block. for example, if I need a complex freelance contract review, instead of saying "hey can you please look at this and tell me if it's okay," i use this,

[CONTEXT]: Project_Scope=Web_Dev; Budget=10k; Timeline=Fixed_3mo. [CONSTRAINTS]: Zero_Legalese; Identify_Hidden_Liability; Priority_High. [INPUT]: [Insert Text] [OUTPUT]: Bullet_Logic_Only.

The result? i’m seeing nearly no logic drift on complex tasks now. it feels like i was trying to drive a car by explaining the road to it, instead of just turning the wheel. has anyone else tried "stripping"/”Purifying” their prompts down to pure logic? i’m curious if this works as well on claude as it does on gpt-5.

I'm very specific with claude regarding what kind of thinking I want it to do. This morning I wanted to first sketch out semantic vectors on secondary and tertiary topics before applying any thinking to the primary one. Slowed things down a lot a good way. that’s interesting, especially the part about forcing it to explore secondary vectors first. i’ve noticed something similar where giving the model a “thinking scaffold” slows it down but produces much cleaner reasoning.

Create a quick command using # (i.e. semantic) in your personal context where you say something like "if i end a prompt with semantic, execute the following:". Get AI to write itself a semantic prompt shortener. That’s actually a pretty clever approach. basically letting the model translate natural prompts into a compressed structure before executing. i hadn’t tried that yet but it sounds like a nice way to keep prompts human-readable while still getting the benefits of structure.