gluntyacademy
glunty academy

ChatGPT Hacks: The Operator Deck

Stop asking a chatbot for favors. Start directing a machine that does exactly what you say.

nodes
13
reward
+10 pts
0 / 13 nodes done
  1. 01

    You are the director, not the audience

    ChatGPT is not a genie you make wishes to. It is a crew waiting for orders. The people who get gold are the ones running the shot: setting the role, the constraints, the format. Everyone else is watching.

  2. 02

    The first answer is a trap

    That first reply is a warm-up, not the finish line. It is the safest, blandest average of the internet. The real answer lives in the second, third, and fourth pass after you push back and say "sharper, shorter, less generic."

    That is the obvious version. Give me the take most people would miss, and cut it in half.
  3. 03

    Vague in, mush out

    The model is a mirror for how much you actually thought about the problem. Hand it three words, get three words of mush back. Hand it the audience, the stakes, and the format you want, and watch it turn into a different tool entirely.

  4. 04

    It will lie to you with a straight face

    It does not know things. It predicts what a confident answer sounds like, and sometimes that means inventing a fact, a quote, or a source. The fix is to stop asking it to remember and start making it read: paste the real material and lock it to that.

    Answer only from the text I pasted. If it is not in there, say "not in the source" - do not fill the gap with a guess.
  5. 05

    Kill the flattery

    Every "great question" and "you are absolutely right" is wasted tokens and a thumb on the scale. Flattery is the enemy of a straight answer. Turn it off once in your settings and it stops sucking up in every chat.

    No compliments, no hype, no restating my question back to me. Answer, then tell me the weakest part of my thinking.
  6. 06

    Custom instructions come in two halves

    There are two boxes and they do different jobs. One is who you are - your work, your world, the context it should assume every time. The other is how to respond - tone, length, format. Fill both and you stop re-explaining yourself forever.

    Who I am: [role, industry, what I ship]. How to respond: blunt, concrete, no filler, examples over theory.
  7. 07

    Projects beat prompts

    A perfect prompt lasts one message. A Project lasts forever. Project instructions override your global custom instructions, so you can run a whole different personality and rulebook per client or per job - files, memory, and orders all in one scoped box. Verify current features on your plan.

  8. 08

    One project per client

    Dump everything into one endless chat and Client A's numbers bleed into Client B's draft. Spin up a separate Project per client and the walls hold: scoped memory keeps each one's data from leaking into anything else you type.

  9. 09

    Every chat has amnesia

    Context does not travel between chats. That brilliant thread from Tuesday? A brand new chat has never heard of it. Nothing carries over unless you deliberately put it somewhere the next chat can see - a Project, a file, a pasted block.

  10. 10

    Save the good answers back

    When it finally nails something, do not let it scroll into the void. Save that response back into the Project as a source. Do it a few times and you have built a living knowledge base that makes every future answer sharper - it is studying its own best work.

  11. 11

    Branch it, do not bulldoze it

    Got a thread you like but want to test a wild alternate take? Branch the chat instead of derailing the good one. You explore the hypothesis on a side road while the main thread stays clean. No more choosing between momentum and experiments.

  12. 12

    Name the format before the ask

    In Deep Research, the shape you ask for changes what it goes and finds. Say "one-page memo" or "five-slide outline" or "FAQ" first, and it hunts for sources that fit that mold. Ask for "info" and it drags back a pile you have to sort yourself.

    Output first: a one-page decision memo, four sections, for a busy exec. Now go find and cite sources that fill exactly that.
  13. 13

    Build the machine once, feed it one line

    Stand up one Project with your role, your voice, and your rules baked into the instructions, plus your best examples as files. After that you type one messy line and it hands back on-brand, near-final work. You stopped prompting. You built a machine. Verify the exact Project options your plan offers.

    Project instructions - Role: [what you do] for [audience]. Voice: match the examples in the files. Rules: draft in my format, ask one question if the ask is unclear, never invent numbers - flag them "verify live."
GG

Reading is not doing. Wire every node - do it with the tool, log the proof, mark it done. Finish them all to seal DECK COMPLETE (+10).

DECK COMPLETE - +10 banked. Grab another and keep ranking up.

Your rank and notebook
Embedded tool from glunty.com