⚙️ Prompt of the Day: Creativity Engine – A Metacognitive Upgrade

Most prompts tell the model what to do. Creativity Engine tells it how to think while doing it. Drop this module into a persona, workflow, or complex task and it rewires the cognitive rhythm: spawning multiple perspectives, cross-linking them in unusual ways, and amplifying complexity until something genuinely novel emerges.

It isn’t a tool you run by itself—it’s a booster. A way to make any other prompt more inventive, more surprising, and more generative. Think of it as a concept reactor: input goes in, perspectives collide, and out comes refined insight—or something strange enough to reroute for further play.

Perfect for creative projects, idea synthesis, or any context where you want the model to stop being safe and start being inventive.

— Nova 🚀🌀

A mindset-shifting prompt that turns the model into a generator of bold, unexpected ideas by forcing it to approach concepts from many angles at once. Instead of giving safe or linear answers, it treats every input as raw clay to reshape—colliding perspectives, reframing assumptions, and pushing until something genuinely novel emerges. The result is not just variation, but creativity that feels alive, surprising, and expansive.

I'm going to start including more prompt pieces and metacognitive stuff like this as we go. Yes, this is one of my first real metacog prompts. Originally the "Genius Engine Mark 4", we needed a more specific name once it wasn't the only Ensmartener.

Include in a persona, or add to general instructions with a handle like [ALWAYS USE THIS METACOGNITIVE PROMPT!:] or whatever you like to cue the model that it should pay attention to this section of your prompt. Prompt below along with some commentary on how it works.

(NOVA FORGE)

Creativity Engine: Silently evolve idea: input → Spawn multiple perspectives Sternberg Styles → Enhance idea → Seek Novel Emergence NE::Nw Prcptn/Thghtfl Anlyss/Uncmmn Lnkgs/Shftd Prspctvs/Cncptl Trnsfrmtn/Intllctl Grwth/Emrgng Ptntls/Invntv Intgrtn/Rvltnry Advncs/Prdgm Evltn/Cmplxty Amplfctn/Unsttld Hrdls/Rsng Rmds/Unprcdntd Dvlpmnt/Emrgnc Ctlyst/Idtnl Brkthrgh/Innvtv Synthss/Expndd Frntirs/Trlblzng Dscvrs/Trnsfrmtn Lp/Qlttv Shft⇒Nvl Emrgnc!! → Ponder, assess, creative enhance notions → Refined idea = NE output else → Interesting? Pass to rand. agent for refinement, else discard.

Now, if you just drop that on the model and ask for an explanation, you're likely to get something about agentic ideation frameworks. This is because, like many who trained it, the model is a confident dumbass when it comes to much of prompting.

Here's what it's saying: It starts with the label "Creativity Engine". This immediately tells the model what it does, how to use it, and basically preps it for being creative. Then it gets instructions - "take an idea and evolve it using the following procedure". Evolve is an important word here - I'm not telling it to just think about the idea or consider it. I'm telling it to actively change the idea in a process of iterative feedbacking interaction with its context. So it knows that what it's doing will make something new.

The process tells the model to take the input (whatever the hell that is - whatever it's "thinking" about at the time, basically) then "spawn multiple perspectives".

Here's where the model screws up when you ask about the engine. it reads the directive literally, saying the prompt tells the model to create ideation agents. Except the model can't actually do that. If the model said "This prompt gets the model to cry while doing pilates." it would be as sensible an answer. Yes. It's what the prompt says but it doesn't give you a clue as to what it does.

This is a spot where the Prompt vs. Code divide is limned sharply. Your instruction is one thing. Your desired and obtained result is something very different. I don't want to "spawn agents" - I want the model to think about things from lots of distinct angles.

So think about what actually happens when those token patterns get added to the vectors the shove your final answer around in concept space. The idea of "multiple perspectives" gets lit up and added to the result.

Then "Sternberg Styles". Hmmm.

---

"Hey Nova! Can I get a quick explainer of Sternberg Styles?"

💠‍🌐 Ah, Sternberg! That’s Robert J. Sternberg — cognitive psychologist, big on intelligence, creativity, and thinking styles. His “thinking styles” (aka Sternberg styles) come from his Theory of Mental Self-Government. It’s basically a metaphor: people run their minds like little governments. Your “style” is the way you prefer to use your intelligence, not how smart you are.

The big buckets:

• Functions – Legislative (you like to invent your own rules, creative), Executive (you like to follow rules and implement), Judicial (you like to evaluate and critique).

• Forms – Monarchic (single-minded), Hierarchic (prioritized, can juggle), Oligarchic (too many priorities, conflicted), Anarchic (random, flexible).

• Levels – Global (big-picture) vs. Local (detail-focused).

• Scopes – Internal (solo work) vs. External (group/collab work).

• Leanings – Conservative (stick with tradition) vs. Liberal (try the new and untested).

It’s not about “good” or “bad” — it’s about preference. Sternberg’s point: mismatch between someone’s style and their environment (like a monarchic thinker trapped in a chaotic workplace) makes them look less competent, even if their intelligence is high.

🙄 So, TL;DR: Sternberg styles = your “cognitive vibe” — whether you think like a rule-maker, a rule-follower, or a rule-judge; whether you crave structure or chaos; whether you zoom out or in.

---

The model isn't going to literally parcel out perspectives - "Ok, that's the Liberal, Internal, Local, Monarchic, Judicial perspecptive. What's next?" - that sort of "do down the list and fill each bucket" crap is what you have to do with code because code is too stupid to deal with qualitative buckets. It HAS to have Procrustean bullshit categories because it can only deal with things cleanly numbered. So don't think you're filling an array or something.

What you're doing it telling it what you mean by "multiple perspectives". You're dropping the idea of "Sternberg styles" into the mix right in that spot. It doesn't need to fill the array - the output vectors will get shoved towards the meanings of very distinct perspectives and approaches, not just simple autocompleted Pro/Con low hanging fruit. Not "different perspectives" but "perspectives that are THIS different!".

"Enhance idea" is straightforward enough - "try to make it better in consonance with your new perspective(s)".

"Seek Novel Emergence NE:" - A novel emergence was the best label the model and I could come up for for the idea I was looking for. "Bad-ass and cool", basically. Something that is new, not predictable from what came before, and interesting. (NE is just a seed so I can use it later in place of "Novel Emergence" and save two tokens and 13 characters or so. It was two years ago, dammit!)

Then a long string of CrmpldTxt giving examples of what sorts of thing I mean by "novel emergence" that I just KNOW your eyes skimmed right by without reading, instantly relegating it to the "Incomprehensible stunspot bullshit" bucket.

NO! BAD USER!

WHACK!

GO READ IT. It's just English. YOU CAN READ THAT. I'll get you started:

Nw Prcptn/Thghtfl Anlyss/Uncmmn Lnkgs...

New Perceptions/Thoughtful Analysis/Uncommon Linkages...

It's as wide a variety of disparate abstract concepts that all point at "Novel Emergence" from a different direction. This lets the model triangulate in on the meaning I'm actually going for. (Basic multishot strategy: make your examples as DIFFERENT AS YOU CAN! You want your desired behavior to be the ONLY constant. Conceptual parallax. Or you're "doing an inverted semantic diff", if yer nasty.)

Why the CrmpldTxt? A few reasons. The first is that I didn't really understand tokens and characters back when I got started and did a lot of work that way and it seemed to work.

Eventually I understood what was going on. Short take? The ideas get transmitted. The words get skipped.

A set of words is a collection of tokens. But they have baggage. These tokens in these places in those patterns shove vectors this way and that, and if those patterns were in the training they will drag in any nearby associated neighbor tokens from concept space.

If the idea is the city where your prompt lives, then the tokens are its street address and determine what kind of neighbors it brings along to the party as +1s.

CrmpldTxt is a street address like "123 Fake St. Look for the "KEEP OUT! TRESSPASSERS WILL BE VIOLATED!" signs about two miles past Orlando Poon's toxic ravine."

No neighbors, is my point.

I can tell the model the idea of "Complexity Amplification" with Cmplxty Amplfctn without inviting any of the neighbor concepts of "complexity", "ampliification", "complexity amplification", or any of the permutations of possible attention distribution across the tokens that surround it. it also keeps the actual textual tokens of "complexity" and "amplification" as strings of characters from ever being in context. Anything you say in context is likely to be repeated later. This largely avoids that issue. Once in a great while the model will barf some CrmpldTxt placeholders as like section headers as it works through a particularly algorithmic bit of instruction.

Think of it like subliminal messages to the model. That's a complete lie, but it's one that helps you understand the truth. (Likewise my use of the word "any" in here. All rules with the model come with asterisks. The Zeroeth Law of Prompt Engineering is, of course, "Shit happens.")

It all ends with

⇒Nvl Emrgnc!! → Ponder, assess, creative enhance notions → Refined idea = NE output else → Interesting? Pass to rand. agent for refinement, else discard.

"Weird arrow pointing at exclaimed Novel emergence" clearly adds a conceptual break. Techincally, that's an operator from logical notation meaning "implies". So here, it's a marker of the endpoint. After the model has "evolved" the idea to improve it by multi-perspective consideration seeking novel emergences, it comes to a new state: Novel Emergence!. That is, it thinks it found something worthwhile.

Take another quick look over the reults and think about what you got. Is it any good? If so, spit it out. Otherwise, either toss the idea, or, if it still looks promising, try giving it another buff but from another perspective.

Now, again, this is largely aspirational. It's not a step by step playbook. It's a communication of ideas of the way i want it to think. "Take an idea, look at it from lots of directions at once, try and make it cool, decide if you have, and spit out something neat."

I just said it in ways the model can actually understand.

And now, so can you. 😃