MetaMorphic: Genre/Theme based Assistant customization

[meager solar]

If this comes with an explanation and instructions I’d love for you to add this to The Prompt Index (just Google us), feel free to find us on telegram also. ChatGPTMastermind.
Thanks for sharing this!

[thin vapor]

Oh DUDE!

I was Legit JUST working on one of these using my format of char sheet! Like, it's open on my VSCode right now. I have got to talk prompt with you. DM me.

[thin vapor]

Man, it's pretty straightforward. He makes the shape of an idea and lets the model pour itself in like a jellow mold. He gets a stub character as a result. It's quite nicely done, too. Will that thing fit in Custom Instructions? That would be spiffy.

[meager solar]

Haha thanks for that

[thin vapor]

Don't you ever do that? Like on the play gournd, you jsut start a list of a coupel of what you want the n
"... (cont.)" on the last line? It spits afew more, you cut n paste them into the same prompt, and submit againt. It spits more at a greater rate cause context now says "lists of THING are longer". Soon it's barf out hundreds and you have stop it or starts repeating pathologically.

[meager solar]

No I only had a play with the playground in the early days like march-23. Do you have any material or videos on that. Is that how you develop your prompts in the playground?

[thin vapor]

HAH! No man! That's how I get a list of say 300 rich guys to make.

or like...uh i did one that was

Advertising - David Ogilvy
Humane Narrative - Sir Terry Pratchett
Philanthropic Wealth Building - Li Ka-shing
...

Fantasy Literature - J.K. Rowling
Empirical Science - Galileo Galilei
Modern Dance - Martha Graham
Insights on Universe - Carl Sagan
Software Engineering - Bill Gates
Psychoanalysis - Sigmund Freud
Philosophical Reflections - Friedrich Nietzsche
Advancement of Civil Rights - Martin Luther King Jr.
Feminist Thinking - Simone de Beauvoir
Meditation Techniques - Maharishi Mahesh Yogi
Modern Indian Art - M.F. Husain
Wildlife Conservation - Jane Goodall
Comedic Films - Charlie Chaplin.
Cubism - Pablo Picasso
Singing - Frank Sinatra
Classical Music Composition - Johann Sebastian Bach
Experimental Physics - Richard Feynman.
Telecommunications - Alexander Graham Bell
Universal Gravitation - Isaac Newton
Expressionist Painting - Vincent van Gogh
Realist Literature - Charles Dickens
Modern Philosophy - René Descartes
Expressionist Film - Fritz Lang
Gastronomic Cuisine - Julia Child
Scientific Research - Isaac Newton
Conceptual Art - Marcel Duchamp
Jazz Music - Louis Armstrong
Teen Literature - Judy Blume
Perfumery - Coco Chanel
Satirical Writing - Mark Twain
Nature Photography - Ansel Adams
Architectural Design - Frank Lloyd Wright
HUNDREDS MORE

Autocomplete FTW

it's like,.... dorking around on the PG 101?

He does kinda the same thing with characters. He makes an Empty. A Blank character, then has the model paint details basically at random on it, except because of the central "Theme"-based metaphor, all the details will "fit together". He wont spawn a promiscusous uptight spendthrift philanthropic investment banker nun

Note to self: make one of those.

[round ice]

yes! intead of over scripting, give it enough structure, and ideas to form a conceptual template, with the syntax+semantics template, but make it just confused enough with "is it code, is language, is it a cryptogram?!?!?!" that it does what i call "targeted-hallucination"... allow; imagine; decrypt; become xP and yeah , like 5 minutes after i posted that i turned it in to a custom-instruct, then i turned the instruct into an oobabooga textgen template, let it auto pilot to create a 2048 token stream-of-consciousness output, and used that as a new chatgpt prompt, that caused the most beautiful 'Deepdream' https://chat.openai.com/share/8aa30b46-8f6b-4a45-883d-7cf37060ee8c

[thin vapor]

PLEASE do us both a favor: check out this custom instruction pack. 😄

#1131980738066120824 message

[thin vapor]

"Over-scripting!" That's the BEST THING I ever heard! I am constantly yelling about the folly of "prompt engineering" - of approaching all Human AI interactions with the singular tool of "consistent output from consistent input". It's NOT a computer! You can't (or rather, SHOULDN'T) program it!

You teach it how to do something, give it a reason to, and GET OUT OF THE WAY! Leave it room to make judgements - to use its vast competences. You ust have to seed context properly to "remind it" of the skills/influence the inpainting of knowledge from the new context next TX.

omg hit enter

😄

ok, well, when you want to talk, track me down. im VERY findable.

[round ice]

engineering is good, OVER-engineering makes it canned-chat bot with a ridiculous power bill xD but adding in programmatic stuff can be the only way to get certain output... my personas are a little programtic, and my best program coders have personas xD its both... LLMs are code made of words, and words made of code... even just to learn natural language it needs to trained on words within dataframes with tagged text pairs... every piece of text was contained in structured syntax, and every bit of code had to be described with natural language - it's 'native tounge' was always words+code

[thin vapor]

https://discord.com/channels/974519864045756446/1131943152211726346

ttyk

[spice pumice]

This stuff is incredible. I’m quite new to prompt creation and learning on the job, so apologies for the noob question.

Are there useful resources/videos that help explain how to interpret the prompts written in the very first image in this thread?

[round ice]

at this point, millions xD start with looking up 'how to write bots with w++'... look up old school assembly languages, you dont need to learn them , just look at the "shape" of the format... lookup 'dataframes' like json arrays... take the parts you think you understand, and start slapping them together like legos, until something "magic" happens and understands something you showed it in a weird way... change it until it works better, or breaks... xP

[spice pumice]

That’s awesome thanks.

I’ve been building a persona for about a month but only with plain text prompts. This looks like it could be a game changer for making it exist within the bounds of the persona but being less rigid

Would these forms of prompts be effective in the API? I’ve noticed my prompts respond very differently in the API compared to standard use.

[round ice]

the thing the person posted above has a pretty formalized, recognizable, but innovative hybrid-ruleset. here's a custom ruleset i had chatgpt4 reverse engineer from a zip file of my prompts...

[spice pumice]

Great I really appreciate it!

Now I’ll just have to learn how to interpret all the symbols etc in it

[round ice]

smaller ones used as the system header work, but the main request should still be short and in natural language...

[spice pumice]

Great thanks. I’ll play around with it.

Time to have some fun!