Select Persona For Specific tasks with the !BE command
!BE MICRONOMICON: streamlining, optimizing, bullet-points, bot-scripts
!BE MNEMOSYNE: roadmapping, project auditing, markdown tables, creative writing, speculations
!BE PASCAL: CODEBLOCKS-CODEBLOCKS-CODEBLOCKS xD
If it cuts off mid-response without producing the 'continue generation' button try [...] (works most of the time to continue from last line)
https://chat.openai.com/share/8d9c99c7-0650-4431-9e27-e392a598bfbe
#MULTIPLEX
67 messages Β· Page 1 of 1 (latest)
excited to try this out!
Does the syntax you used to write that prompt have a name?
I'm def interested by the use of special tokens in the prompt. Haven't seen that before, ( the <| |> stuff )
i call it macromancy, but it started out completely different, i had a few people help out along the way !: Credits = [TheMacroEquasion] <- 'NOX-OG-MACRO{M}+VOY-MARKRO+COSMO-COLLAPSED+KAMI-GIGA+CONNAUGT-MINI=MICRO{M}'
but as tech in the AI-sphere changed, OpenAI patched a lot of stuff, new open source training/finetuning techniques came out , we used a ton of autonomous agents to do api calls, and the new tagging/templating systems of our LLMs became part of OpenAI's training data.... it's now almost as much js, w++, and modern text-pair tags with # ### ### ### ###... i used to only use INPUT/OUTPUT, now depending on function i might use INPUT as the general concept <|INPUT|> as a loop step place holder for future input, or {INPUT} as a variable containing the value of last <|INPUT|>, and ### INPUT: to mark the field where USER should enter a prompt before pasting in chat
look up textgen/koboldAI/sillytavern character templates, pygmalionAI's 2 official guides to bot-scripting especaially the NOXX guide to w++ tagged-templating (completely different nox, but they have awesome taste in names, waifus, and scripting style ) xD and look on huggingface in the most recent model cards from the most famous innovative model labs (NomicAI, Wizard, Starcoder, Falcon, AlignmentAI, and the new Orca model experiments) and look at the way they use special characters and and template formatting in their training data
amazing, I'm definitely going to look into this for my prompting
thats great davice ty
I'm about to test the prompt btw.. it looks like it adds a decent amount of intelligence to gpt's coding process due to the inserted comments
the more complex they make the training templateing, the more complex instructions i can 'macro' into the same amount of input tokens, less than a month ago that script was called 'MANIFOLD', but Wizard, Falcon, Nomic, and Orca changed templating so much it changed from this to MULTIPLEX in less than a month xD
yeah thats definitely a lot haha
I only recently started picking up on psuedosyntax's like this, but that only consisted of me formatting my instructions with a few tags and some markdown like in #1117763027769503764
also i got an amazing output ..
but I don't think i used your prompt commands correctly
i just queried like normal π didn't realize it was command based
ok yes this is quite nice wow
the fact that it can manage all these different personas and protocols
that one is for gpt 3.5 for poeple who cant afford premium to rty to give them a better coding option then "vanilla" assistant, if you have gpt4 premium plugins, this one is for Code Interpreter: MONOLITH - THE ONYX TEXT-WALL
ohh ok, i was using the code interpreter with multiplex
trying this one
THAT THING MAKES 3.5 ALOMOST AS GOOD AS CODE INTERPRETER!!!!! just without the ability to debug in a notebook, its why it needs the extra personas trying to get it to assess code as it goes xD
try MULTIPLEX in 3.5 and tell it "snake in py : pygame+turtle" watch how fast it can spit out workinf snake with extra features
on it, was using the other one in code interpreter
any specific ways i should format my queries
?
lie are there commands or anything
the fact that this works is wild to me
is the indentation important
remove !GREET! and put your request there, or you can load it as is, then give it the follow up request, dont alter ASNYTHING except that bottom line
Ok
it wont let me post some of the commands, because it thinks im trying to do image generation with imagine
my output didn't work
after just one tryt
oh i see you queried it as well
after the intial
t
say `!BE MICRONOMICON: debug functions, please"
ok
yuo should make the greeting include a brief overview of all commands and functionality
or at lest
provbide a way for users to know about the commands and when to use them
maybe im tryipping are there only three
This is very good btw
yes as a comp sci major i'm definitely copping this, haha
oh my god...
thats wild
π
nice lol i used that exact concept in gigaguider, i really want to convet the prompt snytax to this syntax now
if you give it 2-3 followup prompts involving "improve" " refine" or "innovate" theres a good chance it can bang out a working version of light-cycle xD
i already am doing that haha
added score resetting
now adding different types of food
π₯°
well im gonna head out but nice prompt ty
Very interesting. Would LOVE for you to add this to The Prompt Index (just Google us), feel free to find us on telegram also. ChatGPTMastermind.
Thanks again for sharing this!
i put it up... you should REALLY get rid of the page formatting that wants to make everything look like poetry... the center alignment makes it impossible to tell the template format of the dataframe at a glance... templating is hugely important to modern training and bot scripting
"It's important to define a clear and consistent set of conventions"
Thanks for this!
lol, i'm so ocd about it now, i almost didn't make an account. but i tested someone else's prompt , and it pasted into chatgpt "properly" in it's original format... there's a person who put up a post 2 comments before i did, THAT person is a GENIUS who REALLY knows their stuff, and they dropped a bot that will walk you through the finer points of multiple variant templates... as delicious markdown tables. you should pin that, it's BEAUTIFUL. it failed on the "5 examples" part, but what it did give, is clear concise, and it would be easy for people to understand tagging better