Promethean

Four Promethean triple statement rows, each composed of three morphemes, each written in DEX, HEX, and an English translation. From top to bottom: Node, Relation:Template, Template:Spectrum. Node, Relation:Value, Temperature. Node, Relation:Direction, Positive. Node, Relation:Intensity, Weak. Put together these four statements mean "Warm"

An example of the four statements, arranged in multiple different orders. No matter how you order the four statements, the result still means "warm".

Words in #promethean are composed of triple "statements".

Here we have the promethean word for "warm", composed of 4 triple statements.

The statement order doesn't matter, because they all point to the same template.

Order is flexible.

#conlang #conlangs #neography #langsky

Promethean oddly specific vocab. From top top bottom, a Promethean word written in DEX, then HEX, followed by an English definition with examples. The definition is: Template. (noun) - A set of policies which each seems reasonable enough, but as a whole become circular and unreasonable. Example situation 1: "Don't worry, you'll cover this in the next class". "You should have already covered this in the last class". Example situation 2: The help line tells you to call IT. IT tells you to call the help line.

In #promethean it's encouraged to make up words on the spot.

As a result, we end up with words for hyper-specific feelings and circumstances:

◆ "Don't worry, you'll cover this in the next class"
◆ "You should have already covered this in the last class"
◆ 😡

#conlang #neography #langsky

Easy peasy - no homonyms, just an obscene amount of synonyms. This is fine 😄

On the left, a list of six English homonyms. On the right, sets of promethean vocabulary to uniquely distinguish them. From top to bottom, the words are: Light (noun, a source of light; a light bulb). Light (verb, to ignite something; a candle, fuse, etc.). Light (adjective, high in brightness; not dark). Light (adjective, low in color saturation; pale). Light (adjective, low in weight; not heavy). Light (adjective, low in calories, "diet")

Because each word in #promethean is unique and always the same length, there are no homonyms.

You can avoid ambiguity by using precisely the right word for the job. No more pesky context needed!

#conlang #conlangs #neography #langsky

An annotation of the "surprised pikachu" meme. This example shows the original pikachu meme with the text: "me: hmm what happens if i forcibly bend this thing", "thing: *breaks*", "me: " and the image of surprised pikachu. Above the meme is the Promethean word for the meme template. To the right of the meme are three template Promethean parameters. The first represents the action which results in the outcome. The second represents the outcome. The third represents the person that is surprised. Each Promethean word is written in DEX with an equivalent HEX subscript.

Four Promethean dictionary words corresponding to the surprised pikachu meme. Each word is written in DEX with an equivalent HEX subscript and an English definition. The first word's definition is: Template. (meme), A reference to the "surprised pikachu" meme, The feeling of being shocked by an easily predictable outcome. The second word is a parameter of the meme template, representing the action causing the outcome. The third word is also a parameter, representing the entirely predictable outcome. The fourth word is also a parameter, representing the role of the person being surprised.

A little late for #promethean Meme Monday, but here we are:

[⠛⢗⣆⠃⣞⢱⡅⡼⢳⢲⢂⡃⡜⡂⠑⠂] is similar to, but contrasts with [⡂⢡⣷⡂⣬⣼⡀⡂⢉⣾⡎⡢⣐⡤⢠⡘] which refers to a an easily predictable, self-inflicted negative consequence.

#conlang #conlangs #neography #langsky

Fun math fact, despite having two alphabets there are still always 2¹²⁸ possible words:

◆ HEX: 16 letters, 32-letter words: 16³² = 2¹²⁸
◆ DEX: 255 letters, 16-letter words: 256¹⁶ = 2¹²⁸

They're equivalent, but DEX is twice as compact.

A block of 300 Promethean words written in DEX, arranged into 6 columns with 50 lines each. These are randomly generated words with no inherent meaning.

If words in #promethean are generated randomly, won't you get duplicates?

Not likely! There are 2¹²⁸ (~340 undecillion) possible words!

That means even if you generated a billion words per second, it might still take 100 years or so to see a duplicate!

#conlang #conlangs #neography #langsky

A Promethean template word and three example parameter words, written in HEX, DEX, and an English translation. The template word means, "an expression giving both congratulations and condolences simultaneously". The first parameter means "the giver of the expression". The second parameter means "the receiver of the expression". The third parameter means "for, because of, the cause or source warranting something"

Three blocks of Promethean example text, written in DEX, each using the expression that means "congratulations and condolences". The first block is titled "congratulations and condolences" and contains one line of Promethean in DEX. The second block is titled "my congratulations and condolences" and contains four lines of Promethean in DEX. The third block is titled "my congratulations and condolences to you" and contains seven lines of Promethean in DEX.

Because each word in #promethean is a unique identifier, it's easy to invent words for complex or obscure concepts.

Here's a new phrase to express both congratulations and condolences at the same time.

⡮⣡⣏⢋⡮⠓⡄⣩⢵⢬⠱⠭⡙⡠⠌⡤

#conlang #conlanging #neography #langsky

A promethean meme template referencing the "bicycle and the stick" meme. On the right the meme image is shown with three panels - first the person riding a bike with a stick. Second, placing the stick in the bike tire. Third, the crash with the person holding their knee. On the left are a series of Promethean words written in both HEX and DEX. First, the name of the meme template. Second, the word representing the person riding the bike. Third, the word representing the stick, or the act of putting the stick in the bike wheel. Fourth, the word representing the consequence of putting the stick in the bike wheel.

A series of four Promethean dictionary words, written in HEX, DEX, and an English definition. They relate to the "bicycle and stick" meme. In order, the first word is the template name, a reference to the bicycle and stick meme. It refers to an easily predictable, self-inflicted negative consequence. The second word is a parameter representing the person riding the bike. The third word is a parameter representing the stick, or the act of putting the stick in the bike wheel. The fourth word is a parameter representing the consequence of putting the stick in the bike wheel.

It's #promethean meme Monday.

In Promethean, memes are expressed with "templates" and special vocabulary "parameters".

Today, it's the bicycle and the stick.

#conlang #conlanging #neography #langsky

It takes about an hour for morphology posts - the rules are formulaic enough to generate words and expressions by code. Translation posts will take longer because… well it’s hostile by design.

At the top, a dictionary form promethean word written in the DEX alphabet. In the middle, 128 expressed forms of the same word. Using level-3 hostility features, the overall shape and pattern of the expressed words are disguised. The characters used appear completely random at a glance. At the bottom, the "average" number of dots per character are displayed as bands of white with varying opacities. These bands visually display that there is no consistent pattern to each expression. On average the density of characters appears completely random.

(2/2) Specifically for messing with AI readers, #promethean has a "hostility level" you can add which:

◆ Cloaks the pattern
◆ Keeps the word recognizable via a fast formula

The goal of hostility is to throw off bulk corpus analytics, and require pre-processing to block training on raw text.

At the top, a dictionary form promethean word written in the DEX alphabet. In the middle, 128 expressed forms of the same word. Each varies slightly with the DEX characters used, but still has an overall visual pattern to its density. In some areas the dots are sparser, and other they're denser. At the bottom, the "average" number of dots per character are displayed as bands of white with varying opacities. These bands visually display the consistent shape and pattern of these expressed words, even though each expression is unique.

(1/2) In #promethean, every dictionary word is uniquely expressed. You can say the same word repeatedly and it will appear differently each time.

However, a careful observer will notice there's still a consistent "feel" to its shape! Zoom out and you can see a pattern.

#conlang #neography #langsky

At the top is an example paragraph in English: "Who did it?" "The man with the eyepatch. I saw him" "Eyepatch guy? No way." "It was definitely the one-eyed man" They all stopped talking as the man with the eyepatch entered the room. "Well, well", he said. In the middle is a promethean gloss of triple statements identifying the one-eyed man, and creating a unique Promethean word to refer to him by. At the bottom is every english reference to the one-eyed man from the paragraph, such as "him", "he", "eyepatch guy", etc. For every English reference, we could instead use the same unique identifier in Promethean.

More on #promethean coreferences -

For repeated subjects, you only need to describe it once and attach it to a new word.

Subsequently, you can keep using that same word instead of different indirect mentions.

#conlang #neography #langsky

An example English sentence: "She brought her medicine because she heard she was sick". The co-references which refer to the same person are marked - the first and second "she" are marked with node "A". The "her" and third "she" are marked with node "C". Below, a diagram of nodes, with Promethean words identifying them. First is node "A" - representing the woman (person 1) who brought the medicine. Second is node "B" - representing the medicine. Third is node "C" - representing the woman (person 2) who was sick. An arrow passes from A through B to C - representing the act of giving the medicine.

Let's talk about "coreferences" in linguistics:

"She brought her medicine because she heard she was sick"

In English, the reader has to guess who "she" and "her" refers to, but in #promethean we can be very specific. Two people, two words to uniquely identify them.

#conlang #neography #langsky

A table of promethean vocabulary words in dictionary form describing various degrees on a spectrum. In the left column are direction words - positive or negative. In the right column are intensity words, ranging from very weak to strongest. A combination of degree and intensity describes a point on a spectrum. Above the left column is the promethean word for direction, and above the right column is the promethean word for intensity.

An example of building a word to describe a point on the heat spectrum in Promethean. It consists of four lines. The first line marks this statement as a spectrum. The second line denotes it's a temperature spectrum. The third line sets the direction as positive (heat / hotness). The fourth and final line sets the intensity as "weak". All together these four lines mean describe "warm"

In #promethean we can describe degrees like hot vs. cold, cheap vs. expensive, good vs. bad using a "spectrum" template:

◆ value (ex: temperature)
◆ direction (ex: positive)
◆ intensity (ex: very high)

= "scorching hot"

#conlang #neography #langsky

A complex glossed example of Promethean pronoun declension. Six rows of triple words, each rows is a new declension adding detail and granularity to the pronoun. In order, the rows describe "something", "someone", "someone else", "she", "they (female)", and "they (female, honorable)".

An example of a fully inflected pronoun (They, female, honorable) shown in two forms: dictionary form and expressed form.

The #promethean #conlang is extremely compositional.

Declension is done by stacking different properties.

And since all classes are open classes, there's unlimited variations allowed.

#neography #langsky

(2/2) We can use template parameters to customize the meme. For example:

◆ The yelling woman → The #conlang community
◆ The cat → Me using linguistic terms incorrectly

Template parameters are also an open class: someone could add another for "The friend holding the yelling woman back".

An image of the "Woman yelling at a cat" meme. The meme is referenced with a Promethean word. On the left, a second Promethean word references the yelling woman's role. On the right, a third Promethean word references the confused cat's role.

A dictionary listing for the three Promethean words involved in the "woman yelling at a cat" meme. Each word is presented in HEX, DEX, and an English dictionary definition. The first word is defined as `A referece to the "woman yelling at a cat meme" The feeling of being confused by a strong accusation.` The second word is defined as `The role of the yelling woman` The third word is defined as `The role of the confused cat`

(1/2) Memes in #promethean can be even more abstract than a phrase. They could reference an image, situation, or feeling.

#conlang #langsky #neography

On the left, two Promethean words. The first references the "two nickels" meme from Phineas and Ferb. The meme goes: If I had a nickel for every time (something happened), I'd have two nickels. Which isn't a lot, but it's weird that it happened twice. The second Promethean word is a reference to the parameter of that meme - whatever thing may have happened twice. On the right is a picture of the original two-panel meme template with Dr. Doofenshmirtz without text.

Despite being so verbose, in #promethean you can still compress a lot of meaning into a single word.

Memes, references and inside jokes can be distilled into a template.

Similar to Star Trek Tamarian, you'd have to get the reference to understand.

#conlang #conlangs #langsky #neography

A general interest in the upper limits of AI models. That and a near-future short story involving human resistance against the inevitable robot overlords.

An example of 50 English characters broken up into 12 LLM tokens. Most English words become their own unique token.

An example of 50 HEX characters broken up into 27 LLM tokens. This is about a 2x increase over English.

An example of 50 DEX characters broken up into 150 LLM tokens. This is about a 12x increase over English.

(4/4) So at the end of the day, compared to the English alphabet:

◆ HEX: multiplies tokens by ~2x per character
◆ DEX: multiplies tokens by ~12x per character

Simply by using these alphabets, we're increasing AI inference costs by 2-12x. That's a pretty good start!

(3/4) But #promethean's HEX and DEX don't resemble natural letter distributions. HEX is composed of common letters, but chains of HEX don't appear often and so they have less efficient token encodings.

DEX is composed of 8-dot Braille which is so rare that each letter is usually a token by itself!

(2/4) AI models first encodes text into a sequence of "tokens".

It's an extremely efficient encoding, so 1 token generally equals ~3/4th of a word. Common English words are almost always their own token.

You can play around with it here:

platform.openai.com/tokenizer

Three text samples of equal length in English, HEX, and DEX. Alongside each sample is a letter distribution graph showing the most frequent letters and what percent of the sample they represent. The distribution for HEX is flatter than English, and the distribution for DEX is substantially flatter than English.

(1/4) For LLM-hostile design, we start with choice of alphabet.

#promethean's two alphabets (HEX and DEX) greatly increase an AI's token count per character.

#conlang #conlangs #langsky #neography

(3/3) So #promethean exists to trip up LLMs by:

◆ Driving up token counts and compute costs by ~2-3 orders of magnitude
◆ Sparking more hallucinations
◆ Exceeding a model's effective context window

Here there be dragons.

(2/3) At the moment, LLMs still struggle a bit with:

◆ Complex graphical structures & reasoning
◆ IDs and randomness without guessable meaning
◆ Obscure or less-used symbols
◆ Length of text and using their full context window

An example translation of two English texts to Promethean. The first English text is "me", which takes 640 characters in the DEX alphabet. The second English text is "my cat" which consumes 1536 characters in the DEX alphabet. By design, it takes a great deal more characters to express concepts in Promethean.

(1/3) It takes really a long time to say anything in #promethean.

Why is that?

Promethean was created to be "LLM-hostile" - theoretically readable to a human with inhuman patience, but to absolutely throngle any AI models that stumble across it.

#conlang #conlangs #langsky #neography

Three rows of Promethean text divided into three sections each. The first row shows fully expressed 64-character morphemes. The second row shows the underlying pairs of 32-character morphemes split apart, marked in green for morphemes with meaning and marked in white for random decorative noise. The third and final row shows only the morphemes with real meaning, marked in green.

In #promethean we express a morpheme by doubling its length - interleaving decorative embellishments.

This means that when reading a piece of text, we'll ultimately discard half its volume!

The true meaning lies in the remaining half.

#conlang #conlangs #langsky #neography

One advantage of making a needlessly complex, visually inscrutable conlang like Promethean:

Nobody's going to check your work for mistakes ;)

#conlang #conlangs #langsky #neography

#promethean syntax is always arranged into blocks of three morphemes called "statements"

You could imagine it as each new statement adding to the nodes and edges of a visual diagram.

#conlang #conlangs #langsky #neography