++++
2/26/2024
 ++++


It does not appear there is a LLM right now that can write 
in Pilish, which is a constrained writing system where you 
take the digits of Pi and write where the number of letters 
in each word correspond with the digits in Pi.  For 
reference, here's a start 
3.141592653589793238462643383279502884

(I wanted to get the string of pi long enough to include a 
zero and then a bit more ... if you are interested in how 
that is handled in Standard Pilish see [1]). 

    3   1  4   1 5     9
  | Can I bite a funny chocolate?

Is an example sentence in Pilish. 

The free GPT will just screw it up over and over again, and 
do that attempt at convincingly lying thing it does. Someone 
who is paying for GPT-4 had it try, and the first result it 
spit out was a plagiarized poem [2], and then it failed any 
further attempts. 

... It's really funny to watch a program make arthritic 
errors, IE a computer not computing correctly. It's like a 
like a funny chocolate to bite right into, on the 
metaphorical level, which chatGPT is very much able to 
manipulate and discuss. 

It just can't count.

I'll leave the why to those who are more technically 
inclined than me. I'll close out with a Pilish sentence 
mocking GPT: 

  | GPT: "I need a clear tradition to simply write any 
  | thing."

And the best piece of Pilish I have written: 

  | Dao: a love, a twist, tiptoeing in cosmic jokes and 
  | minor delights, localized utopias; Mysticism -- wei wu 
  | wei.


==

[1] How do you deal with zeroes in Pilish? The convention 
that has been settled on is 10 letter word. Here seems like 
the only place to point out that if you want to work in 
words over 10 letters, they will count as two digits in Pi. 
So clearly, they can only be worked into very specific 
places.

[2] The website in question has a clear notice that it 
doesn't want anything copied without written permission. 
While I don't personally like copyright, it is the law of 
the land, and this site is at the very least not releasing 
anything beyond fair use... And even fair use would require 
some kind of attribution, which the trainers of LLMs 
refused to make a consideration. 

==

This work (unlike the site openAI stole from) is hereby in 
the public domain.

Do what you want with it.