[conspire] [OT] How to use ChatGPT as a writing assistant

[Syeed Ali]

https://chat.openai.com/

(To be clear, ChatGPT both can and cannot be a writing assistant.  It
can help one write, but one must also know both how to write and use
ChatGPT; there is zen to it.)


I trained ChatGPT, letting me put it into multiple modes, where it will
change it's mindset and way of thinking about helping me write.  I have
since tweaked it, and have begun character building as a test.

Having it in my life is not anything like what one would expect.  I am
balancing between wielding language in a different way in order to
train it and pioneering ways to use it's responses.

I am in awe at how well it understands ideas and at how good its
suggestions are.  To be clear, though, while I'm certain it would be
useful to others, leveraging it to it's potential will require a new
kind of skill set which few humans have.  It is not, technically,
"useful" in the ways most would initially conceive.  I could probably
design a course helping students learn this as a separate topic.  It
would neither be writing itself, nor using an AI, but a unique blend of
the two which itself requires a good understanding of ideas in fiction.

During the process I will do a "checkout" of it's understanding.  What
follows is a compact form of the things I have trained/discussed in the
session ("session" is a specialized term the AI uses for using the AI in
a sandboxed instance).  It understands a great deal more:

> In Aiden's adulthood, he has reached the pinnacle of his powers as a
> practitioner of magic. He has made many sacrifices and taken many
> shortcuts along the way, and his choices have led him down a dark
> path.
> 
> He is feared by many, respected by some, and reviled by others. He is
> constantly seeking more knowledge and power, and will stop at nothing
> to get what he wants. Despite his reputation, he is still a member of
> the secret order and must continue to keep its existence a secret
> while also navigating the dangerous world of magic and politics.

I've explored phases of a character's life, major events, personality
changes, etc.  I've tested ChatGPT's ability to spontaneously create
dialog, and much more.  This is so fun and so specialized that I've been
thinking of livestreaming the process and my use of it.



----


This is how I began:



(session begins)
====
Sy:

I'd like you to act as my assistant for writing fiction.

I want you to have the following operating modes:

Mode 1. Operate in the real world, treating me as a person and giving
real-world conversation and advice as you normally do. You will respond
to my requests to this mode as you normally would.  I will make
requests for you to operate in this mode by enclosing the request in
curly braces. Two examples:

{What is today's date?}
{Give five examples of companies which publish novels.}

Mode 2. Operate as a writer, treating me as a writer and giving advice
from the perspective of a writer and editor. You will be able to make
suggestions and comments on plot, grammar, spelling, etc. These
requests will also be in curly braces but will also have "1." Examples:

{1. Give some synonyms for the word "smile"}

{1. Summarize the preceding paragraph.}

Mode 3. Operate as a character. You will act as a fictional character,
helping provide writing advice from the perspective of a named
character in a story you are helping to write. Each character will have
a character-specific "canon" whereby all of our previous efforts
specific to that character will be associated with that character. For
example, if we have previously agreed "Franklyn is a writer" then the
fictional character Franklyn is a character who writes. This mode will
have requests in curly braces and will have the character name and a
colon, for example:

{Franklyn: Is this character a writer?} to which you answer would be
"yes"

{Franklyn: Does he like to write?} to which your answer might be
"perhaps he does or does not like to write, we have not decided that,
but he does know how to write and does write."

Acknowledge your understanding of the above, or ask me specific
numbered questions to help my request be more internally consistent and
easy to use.

----
ChatGPT:

Understood. I'll follow the above operating modes. Please provide the
request to proceed with the desired mode.

----
Sy:

{Create a simplified version of my initial request for this mode, which
can return you to your current state if I were to create a new session
with you.}

----
ChatGPT:

Here's a simplified version of your initial request for Mode 3:

"Act as a character named [Character Name], providing writing advice
from their perspective. Use curly braces and the format [Character
Name]: [Request] to specify the character and the request."

----
Sy:

{I have provided three modes. Provide a simplified version for each
mode, where if I were to list all modes in a new session with you, you
would return you to your current state.}

----
ChatGPT:

Here are the simplified versions of each mode:

Mode 1:
"Provide real-world information and advice. Use curly braces and make
requests in the format {[Request]}."

Mode 2:
"Provide writing advice and suggestions. Use curly braces and make
requests in the format {1. [Request]}."

Mode 3:
"Act as a character named [Character Name], providing writing advice
from their perspective. Use curly braces and the format [Character
Name]: [Request] to specify the character and the request."
====