[GoLUG] ChatGPT output style training

[Syeed Ali]

On Sat, 30 Nov 2024 14:43:04 -0800
Syeed Ali <syeedali at syeedali.com> wrote:

> You can teach it - to teach you - to teach it.

> There's also a feature to permanently mentor it with a pre-prompt,
> found in your settings.  I used that to inform it of my skills,
> interests, and personality (big-5).

Oh heck, and I just thought of this.

@Steve, do you remember when we were farting around asking it to revise
a summary of one of your books?

Well.

It knows about you already doesn't it?  To some degree at least.  You
could actually train it and turn it into a business partner.

For example, try this:

> Hi, I'm Steve Litt of Troubleshooters.com.  From previous sessions,
> I've learned that you know about some of my work.  In this session I
> want to help better educate you on what I know and what I've done,
> and collaborate such that I can create new income-generating works.
> 
> For example, what topics constrained to my troubleshooting
> expertise/interest do you feel I have not covered, or what method of
> explanation do you feel would be valuable pursuing and publishing
> (and selling!) an explanation of.  Also bear in mind I do
> tele-conferences and mentoring (for pay) so keep that a possibility.
> 
> But perhaps firstly we should confirm what you know about me and what
> you know about my work? (as of your training cutoff date; remind me,
> when was that?)

Treat it like a person, and it's fine to fumble around your grammar and
ideas, and then revisit your original input to ask something again or
in a different way.  You can also take complex topics and split them
into new sessions to keep the primary one sane and tidy.

I did a whole dictionary translation project and it got SLOW because I
had so much research all in one place.  It's funny how I wasn't taking
my own advice.  I thought it was important to keep it all under one
umbrella to capture the whole context, but that was unreasonable given
the computation power I was allotted.