#Thread
last shale
This is how it works.
Basically, ChatGPT only sees this as natural language. While it looks confusing and weird, it could be as simple as SUM: in that outline above.
This is the more specific info-dense details about the process.
The main reason this works really well is because in Chat, there is something called a Named Entity Recognition. Chat is able to see "SUM:" as a kind of made-up word, and then it defines that word as a process.
dusky wing
is there a list of prompts somewhere?
last shale
I make these on the fly.
I am trying to go into more detail but it is a WIP
OH! And then you can reference that process in a session pretty easily (depending on length and complexity)
So, I can keep going with just SUM: or whatever
dusky wing
even if i delete the chat?
last shale
no
and it only lasts for about 10 or so inputs before it forgets and that 4K token space is filled up with other stuff
It only has a 4K attention space.
dusky wing
this is sick tho
using prompts u can extract even more usefulness from this ai
its really great at writing copy
last shale
I bet I could write a way more advanced Mecromancy bot that made these types of prompts 🤣 But it would not be as good as just figuring out the process.
This is my Advanced bot format
dusky wing
what does the memory function do
does it call on the database the ai was trained with
last shale
You mean the 4K?
play around with this and you can get an idea of what 4K worth of tokens in Text looks like.
it is a numerical representation of words, basically. it kind-of compresses some data better than others.
for example, "the quick brown fox jumped over the lazy dog" would not compress very much.
dusky wing
thanks. im still figuring this bot out. just started using it last weekend
ill mess around with this when chat gpt starts feeling like smarterchild
last shale
This is the perfect way to think about Chat. I love that this metaphor keeps reappearing.