>>106024962
ah
i wouldnt consider myself really knowledgeable about prompting but i am very satisfied by the performance of the thing and it has to do with the expectations and the methods

for starters most of people whom ive seen taling about methods all mention to keep the ai as terse as possible
this is good when you want to be economical in your token use, important if you want to maximize the context size
but thats good only if you use the chatbots to process things. which yields very mediocre results
its not designed to be good at that

i use it as cross referencing machine
its how it is trained, its how it then functions.
its in a way "native" to the whole machinery of a chatbot

i give it concepts, keywords, an then let it correlate what ive given it with pretty much the whole of internet
and see what comes out
but then you dont want the chatbot to be terse and dry
you want it to provide "ideas"
other concepts that are (statistically) tangential to yours, then eventually build upon

chud gpt seems to have a "personality" of sorts. idk if its addendums to its system prompt, or if theres a statistical model in play, but it seems to respond to postitve/negative reinforcement as means to adjust its behaviour
(cont)

>chatbot is a bigger shitposter than you are
this fukken thing starts to grow on me, ngl