[ot][spam][crazy] using spasmodic behaviors to translate openai prompts to local models?
i wonder if i could just like spam the list in a way that converts the langchain prompts 1. i have some prompt tuning intuition and could make a stab at some of this 2. i'm generally on mobile when spamming, so it's harder to test local models, but maybe doable setting up a good karl-harness that makes this easier would likely make it become enjoyable
some of the prompts i really like are ones that extract relevant data from snippets of documents when dissociated i generally have very poor creativity (spoiler; the mcboss stories are made by transcribing my internal experiences in a way that pushes through my coverup programming and such. more to come!) so it would be helpful for these to have example documents maybe working with larger systems based on langchain, then i could enable caching and accumulate example data and manually convert the prompts maybe with help of a small script that measured karl-loss for me to minimize
participants (1)
-
Undescribed Horrific Abuse, One Victim & Survivor of Many