>>106349402
Like I said the examples I posted earlier were shit. Here's a better example that I think better illustrates what I'm trying to do:
Stage 1: raw snippets from the story:
{
"prompt": "She felt a blush creep up her neck,",
"response": "a tell-tale heat that betrayed her composure."
}
Stage 2: smart LLM explains the " WHY" of the sensation
{
"prompt": "She felt a blush creep up her neck,",
"response": "as an emotional stimulus caused the vasodilation of capillaries in the dermis, increasing peripheral blood flow."
}
This further grounds the model into learning WHY people feel certain emotions and what trigger them. It gets better at understanding emotional intelligence
State 3:
{
"prompt": "She felt a blush creep up her neck,",
"chosen": "the rush of blood to her skin a clear signal of her lost composure.",
"rejected": "as an emotional stimulus caused the vasodilation of capillaries in the dermis, increasing peripheral blood flow."
}
Model still understands human emotions, why we feel pleasure, etc, but it is then nudged into explaining them like a normal person instead of a textbook or someone with giga Aspergers. This insurance it knows WHY a dude feels good when you tug on his penis AND that its able to actually weave that into a story..... probably...maybe...idk I've never tried something like this in particular
¯\_(ツ)\_/¯
